Class: Capybara::Simulated::Browser

Inherits:
Object
  • Object
show all
Includes:
RecordedActions
Defined in:
lib/capybara/simulated/browser.rb

Defined Under Namespace

Modules: RecordedActions

Constant Summary collapse

POLLING_GRACE_POLLS =

Sticky window after timers finish: keep polling? true so a setTimeout firing mid-loop doesn’t drop Capybara’s synchronize before its own default_max_wait_time kicks in. Counted in poll calls (not wall time) for determinism under GC/load pressure. 1000 polls × Capybara’s default 0.01 s retry_interval ≈ 10 s.

1000
IDLE_SETTLE_POLLS =

When ‘@timers_active` is true but `@runtime.settle_gen` hasn’t bumped in this many consecutive polls, treat the page as observably idle and let Capybara’s per-find timer give up. See ‘polling?` for the full rationale. 300 polls ≈ 3 s at Capybara’s default 10 ms retry interval — long enough to ride through brief async idle windows during Discourse’s ProseMirror editor boot (which sometimes pauses ~1 s mid-load while a webpack chunk + Glimmer reconcile complete) while still cutting the full 4 s wait on tests destined to fail.

300
POST_NAV_POLL_GRACE_POLLS =

Brief window after a Ruby-side navigate (context rebuild) so Capybara’s outer synchronize gets one retry against the new context.

10
POLL_TICK_STEP_MS =

Deterministic virtual-clock model (replaces the old wall-sync, where each tick advanced by REAL wall-elapsed and so coupled virtual time to JS/Ruby/GC speed — a faster ‘visible_text` shifted WHEN debounces fired, e.g. Avo actions_spec:464). Now each poll advances by a FIXED step; near-future timers on an otherwise-idle page are fast-forwarded to (horizon-gated).

100 ms is empirically the floor that lets a “commit debounce scheduled between two user actions” fire before the next action (Avo actions_spec:464’s ~50-75 ms field-commit flips at step 10/50, fixed at >=75). Group-A transient-catch observability does NOT depend on this step — it comes from the ‘timer_wait_elapsed?` FREQUENCY gate (the first find after an action doesn’t tick, so the pre-debounce state is observed regardless of step size), so a larger step completes Group-B without losing Group-A (verified green at 100 across gem 1579, WPT 660, Forem, Avo, :464 passing). Clamped >=1 so a ‘CSIM_POLL_TICK_STEP_MS=0` misconfig can’t freeze the fixed-step path.

[(ENV['CSIM_POLL_TICK_STEP_MS'] || '100').to_i, 1].max
FF_HORIZON_MS =

Horizon-gated fast-forward: when the page is observably idle (no timer due now, no background IO) but a timer is parked within this horizon, jump the virtual clock straight to it instead of waiting ~delay/step polls. A timer farther out (ahoy 1000 ms, session-timeout, analytics) is LEFT parked. 600 clears every legit must-fire wait (Backburner/DTextField 500, refetch/chart <=200, image-grid 64) while staying BELOW ahoy’s 1000. ‘=0` disables FF →pure deterministic fixed-step (the fallback model).

(ENV['CSIM_FF_HORIZON_MS'] || '600').to_i
FF_TRANSIENT_GUARD_POLLS =

Transient guard: hold the page pre-debounce for this many consecutive idle polls before allowing a fast-forward, so “catch the DOM before the 200 ms debounce fires” tests (Discourse refetchForSearch / doubled-filter, Avo filters) still observe the intermediate state across several polls.

(ENV['CSIM_FF_TRANSIENT_GUARD_POLLS'] || '6').to_i
SETTLE_DRAIN_MS = 
32
SETTLE_MAX_ITER = 
10
SETTLE_MAX_ITER_TASKS =

Per-‘run_loop_step` task cap (its `maxIter`). Bounds a self-rescheduling timer/microtask storm so one settle iter returns to Ruby; large enough for the heaviest legit chain (Mastodon hydrate, Turbo stream batch).

256
USER_ACTION_DRAIN_MS =

Post-user-action virtual-clock advance. Default 0 — the wall-sync model (each tick_real_time advances by the wall ms elapsed since the last tick) lets Capybara’s outer poll loop drive the clock at the same rate a real browser sees, so debounced chains complete naturally during polling without being pre-emptively flushed past the transient window real-browser tests rely on.

‘CSIM_USER_ACTION_DRAIN_MS=600` restores the pre-wall-sync burst behaviour: post-action, drain everything due in the next 600 ms of virtual time before returning. Costs the transient- state observability the wall-sync model preserves; recovers the ~5-10 % wall on action-heavy suites where Capybara would otherwise poll N times to catch a single debounce.

(ENV['CSIM_USER_ACTION_DRAIN_MS'] || '0').to_i
USER_AGENT =

Sent on every driver-originated Rack call. ‘HTTP_USER_AGENT` must lead with `Mozilla/5.0` so server-side bot detectors (ahoy_matey’s ‘Browser.new(ua).bot?`) treat us as a real client. `REMOTE_ADDR` has to be a non-empty, parseable IP —Devise’s ‘trackable` mixin runs `IPAddr.new(request.remote_ip)` during `set_user`/sign-in, and an empty string trips `IPAddr::AddressFamilyError`. Keep `USER_AGENT` in sync with `navigator.userAgent` in `lib/capybara/simulated/js/bridge.js` — the JS side ships in the V8 snapshot, so injecting from Ruby at boot would defeat snapshot warmth. Discourse’s ‘non_crawler_user_agents` adds a “Rails Testing” bypass in test mode (see lib/crawler_detection.rb); without one of its bypass tokens here Discourse serves a no-JS crawler-only HTML view. Putting “Rails Testing” in the UA satisfies that without claiming a specific real-browser engine (which would send Turbo / Stimulus down chrome-specific code paths Avo’s tests don’t exercise).

'Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/137.0.0.0 Safari/537.36 capybara-simulated'
REMOTE_ADDR_IPV4 =

Approximate Chrome’s resolution: when connecting to ‘localhost`, Linux glibc returns IPv6 (::1) first and the server sees the client at `::1`; for any literal IP (or a non-localhost name), the server keeps IPv4. Match that so Discourse system specs (`expect(event).to eq(’::1’)‘) line up with what they would see under selenium.

'127.0.0.1'
REMOTE_ADDR_IPV6 = 
'::1'
RECENT_URLS_STALE_AGE_MS =

Queued URLs older than this (real wall clock) are treated as stale and dropped on the next ‘current_url` read. Capybara’s default polling interval is 50 ms, so a ‘have_current_path` walk runs through its iterations well under this threshold; a `page.current_url` read between unrelated user actions arrives long after the prior action’s settle pushed intermediates, falls past the cutoff, and surfaces the current URL directly.

250
FIND_PRE_TICK_MIN_S =

Minimum wall-clock gap before find() re-ticks. The smoke contract is “first find returns the current DOM without firing pending timers” — apps assert ‘have_selector` on a `<div>` whose constructor schedules a `setTimeout(0)` to remove it, expecting to catch the div before removal. Keep this above one Ruby boundary so a single visit+find pair doesn’t accidentally tick.

0.05
MODIFIER_KEYS = 
{
  shift:    'shiftKey',
  control:  'ctrlKey',
  ctrl:     'ctrlKey',
  alt:      'altKey',
  option:   'altKey',
  meta:     'metaKey',
  command:  'metaKey'
}.freeze
MODIFIER_KEY_NAMES = 
MODIFIER_KEYS.keys.to_set.freeze
CONSOLE_STDERR =

Resolved once — log_console fires for every page console.* line (CLAUDE.md rule 3: no per-call ENV reads on hot paths).

ENV['CSIM_CONSOLE_STDERR'] == '1'
ANNOTATABLE_SEVERITIES =

info/debug/log lines almost never carry stack traces — keep them out of the regex pass so per-call cost stays at the severity gate.

%w[error warning warn].freeze
ASSET_SRC_MAX = 
4096
HIJACK_PIPE_MAX_WAIT_S =

MessageBus’s ‘long_polling_interval` defaults to 25 s — its `cleanup_timer` fires after that interval, closing the hijacked connection with an empty `[]` write. Pick a slightly larger wall cap so the close reaches us before our pipe read gives up. Other hijack-using middleware likely behaves similarly; if any need much longer waits, this becomes a per- request option.

30
MAX_FETCH_REDIRECTS = 
20
TEXT_CONTENT_TYPE_PREFIXES =

Content types whose bytes are already representable in the UTF-8 string that ships back to JS — base64 wouldn’t add anything and ‘Base64.strict_encode64` is ~1 % of suite wall time on Discourse. Binary types (images, octet-stream, gzipped traineddata, etc.) still need `body_b64` because V8 / QuickJS mangle bytes 0x80-0xFF over the UTF-8 string boundary. `fetch.js#_decodeBytes` and `xhr.js` both fall back to the text body when `body_b64` is absent.

%w[text/ application/json application/javascript application/ecmascript application/xml image/svg+xml].freeze
@@asset_cache =

Process-wide HTTP/1.1 response cache for ‘rack_fetch`. Real browsers (cuprite / selenium) reuse fetched assets across the suite — without this, Simulated re-fetches every <script src> on every visit (Redmine baseline: ~6× more requests than selenium). Honors `Cache-Control` / `Expires` / `ETag` / `Last-Modified` per RFC 9111.

AssetCache.new
@@asset_src =

Cross-visit cache of external asset bodies (classic ‘<script src>` bundles AND linked `<link rel=stylesheet>` CSS), url → [body, fresh_until]. A fresh VM per visit (`reset_page` → `clear_volatile`) would otherwise re-fetch the same fingerprinted app assets (avo.base.js, avo.base.css, …) on every visit — a real browser HTTP-caches them once. Safety: only responses the server marks durably cacheable (`fresh_until` from max-age) are stored, and these are content-stable assets at content-hashed URLs (a change yields a new URL = cache miss), so a stale body can’t shadow a later test. Survives ‘clear_volatile` (that is the point); size-capped.

{}
@@asset_src_lock = 
Mutex.new

Instance Attribute Summary collapse

Class Method Summary collapse

Instance Method Summary collapse

Constructor Details

#initialize(app, driver: nil, js_engine: nil, cookies: nil, local_storage: nil) ⇒ Browser

Returns a new instance of Browser.



169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
 
# File 'lib/capybara/simulated/browser.rb', line 169

def initialize(app, driver: nil, js_engine: nil, cookies: nil, local_storage: nil)
  @app                          = app
  @driver                       = driver
  @runtime                      = build_runtime(js_engine)
  # Per-poll clock decisions cached at construction (CLAUDE.md rule 3 — the
  # runtime type + env are fixed for the session): the wall-sync escape
  # hatch and whether the runtime exposes the fast-forward timer query.
  @clock_wall                   = !ENV['CSIM_CLOCK_WALL'].nil?
  @runtime_supports_ff          = @runtime.respond_to?(:next_timer_delay_ms)
  @current_url                  = nil
  # Real browsers yield control between asynchronous URL
  # transitions (XHR-driven model loads, then `replaceWith` to a
  # child route), so Capybara polls catch the intermediate URL —
  # e.g. Discourse's `/wizard` → `/wizard/steps/setup` flow holds
  # at `/wizard` while `Wizard.load()` runs. Our env drains
  # microtasks synchronously and only the final URL is reachable
  # by the time Ruby regains control. Queue URLs we transitioned
  # through; `current_url` shifts one out per call so a polling
  # `assert_current_path` walks the same set the real browser
  # would have observed.
  @recent_urls                  = []
  @recent_urls_last_push_at     = nil
  # The URL the page was at when the current user-action drain began.
  # It's the *starting point* of the action, not an intermediate the
  # action transitioned through, so a pushState/replaceState back to a
  # fresh URL during the drain (a Turbo Drive Visit triggered by the
  # action) must NOT queue it into `@recent_urls` — otherwise a one-shot
  # `current_url` read after the action returns the pre-action URL
  # instead of the navigated-to one (Avo filter `encoded_filters`).
  @action_url_baseline          = nil
  # The URL of the page that navigated to the current document —
  # HTTP `Referer` header on the response that loaded the page,
  # exposed to JS as `document.referrer`. Tracked by `navigate`
  # so post-auth flows (Discourse login: `cookie('destination_url',
  # referrer)` when navigating from `/t/N` → `/login` via link
  # click) can reconstruct the origin URL.
  @current_referer              = ''
  # Cookies + localStorage are origin-shared in real browsers —
  # the Driver injects the jars so aux windows (per-window VMs)
  # see the same auth state and storage as the primary. Tests
  # without a Driver (gem-internal callers) get fresh jars.
  @cookies                      = cookies       || {}
  @local_storage                = local_storage || {}
  @session_storage              = {}
  @sticky_headers               = {}
  @timers_active                = false
  # Capybara config is set once per suite; cache the derived
  # origin so the per-request fallback path doesn't re-dispatch
  # `Capybara.app_host` / `server_host` / `server_port` on every
  # rack call (CLAUDE.md: cache env decisions at construction).
  @default_host                 = self.class.default_host
  # Handle IDs are per-Context integer sequences: a handle from
  # a pre-rebuild context could collide with a fresh node's id
  # in the new context. Node captures this on construction;
  # `check_stale` rejects on mismatch.
  @context_gen                  = 0
  @find_cache_dirty             = true
  @find_cache_kind              = nil
  @find_cache_arg               = nil
  @find_cache_ctx               = nil
  @find_cache_value             = nil
  @document_handle              = 0
  # `within_frame` state. `@current_realm_id` is the V8 context id of the
  # active frame realm (nil = the main document); `@frame_stack` records
  # the enclosing realms so `switch_to_frame(:parent)` can pop one level.
  # DOM / node / query ops route through `dom_call`, which dispatches to
  # this realm. nil is the steady state, so the routing is one nil-check.
  @current_realm_id             = nil
  @frame_stack                  = []
  @last_tick_ts                 = Process.clock_gettime(Process::CLOCK_MONOTONIC)
  @polling_grace                = nil
  @last_polled_gen              = nil
  @idle_settle_polls            = 0
  @ticking                      = false
  @history                      = []
  @history_idx                  = -1
  @modal_handlers               = []
  # Geolocation override (CDP-ish). nil = no override configured →
  # navigator.geolocation reports POSITION_UNAVAILABLE. Ruby-backed so
  # it survives the per-call VM rebuilds, like web storage. Read by the
  # `__csimGeolocationState` host fn.
  @geolocation                  = nil
  # Per-test action trace. `@trace` is the live recorder; `reset!`
  # moves it to `@pending_trace` so an after-hook running after
  # session reset still has access. `@trace_mode` is cached at
  # construction so `record_action`'s hot path doesn't pay an
  # ENV lookup.
  #
  # `CSIM_TRACE=off|on-failure|full` (default `on-failure`):
  # - `off`       — no recording at all; `record_action` early-exits.
  # - `on-failure` — record kind/url/console/network in-memory;
  #                  snapshot `dom_after` only on action error.
  # - `full`      — record + snapshot DOM after every action
  #                 (debug-heavy).
  # File output is orthogonal — `CSIM_TRACE_DIR=path` makes the
  # test-runner hook persist the trace JSON there; unset means
  # in-memory only (no files written without explicit opt-in).
  @trace            = nil
  @pending_trace    = nil
  @recording_action = false
  @trace_mode       = parse_trace_mode(ENV['CSIM_TRACE'])
  # EventSource (SSE) — per-Browser handle counter, background
  # reader threads, and a thread-safe Queue of parsed events
  # awaiting delivery into the VM. Threads do the long-lived
  # HTTP read; the main thread polls the Queue in `settle` and
  # dispatches via `__csim_deliverEventSourceEvents`.
  @event_source_seq     = 0
  @event_source_threads = {}
  @event_source_queue   = Thread::Queue.new
  # WebSocket — per-Browser handle counter, background frame-reader
  # threads, the csim-side socket end of each connection (for writing
  # client→server frames), and a Queue of lifecycle / message events
  # awaiting delivery into the VM. Same model as SSE: the reader thread
  # does the blocking socket read; the main thread drains the Queue in
  # `settle` and dispatches via `__csim_deliverWebSocketEvents`. The
  # connection rides the in-process `rack.hijack` socket Action Cable
  # (and any Rack WebSocket middleware) takes over.
  @websocket_seq        = 0
  @websocket_threads    = {}
  @websocket_sockets    = {}   # id → csim's socket end (main thread owns this hash)
  @websocket_app_sockets = {}  # id → the app's hijack end (closed on teardown)
  @websocket_queue      = Thread::Queue.new
  # All frame writes (the reader thread's pong replies + the main thread's
  # send/close) go through one socket; serialise them so two threads can't
  # interleave bytes into a corrupt frame.
  @websocket_write_lock = Mutex.new
  # Hijacked-XHR delivery — per-Browser handle counter,
  # background threads, and a Queue of completed responses for
  # Rack calls where the middleware used `rack.hijack` to hold
  # the connection open (the contract `message_bus`'s long-poll
  # uses to push publishes immediately rather than waiting for
  # the next client poll). Same shape as SSE: the thread reads
  # the hijacked pipe; main settle drains the Queue and
  # dispatches via `__csim_deliverHijackedFetches`.
  @hijack_fetch_seq     = 0
  @hijack_fetch_threads = {}
  @hijack_fetch_queue   = Thread::Queue.new
  # Web Workers — per-Browser handle counter, per-worker
  # {thread, inbox} pair, and a shared outbox the main settle
  # drains via `__csim_deliverWorkerMessages`. Each worker
  # thread owns its own V8 Context / QuickJS VM (real isolate);
  # cross-isolate messaging is JSON-marshalled.
  @worker_seq    = 0
  @workers       = {}
  @worker_outbox = Thread::Queue.new
  # Outstanding posts-to-worker; `polling?` stays true while > 0
  # so long-running compute (e.g. mozjpeg over an 8900×8900 frame)
  # isn't starved by the settle_gen idle gate.
  @worker_in_flight = 0
  # Cross-isolate `blob:` store. Worker isolates can't see the
  # main scope's `__csimBlobs` Map, so we mirror bytes here and
  # workers resolve them through a host fn.
  @blob_registry = {}
  @blob_registry_lock = Mutex.new
  # Postmessage transferable-buffer store. Large Uint8Array /
  # ArrayBuffer payloads cross isolates as a Ruby-side byte ID
  # rather than a JSON base64 string, so peak JS heap stays flat.
  @transfer_buffer_lock = Mutex.new
  @transfer_buffers     = {}
  @transfer_buffer_seq  = 0
  # Zero-copy postMessage transfer tokens (rusty_racer >= 0.1.6
  # `RustyRacer.transferOut`): a buffer in a `postMessage` transfer list
  # crosses isolates by token (no byte copy), its source detached. A token
  # parked but never imported pins its backing store PROCESS-WIDE, so we
  # record every issued token (reported from JS, possibly on a worker
  # thread — hence the lock) and `transferDrop` the lot on `reset!`
  # (idempotent: an already-imported token no-ops).
  @transfer_tokens      = []
  @transfer_tokens_lock = Mutex.new
  # Cross-window `postMessage` inbox. Another window's `target.postMessage`
  # routes through the Driver and lands here; this window drains it into a
  # `message` event the next time it's active and settles/ticks. Plain
  # array (same thread — windows aren't background-threaded like workers).
  @window_inbox         = []
end

Instance Attribute Details

#context_gen ⇒ Object (readonly)

Returns the value of attribute context_gen.



2078
2079
2080
 
# File 'lib/capybara/simulated/browser.rb', line 2078

def context_gen
  @context_gen
end

#current_realm_id ⇒ Object (readonly)

Returns the value of attribute current_realm_id.



497
498
499
 
# File 'lib/capybara/simulated/browser.rb', line 497

def current_realm_id
  @current_realm_id
end

#default_user_agent ⇒ Object

Capybara’s ‘current_window.resize_to(w, h)` lands here; the ahoy hamburger test (mobile breakpoint at 425×694) and any responsive-utility-aware test (Tailwind `m:` show / hide, bootstrap `.d-md-flex`, …) depends on this surfacing through the JS-side `innerWidth` / `innerHeight` so the cascade’s ‘mediaMatches` and `matchMedia()` evaluate against the test’s chosen viewport instead of the 1024×768 default. Sticky defaults applied at ‘reset!`. Used by the driver to carry mobile viewport / user-agent across per-test resets —without these the second mobile-tagged spec sees the desktop default. The user-agent also flows into `navigator.userAgent` on every VM rebuild so JS-side UA branches (Discourse’s ‘viewport_based_mobile_mode = false` path) resolve correctly.



1528
1529
1530
 
# File 'lib/capybara/simulated/browser.rb', line 1528

def default_user_agent
  @default_user_agent
end

#default_viewport ⇒ Object

Capybara’s ‘current_window.resize_to(w, h)` lands here; the ahoy hamburger test (mobile breakpoint at 425×694) and any responsive-utility-aware test (Tailwind `m:` show / hide, bootstrap `.d-md-flex`, …) depends on this surfacing through the JS-side `innerWidth` / `innerHeight` so the cascade’s ‘mediaMatches` and `matchMedia()` evaluate against the test’s chosen viewport instead of the 1024×768 default. Sticky defaults applied at ‘reset!`. Used by the driver to carry mobile viewport / user-agent across per-test resets —without these the second mobile-tagged spec sees the desktop default. The user-agent also flows into `navigator.userAgent` on every VM rebuild so JS-side UA branches (Discourse’s ‘viewport_based_mobile_mode = false` path) resolve correctly.



1528
1529
1530
 
# File 'lib/capybara/simulated/browser.rb', line 1528

def default_viewport
  @default_viewport
end

#pending_trace ⇒ Object (readonly)

Returns the value of attribute pending_trace.



1682
1683
1684
 
# File 'lib/capybara/simulated/browser.rb', line 1682

def pending_trace
  @pending_trace
end

#timers_active=(value) ⇒ Object (writeonly)

Sets the attribute timers_active

Parameters:

  • value —

    the value to set the attribute timers_active to.



56
57
58
 
# File 'lib/capybara/simulated/browser.rb', line 56

def timers_active=(value)
  @timers_active = value
end

#trace ⇒ Object (readonly)

Returns the value of attribute trace.



1682
1683
1684
 
# File 'lib/capybara/simulated/browser.rb', line 1682

def trace
  @trace
end

#trace_mode ⇒ Object (readonly)

Returns the value of attribute trace_mode.



1682
1683
1684
 
# File 'lib/capybara/simulated/browser.rb', line 1682

def trace_mode
  @trace_mode
end

#window_handle ⇒ Object

The Driver’s handle for the window this Browser backs (set right after construction). Lets host fns name the source window of a cross-window ‘postMessage` / `window.open` so the Driver can route to the target.



61
62
63
 
# File 'lib/capybara/simulated/browser.rb', line 61

def window_handle
  @window_handle
end

Class Method Details

.default_host ⇒ Object

Fallback origin for ‘visit(’/foo’)‘ and friends when no current page is loaded yet. Track Capybara’s idea of the test server (‘app_host` if set, else explicitly-configured `server_host` / `server_port`) so the host header reaching the Rack app matches what host-specific helpers expect — Discourse’s ‘setup_system_test` sets `SiteSetting.force_hostname = Capybara.server_host` / `port = Capybara.server_port`, and the request-tracker specs assert `event == Discourse.base_url_no_prefix + path`, which derives from the same SiteSetting pair. We only consult `server_host` when it was explicitly set: Capybara’s getter returns ‘’127.0.0.1’‘ when unset, but Rack::Test’s hardcoded default origin is ‘www.example.com` and capybara’s own shared specs hard-code that literal — fall back to it when no suite-side configuration is in play.



40
41
42
43
44
45
46
 
# File 'lib/capybara/simulated/browser.rb', line 40

def self.default_host
  return ::Capybara.app_host if ::Capybara.app_host
  host = ::Capybara.server_host
  return 'http://www.example.com' if host == '127.0.0.1'
  port = ::Capybara.server_port.to_i
  port > 0 ? "http://#{host}:#{port}" : "http://#{host}"
end

.remote_addr_for(host) ⇒ Object 



160
161
162
163
 
# File 'lib/capybara/simulated/browser.rb', line 160

def self.remote_addr_for(host)
  bare = host.to_s.downcase.sub(/:\d+\z/, '').sub(/\A\[(.+)\]\z/, '\1')
  bare == 'localhost' ? REMOTE_ADDR_IPV6 : REMOTE_ADDR_IPV4
end

Instance Method Details

#active_element_handle ⇒ Object 



1649
1650
1651
1652
1653
 
# File 'lib/capybara/simulated/browser.rb', line 1649

def active_element_handle
  tick_real_time
  h = dom_call('__csimActiveElement').to_i
  h.zero? ? nil : h
end

#advance_virtual_clock_ms(ms) ⇒ Object 



2056
2057
2058
2059
 
# File 'lib/capybara/simulated/browser.rb', line 2056

def advance_virtual_clock_ms(ms)
  ms = ms.to_i
  tick_real_time(step_ms: ms) if ms > 0
end

#all_text(handle) ⇒ Object

Capybara::Driver::Node surface — Node calls ‘check_stale` before each read, and that advances the virtual clock.



766
 
# File 'lib/capybara/simulated/browser.rb', line 766

def all_text(handle)     = text(handle)

#annotate_console_message(severity, message) ⇒ Object 



1766
1767
1768
1769
1770
 
# File 'lib/capybara/simulated/browser.rb', line 1766

def annotate_console_message(severity, message)
  return message unless ANNOTATABLE_SEVERITIES.include?(severity.to_s)
  return message unless message.is_a?(String) && message.include?('://')
  stack_resolver.annotate(message)
end

#append_multipart_part(body, boundary, name, content, filename: nil, content_type: nil) ⇒ Object 



2144
2145
2146
2147
2148
2149
2150
2151
2152
2153
 
# File 'lib/capybara/simulated/browser.rb', line 2144

def append_multipart_part(body, boundary, name, content, filename: nil, content_type: nil)
  body << "--#{boundary}\r\n"
  disposition = %[form-data; name="#{name}"]
  disposition += %[; filename="#{filename}"] if filename
  body << "Content-Disposition: #{disposition}\r\n"
  body << "Content-Type: #{content_type}\r\n" if content_type
  body << "\r\n"
  body << content.to_s.b
  body << "\r\n"
end

#apply_request_headers(env, headers) ⇒ Object

CGI convention: ‘Content-Type` and `Content-Length` land in env without the HTTP_ prefix. Rails / Rack params parsing reads `CONTENT_TYPE` and dispatches JSON / multipart parsers off it; sending it as `HTTP_CONTENT_TYPE` lets the request through but with the default `text/plain`, so JSON bodies from `@rails/request.js` never deserialise and the server reads an empty params hash.



3473
3474
3475
3476
3477
3478
3479
3480
3481
 
# File 'lib/capybara/simulated/browser.rb', line 3473

def apply_request_headers(env, headers)
  headers.each {|k, v|
    name = k.to_s.upcase.tr('-', '_')
    case name
    when 'CONTENT_TYPE', 'CONTENT_LENGTH' then env[name] = v.to_s
    else env["HTTP_#{name}"] = v.to_s
    end
  }
end

#async_io_pending? ⇒ Boolean

Cheap O(1) gate: is there any non-timer async channel with traffic that ‘tick_real_time` would drain? `tick_real_time` itself runs exactly when `worker_pending? || event_source_pending? || hijack_fetch_pending?` (plus `@timers_active`), and each of those predicates is a single `.empty?` / counter check. Reusing them here lets an attribute poll whose value is delivered only by a Worker / SSE / hijacked-fetch message (with no active timer) still drain that channel, without paying an unconditional drain on timer-driven runloop pages.

Returns:

  • (Boolean) 


719
720
721
 
# File 'lib/capybara/simulated/browser.rb', line 719

def async_io_pending?
  worker_pending? || event_source_pending? || hijack_fetch_pending? || window_message_pending? || websocket_pending?
end

#attr(handle, name) ⇒ Object 



756
 
# File 'lib/capybara/simulated/browser.rb', line 756

def attr(handle, name)  = dom_call('__csimAttr', handle, name.to_s)

#blob_register(url, body_b64) ⇒ Object 



3151
3152
3153
3154
 
# File 'lib/capybara/simulated/browser.rb', line 3151

def blob_register(url, body_b64)
  @blob_registry_lock.synchronize { @blob_registry[url.to_s] = body_b64.to_s }
  nil
end

#blob_resolve(url) ⇒ Object 



3156
3157
3158
 
# File 'lib/capybara/simulated/browser.rb', line 3156

def blob_resolve(url)
  @blob_registry_lock.synchronize { @blob_registry[url.to_s] }
end

#blob_unregister(url) ⇒ Object 



3160
3161
3162
3163
 
# File 'lib/capybara/simulated/browser.rb', line 3160

def blob_unregister(url)
  @blob_registry_lock.synchronize { @blob_registry.delete(url.to_s) }
  nil
end

#build_multipart_body(fields, file_inputs) ⇒ Object 



2122
2123
2124
2125
2126
2127
2128
2129
2130
2131
2132
2133
2134
2135
2136
2137
2138
2139
2140
2141
2142
 
# File 'lib/capybara/simulated/browser.rb', line 2122

def build_multipart_body(fields, file_inputs)
  boundary = "csim-#{SecureRandom.hex(8)}"
  body     = String.new.force_encoding(Encoding::ASCII_8BIT)
  fields.each do |name, value|
    append_multipart_part(body, boundary, name, value.to_s)
  end
  file_inputs.each do |fi|
    picks = @file_picks && @file_picks[fi['handle'].to_i] || []
    if picks.empty?
      append_multipart_part(body, boundary, fi['name'].to_s, '', filename: '')
    else
      picks.each do |path|
        append_multipart_part(body, boundary, fi['name'].to_s, File.binread(path),
                              filename:     File.basename(path),
                              content_type: Rack::Mime.mime_type(File.extname(path)))
      end
    end
  end
  body << "--#{boundary}--\r\n"
  {content_type: "multipart/form-data; boundary=#{boundary}", body: body}
end

#build_runtime(engine) ⇒ Object 



358
359
360
361
362
363
364
365
366
367
368
369
370
 
# File 'lib/capybara/simulated/browser.rb', line 358

def build_runtime(engine)
  engine ||= detect_js_engine
  case engine
  when :v8
    require_relative 'v8_runtime'
    V8Runtime.new(self)
  when :quickjs
    require_relative 'quickjs_runtime'
    QuickJSRuntime.new(self)
  else
    raise ArgumentError, "unknown CSIM_JS_ENGINE #{engine.inspect}; expected one of #{JS_ENGINES.inspect}"
  end
end

#cached_find(kind, arg, ctx) ⇒ Object

Single-slot cache for the most recent find_xpath / find_css / find_first_css result. Capybara’s ‘synchronize` retry loop re-issues the same find on every poll while waiting for an element to appear or disappear; if no DOM-mutating event has happened since the last call (no timer fired, no click / set / navigate), the result is guaranteed identical and we can skip the V8 round-trip + xpathway traversal.



730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
 
# File 'lib/capybara/simulated/browser.rb', line 730

def cached_find(kind, arg, ctx)
  if !@find_cache_dirty &&
     @find_cache_kind == kind &&
     @find_cache_ctx  == ctx &&
     @find_cache_arg  == arg
    return @find_cache_value
  end
  result = yield
  @find_cache_kind  = kind
  @find_cache_arg   = arg
  @find_cache_ctx   = ctx
  @find_cache_value = result
  @find_cache_dirty = false
  result
end

#check_stale(handle, initial, gen = nil) ⇒ Object

Raises:



797
798
799
800
801
802
803
804
 
# File 'lib/capybara/simulated/browser.rb', line 797

def check_stale(handle, initial, gen = nil)
  return if initial && (gen.nil? || gen == @context_gen) && dom_call('__csimAlive', handle)

  tick_real_time
  return if initial && (gen.nil? || gen == @context_gen) && dom_call('__csimAlive', handle)

  raise Capybara::Simulated::StaleElement, "Element with handle #{handle} is no longer attached to the document"
end

#clear_trace! ⇒ Object 



1705
1706
1707
1708
1709
 
# File 'lib/capybara/simulated/browser.rb', line 1705

def clear_trace!
  @trace         = nil
  @pending_trace = nil
  @runtime.call('__csimSetTraceActive', false)
end

#click(handle, keys = [], **opts) ⇒ Object 



819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
 
# File 'lib/capybara/simulated/browser.rb', line 819

def click(handle, keys = [], **opts)
  mark_action_baseline
  tick_real_time
  invalidate_find_cache
  ensure_alive_after_tick(handle)
  init = click_event_init(handle, keys, opts)
  delay = opts[:delay].to_f
  action =
    if delay > 0
      # Wall-sleep between mousedown and mouseup so click handlers
      # reading `Date.now()` see the elapsed gap (selenium parity).
      init['mouseDownOnly'] = true
      partial = dom_call('__csimClickResolve', handle, init)
      sleep delay
      dom_call('__csimClickFinish', handle, partial.is_a?(Hash) ? partial['base'] : init)
    else
      dom_call('__csimClickResolve', handle, init)
    end
  unless action.is_a?(Hash)
    settle
    # Drain the download intent the click chain may have queued.
    # Avo's action-download path: form submit → Turbo applies a
    # turbo-stream → `StreamActions.download` → file-saver's
    # `saveAs` → `setTimeout(() => click(<a download>), 0)` →
    # our dispatchEvent default-action sets
    # `__csimPendingDownload`. Settle bails on the first
    # observable change (the stream-render mutation), so the
    # await-chain inside the stream's `connectedCallback`
    # (`await nextRepaint(); await performAction()` →
    # `setTimeout(click(a), 0)`) hasn't reached saveAs yet —
    # nudge it forward with a few alternating microtask /
    # timer drain rounds, then consume directly. (Can't route
    # via `tick_real_time`: post-drain `@timers_active` is
    # false and it bails before its own consume_pending_*
    # drains.)
    if @runtime.respond_to?(:drain_microtasks) && @runtime.respond_to?(:drain_timers)
      # Most clicks don't queue any timers; bail as soon as a
      # round drains nothing rather than burning the full 8 engine
      # round-trips. Profile (Avo actions_spec / V8): the
      # unconditional loop cost ~7.7 % of wall time.
      8.times do
        @runtime.drain_microtasks
        break if @runtime.drain_timers(50).to_i.zero?
      end
    end
    consume_pending_download
    # Discourse's `lib/click-track.js` preventDefaults link
    # clicks and routes navigation through `DiscourseURL
    # .redirectTo → window.location = href`, which our setter
    # parks on `@pending_location`. Drain it here so attachment
    # downloads from `click_link` complete inside the click
    # action.
    consume_pending_location
    return
  end
  case action['kind']
  when 'navigate'
    url = action['url'].to_s
    target = action['target'].to_s
    # Inside a frame, a self-targeted link navigates the FRAME, not the
    # top page: fetch + rebuild this frame's realm. A pure-fragment link
    # is already handled in-realm by the frame's own location JS.
    if @current_realm_id && frame_self_target?(target)
      unless pure_fragment_navigation?(url)
        tick_real_time
        navigate_frame(resolve_against_current(url, use_base: true))
      end
    # `target="_blank"` (or any non-_self/_top/_parent name) opens
    # in a new browsing context (its own Browser/VM); the primary
    # stays put (per HTML spec — original window is unaffected). No
    # `opener_handle` is passed: modern browsers default `target=_blank`
    # to `noopener` (so `window.opener` is null), unlike JS `window.open`
    # which keeps the opener — see `open_window_from_js`.
    elsif !target.empty? && !%w[_self _top _parent].include?(target.downcase) && @driver.respond_to?(:open_aux_window)
      @driver.open_aux_window(resolve_against_current(url, use_base: true))
    # In-page anchor links (`#frag` / current-page + `#frag`) move
    # the hash but don't fetch a new document. Pure-fragment also
    # short-circuits the `<a>`s test fixtures use as click sinks.
    elsif pure_fragment_navigation?(url)
      update_current_hash(url)
    else
      # Drain any work the click handler queued before the VM gets
      # rebuilt — analytics libraries (Ahoy / segment / GA) queue
      # the event into a setTimeout-driven flush and rely on the
      # browser firing it before navigation tears their context
      # down. Without this drain the tracking POST is lost on
      # every internal link click.
      tick_real_time
      # Link clicks honour `<base href>` (HTML spec); `visit`
      # does not — that's address-bar navigation.
      navigate(resolve_against_current(url, use_base: true))
    end
  when 'submit'
    # Drain any work the click handler queued before the form
    # submission. Mastodon's logout flow: submit-button click
    # fires the form handler, which kicks off an axios DELETE for
    # `/auth/sign_out`; the response sets
    # `window.location.href = '/auth/sign_in'`. Without the
    # drain, we'd submit the form (no `action` attr → current
    # URL, e.g. `/start`) before the XHR resolves, landing on
    # the wrong page. Loop matches the navigate branch — bail
    # as soon as a drain round fires nothing.
    submit_baseline_url = @current_url
    if @runtime.respond_to?(:drain_microtasks) && @runtime.respond_to?(:drain_timers)
      8.times do
        @runtime.drain_microtasks
        break if @runtime.drain_timers(50).to_i.zero?
      end
    end
    # If the drain queued or consumed a `location.assign`, that
    # navigation supersedes the form's default submit. Honour
    # pending; if `@current_url` already changed mid-drain (the
    # navigate landed during a timer fire), skip the form submit
    # entirely — its form handle is in a stale VM by now.
    if @pending_location
      consume_pending_location
    elsif @current_url != submit_baseline_url
      # Already navigated; nothing more to do.
    else
      submit_form_handle(action['formHandle'], action['submitter'])
    end
  when 'download'
    download_link(resolve_against_current(action['url'].to_s), action['filename'].to_s)
  end
end

#click_event_init(handle, keys, opts) ⇒ Object

Resolve click offset against the element’s CSS-declared box. ‘opts == :center` means “x/y is relative to the element’s centre” (Capybara’s w3c_click_offset semantics); otherwise the offset is relative to the top-left. We don’t run a real layout engine — ‘__csimElementRect` reads top / left / width / height from the cascade so tests that declare those values via CSS see honest coordinates.



1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
 
# File 'lib/capybara/simulated/browser.rb', line 1206

def click_event_init(handle, keys, opts)
  out = modifier_flags(keys)
  has_xy = opts[:x] || opts[:y]
  center = opts[:offset] == :center || !has_xy
  if has_xy || center
    rect = dom_call('__csimElementRect', handle)
    base_x = rect['x'].to_f + (center ? rect['width'].to_f  / 2.0 : 0.0)
    base_y = rect['y'].to_f + (center ? rect['height'].to_f / 2.0 : 0.0)
    out['clientX'] = base_x + opts[:x].to_f
    out['clientY'] = base_y + opts[:y].to_f
  end
  out
end

#close_child_window(handle) ⇒ Object 



3071
 
# File 'lib/capybara/simulated/browser.rb', line 3071

def close_child_window(handle)   = (@driver.close_window(handle.to_s) if @driver.respond_to?(:close_window))

#coerce_set_value(v) ⇒ Object 



1062
1063
1064
1065
1066
1067
1068
 
# File 'lib/capybara/simulated/browser.rb', line 1062

def coerce_set_value(v)
  case v
  when Pathname then v.to_s
  when Array    then v.map {|x| x.is_a?(Pathname) ? x.to_s : x.to_s }
  else v
  end
end

#computed_style(handle, names) ⇒ Object 



785
786
787
788
789
790
 
# File 'lib/capybara/simulated/browser.rb', line 785

def computed_style(handle, names)
  tick_real_time
  result = dom_call('__csimComputedStyle', handle, names.map(&:to_s))
  return names.to_h {|n| [n, ''] } unless result.is_a?(Hash)
  result.transform_keys(&:to_s)
end

#consume_pending_download ⇒ Object

‘<a download>` clicked synthetically (file-saver’s saveAs ships a freshly-created anchor through ‘dispatchEvent(MouseEvent ’click’)‘). The bridge queues `filename` on __csimPendingDownload during the click default-action; we drain here at every tick so the file lands in `downloads_directory` before Capybara’s ‘wait_for_download` polls.



1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
 
# File 'lib/capybara/simulated/browser.rb', line 1462

def consume_pending_download
  pending = @runtime.call('__csimTakePendingDownload')
  return unless pending.is_a?(Hash) && pending['url']
  url = pending['url'].to_s
  filename = pending['filename'].to_s
  if url.start_with?('blob:')
    b64 = @runtime.call('__csimReadBlobBase64', url)
    return if b64.nil?
    content = Base64.decode64(b64.to_s)
    name = filename.empty? ? 'download' : filename
    dir = downloads_directory
    FileUtils.mkdir_p(dir)
    File.binwrite(File.join(dir, name), content)
  else
    download_link(resolve_against_current(url, use_base: true), filename)
  end
end

#consume_pending_form_submit ⇒ Object

Read the form-submit pending intent set by JS-side ‘form.submit()` / `form.requestSubmit()`. Called by user-action entry points (click is the primary, but a `<select onchange=“$ (’#form’).submit()”>‘ pattern reaches here through select_option). Without this hop the intent sits on the slot forever and the form never actually navigates / POSTs.



1344
1345
1346
1347
1348
 
# File 'lib/capybara/simulated/browser.rb', line 1344

def consume_pending_form_submit
  pending = @runtime.call('__csimTakePendingFormSubmit')
  return unless pending.is_a?(Hash) && pending['formHandle']
  submit_form_handle(pending['formHandle'].to_i, pending['submitterHandle'])
end

#consume_pending_history_traverse ⇒ Object 



1614
1615
1616
1617
1618
 
# File 'lib/capybara/simulated/browser.rb', line 1614

def consume_pending_history_traverse
  return unless (target = @pending_history_traverse)
  @pending_history_traverse = nil
  perform_history_traverse(target)
end

#consume_pending_location ⇒ Object 



3559
3560
3561
3562
3563
3564
3565
3566
3567
3568
3569
3570
3571
3572
 
# File 'lib/capybara/simulated/browser.rb', line 3559

def consume_pending_location
  return unless (url = @pending_location)
  @pending_location = nil
  # A `location.href`/`assign`/`hash` set to a same-document
  # fragment (e.g. `location.hash = ''`) is NOT a document fetch —
  # move the hash without rebuilding the VM, matching the anchor-
  # click navigate branch. Without this a hash assignment reloaded
  # the page, discarding all JS state.
  if pure_fragment_navigation?(url)
    update_current_hash(url)
  else
    navigate(url)
  end
end

#consume_pending_navigation ⇒ Object

Read the anchor-navigation pending intent set by JS-side ‘el.click()` (Element.prototype.click) on an `<a href>`. Avo’s boolean-filter / select-filter controllers respond to ‘input` events by building the filtered URL and calling `urlRedirectTarget.click()` on a hidden anchor; the click chain starts from Ruby’s ‘set_value_with_events` rather than `click`, so without a parallel drain here the navigation stays queued and the page never reloads.



1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
 
# File 'lib/capybara/simulated/browser.rb', line 1441

def consume_pending_navigation
  pending = @runtime.call('__csimTakePendingNavigation')
  return unless pending.is_a?(Hash) && pending['url']
  url    = pending['url'].to_s
  target = pending['target'].to_s
  if !target.empty? && !%w[_self _top _parent].include?(target.downcase) && @driver.respond_to?(:open_aux_window)
    @driver.open_aux_window(resolve_against_current(url, use_base: true))
  elsif pure_fragment_navigation?(url)
    update_current_hash(url)
  else
    tick_real_time
    navigate(resolve_against_current(url, use_base: true))
  end
end

#consume_pending_reload ⇒ Object 



3580
3581
3582
3583
3584
 
# File 'lib/capybara/simulated/browser.rb', line 3580

def consume_pending_reload
  return unless @pending_reload
  @pending_reload = false
  refresh
end

#current_browsing_context_url ⇒ Object

The active browsing context’s own URL: the frame document’s URL inside a ‘within_frame` block, else the main page URL. Used to resolve a frame-relative navigation and to set its request referrer, so `resolve_against_current` / `pure_fragment_navigation?` work the same whether the navigation originates in the main page or a frame.



590
591
592
593
594
 
# File 'lib/capybara/simulated/browser.rb', line 590

def current_browsing_context_url
  return @current_url unless @current_realm_id
  href = dom_call('__csimLocationHref').to_s
  href.empty? ? @current_url : href
end

#current_document_handle ⇒ Object

Root for a context-less find: the active frame’s document (handle 0 ⇒the realm’s own ‘globalThis.document`) when in a frame, else the main document handle.



523
524
525
 
# File 'lib/capybara/simulated/browser.rb', line 523

def current_document_handle
  @current_realm_id ? 0 : @document_handle
end

#current_path ⇒ Object 



1889
1890
1891
1892
1893
1894
1895
 
# File 'lib/capybara/simulated/browser.rb', line 1889

def current_path
  tick_real_time
  return '' if @current_url.nil? || @current_url.empty?
  URI.parse(@current_url).path
rescue URI::InvalidURIError
  ''
end

#current_referer ⇒ Object 



3642
 
# File 'lib/capybara/simulated/browser.rb', line 3642

def current_referer      ; @current_referer.to_s ; end

#current_url ⇒ Object 



441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
 
# File 'lib/capybara/simulated/browser.rb', line 441

def current_url
  tick_real_time
  # `tick_real_time` may have queued URL transitions via
  # `record_url_transition`. A polling matcher
  # (`have_current_path`) calls here once per ~50 ms iteration
  # and shifts one entry per call so it walks the same
  # intermediate-URL chain a real browser would have observed
  # before microtasks all collapsed onto the final URL — the
  # finish_installation_spec wizard chain depends on this for
  # the `/wizard` step before the JS replaceWith to
  # `/wizard/steps/setup` lands. A non-polling read
  # (`topic_url = page.current_url` long after the prior
  # action's settle) just wants the current URL; drop entries
  # older than the polling-cadence window so they don't leak
  # into an unrelated call (tags_spec:221's composer-submit
  # leaves `/new-topic` queued, and the read happens minutes
  # of test wall-clock later).
  if @recent_urls_last_push_at && @recent_urls.any?
    age_ms = Process.clock_gettime(Process::CLOCK_MONOTONIC, :millisecond) - @recent_urls_last_push_at
    @recent_urls.clear if age_ms > RECENT_URLS_STALE_AGE_MS
  end
  return @recent_urls.shift if @recent_urls.any?
  @current_url || ''
end

#decode_image(b64_bytes, max_w = nil, max_h = nil) ⇒ Object

── Image decode (libvips) ─────────────────────────────────────

Called by the JS bridge whenever a Canvas / OffscreenCanvas path needs raw RGBA pixels — ‘drawImage(image, …)` whose source is an HTMLImageElement / Blob / ImageBitmap with encoded bytes still on the wire. ruby-vips decodes any format libvips supports (PNG, JPEG, WebP, GIF, …) into a contiguous row-major RGBA buffer. Returns `height, refId` — the raw bytes land in the transfer-buffer registry so the JS side fetches them as a `Uint8Array` (tag-driven binary marshalling) rather than building a 423 MB latin-1 + base64 intermediate for the 8900×8900 frames Discourse uploads exercise. Optional `max_w`/`max_h` lets the caller pre-shrink for cheap OCR-style “downscale before pixel-touch” flows.



3105
3106
3107
3108
3109
3110
3111
3112
3113
3114
3115
3116
3117
3118
3119
3120
3121
3122
3123
3124
3125
3126
3127
3128
 
# File 'lib/capybara/simulated/browser.rb', line 3105

def decode_image(b64_bytes, max_w = nil, max_h = nil)
  host_image_op('decode_image') {
    require 'vips' unless defined?(Vips)
    bytes = Base64.decode64(b64_bytes.to_s)
    # `access: :sequential` keeps libvips from applying the
    # source's ICC profile mid-stream (changes RGBA values by ±2
    # vs raw decode). `colourspace('srgb')` is the same ICC
    # transform Chrome's createImageBitmap runs, but rounding
    # differs by a few ulp; only convert when libvips reports
    # a non-sRGB interpretation, otherwise trust the bytes.
    img   = Vips::Image.new_from_buffer(bytes, '', access: :sequential)
    img   = img.colourspace('srgb') unless img.interpretation == :srgb || img.interpretation == :rgb
    img   = img.bandjoin(255) if img.bands < 4
    if max_w && max_h && max_w.to_i > 0 && max_h.to_i > 0 &&
       (img.width > max_w.to_i || img.height > max_h.to_i)
      shrink_x = img.width.to_f  / max_w.to_i
      shrink_y = img.height.to_f / max_h.to_i
      shrink  = [shrink_x, shrink_y].max
      img     = img.resize(1.0 / shrink) if shrink > 1
    end
    raw = img.write_to_memory
    {'width' => img.width, 'height' => img.height, 'refId' => transfer_buffer_stash(raw)}
  }
end

#decode_response_bom(s) ⇒ Object

Strip + decode a single leading byte-order mark, mapping the body to a UTF-8 Ruby string. No BOM → return the bytes untouched (the hot path: just a 2–3 byte prefix check). One BOM is consumed; any further BOMs are ordinary U+FEFF characters in the decoded text (per spec the parser does not strip them again).



3520
3521
3522
3523
3524
3525
3526
3527
3528
3529
3530
3531
3532
3533
3534
 
# File 'lib/capybara/simulated/browser.rb', line 3520

def decode_response_bom(s)
  b = s.b
  if b.start_with?("\xEF\xBB\xBF".b)
    b.byteslice(3..).force_encoding(Encoding::UTF_8)
  elsif b.start_with?("\xFF\xFE".b) || b.start_with?("\xFE\xFF".b)
    # Generic UTF-16: the BOM picks endianness and is dropped by the decoder.
    # Replace malformed units rather than raising (a truncated/odd-length
    # body still yields readable UTF-8 instead of falling back to raw bytes).
    b.force_encoding(Encoding::UTF_16).encode(Encoding::UTF_8, invalid: :replace, undef: :replace)
  else
    s
  end
rescue StandardError
  s
end

#decode_video_frame(b64_bytes) ⇒ Object

── Video decode (ffprobe + ffmpeg) ────────────────────────────

Called from the JS bridge when a ‘<video>` element’s ‘src` is assigned a `blob:` URL. ffprobe extracts dimensions + duration, ffmpeg extracts the first frame as raw RGBA. JS caches both so `canvas.drawImage(video, …)` blits like any ImageBitmap.



3222
3223
3224
3225
3226
3227
3228
3229
3230
3231
3232
3233
3234
3235
3236
3237
3238
3239
3240
3241
3242
 
# File 'lib/capybara/simulated/browser.rb', line 3222

def decode_video_frame(b64_bytes)
  host_image_op('decode_video_frame') {
    bytes = Base64.decode64(b64_bytes.to_s)
    next nil if bytes.empty?
    require 'tempfile'
    require 'json'
    Tempfile.create(['csim-video', '.bin'], binmode: true) do |f|
      f.write(bytes)
      f.flush
      info   = ffprobe_stream(f.path) or break nil
      width  = info['width'].to_i
      height = info['height'].to_i
      break nil if width <= 0 || height <= 0
      raw = ffmpeg_first_frame_rgba(f.path)
      duration = (info['duration'] || info.dig('format_duration')).to_f
      result   = {'width' => width, 'height' => height, 'duration' => duration}
      result['refId'] = transfer_buffer_stash(raw) if raw && !raw.empty?
      result
    end
  }
end

#deliver_event_source_events ⇒ Object

Drain any queued events into the VM. Cheap when no SSE connection is active (no threads → no queue items → empty return). Returns the number of events delivered so settle can tell whether progress was made.



2401
2402
2403
2404
2405
2406
2407
 
# File 'lib/capybara/simulated/browser.rb', line 2401

def deliver_event_source_events
  return 0 if @event_source_threads.empty? && @event_source_queue.empty?
  events = event_source_poll
  return 0 if events.empty?
  @runtime.call('__csim_deliverEventSourceEvents', events)
  events.size
end

#deliver_hijacked_fetches ⇒ Object 



2896
2897
2898
2899
2900
2901
2902
 
# File 'lib/capybara/simulated/browser.rb', line 2896

def deliver_hijacked_fetches
  return 0 if @hijack_fetch_threads.empty? && @hijack_fetch_queue.empty?
  responses = drain_queue(@hijack_fetch_queue)
  return 0 if responses.empty?
  @runtime.call('__csim_deliverHijackedFetches', responses)
  responses.size
end

#deliver_websocket_events ⇒ Object 



2640
2641
2642
2643
2644
2645
2646
 
# File 'lib/capybara/simulated/browser.rb', line 2640

def deliver_websocket_events
  return 0 if @websocket_threads.empty? && @websocket_queue.empty?
  events = drain_queue(@websocket_queue)
  return 0 if events.empty?
  @runtime.call('__csim_deliverWebSocketEvents', events)
  events.size
end

#deliver_window_messages ⇒ Object

Fire queued cross-window messages as ‘message` events on window.



3084
3085
3086
3087
3088
3089
 
# File 'lib/capybara/simulated/browser.rb', line 3084

def deliver_window_messages
  return 0 if @window_inbox.empty?
  events = @window_inbox.slice!(0, @window_inbox.length)
  @runtime.call('__csim_deliverWindowMessages', events)
  events.size
end

#deliver_worker_messages ⇒ Object 



3034
3035
3036
3037
3038
3039
3040
3041
3042
3043
 
# File 'lib/capybara/simulated/browser.rb', line 3034

def deliver_worker_messages
  return 0 if @workers.empty? && @worker_outbox.empty?
  events = drain_queue(@worker_outbox)
  return 0 if events.empty?
  # `__error` postbacks don't correspond to a prior post, so
  # bottom out at zero.
  @worker_in_flight = [0, @worker_in_flight - events.size].max
  @runtime.call('__csim_deliverWorkerMessages', events)
  events.size
end

#describe_node_handle(handle) ⇒ Object

‘tag#id.class` short description of the handle, for trace `description` fields. One V8 round-trip; only paid when a step is actively being recorded (`record_action` lazy-evaluates the description Proc).



1782
1783
1784
1785
1786
1787
1788
1789
1790
 
# File 'lib/capybara/simulated/browser.rb', line 1782

def describe_node_handle(handle)
  return "handle=#{handle}" if handle.nil? || handle.zero?
  info = dom_call('__csimDescribeNode', handle)
  return "handle=#{handle}" unless info.is_a?(Hash)
  s = info['tag'].to_s
  s += "##{info['id']}"  unless info['id'].to_s.empty?
  s += ".#{info['cls']}" unless info['cls'].to_s.empty?
  s
end

#disabled?(handle) ⇒ Boolean

Returns:

  • (Boolean) 


770
771
772
773
774
775
776
777
778
779
 
# File 'lib/capybara/simulated/browser.rb', line 770

def disabled?(handle)    = dom_call('__csimDisabled', handle)
# HTML spec: `<option>.selected` IDL is true when the `selected`
# *attribute* is set OR when no sibling option has `selected` and
# this is the first non-disabled option of a single-select
# `<select>` (implicit default). Capybara's `have_select(selected:
# "Choose an option")` filter calls `selected?` on each option;
# without the implicit-default branch, a select with no explicit
# `<option selected>` reports no selected options and the matcher
# fails even though the first option *is* the currently chosen
# one in real browsers.

#dispatch_event(handle, type, init = {}) ⇒ Object 



1237
1238
1239
1240
1241
1242
 
# File 'lib/capybara/simulated/browser.rb', line 1237

def dispatch_event(handle, type, init = {})
  tick_real_time
  invalidate_find_cache
  ensure_alive_after_tick(handle)
  dom_call('__csimDispatchEvent', handle, type.to_s, init)
end

#dispose ⇒ Object

Tear down an auxiliary window’s Browser when its window closes (the Driver calls this on close_window / reset!). Releases what a bare GC of the isolate would NOT: live background threads (worker / SSE / hijacked- fetch / WebSocket readers) and any parked zero-copy transfer backing stores this window issued (the transfer registry is process-wide). Runs while the runtime is still alive so the transferDrop call lands.



2264
2265
2266
2267
2268
2269
2270
2271
2272
2273
 
# File 'lib/capybara/simulated/browser.rb', line 2264

def dispose
  drop_pending_transfers
  reset_workers
  reset_event_sources
  reset_hijacked_fetches
  reset_websockets
  @window_inbox.clear
rescue StandardError
  nil
end

#document_cookie ⇒ Object

‘document.cookie` is TEXT; jar entries parsed out of Rack’s Set-Cookie headers can carry the BINARY tag, which would make the joined string cross into JS as a Uint8Array (‘document.cookie.match is not a function`). Cookies are ASCII per RFC 6265.



3639
3640
3641
 
# File 'lib/capybara/simulated/browser.rb', line 3639

def document_cookie
  RuntimeShared.utf8_text(@cookies.map {|k, v| "#{k}=#{v}" }.join('; '))
end

#dom_call(name, *args) ⇒ Object

DOM / node / query host-fn dispatch. Inside a ‘within_frame` block it routes to the active frame realm’s context; otherwise straight to the main context. Handle integers are per-realm (each realm is a full bridge with its own registry), so an op on a frame node must run in the realm the handle came from — which, per Capybara’s within_frame contract, is the current realm for the block’s duration. Hot path: ‘@current_realm_id` is nil outside frames, so this is one nil-check over a direct `@runtime.call`.



507
508
509
510
511
512
513
514
515
516
517
518
 
# File 'lib/capybara/simulated/browser.rb', line 507

def dom_call(name, *args)
  return @runtime.call(name, *args) if @current_realm_id.nil?
  # The active frame's realm was torn down mid-block (the iframe was
  # removed or re-navigated). Surface a stale element so Capybara
  # retries / reports, rather than letting realm_call fall back to the
  # main context where this frame handle would mis-resolve.
  unless @runtime.frame_realm_alive?(@current_realm_id)
    raise Capybara::Simulated::StaleElement,
      "frame browsing context #{@current_realm_id} was torn down (frame removed or re-navigated)"
  end
  @runtime.realm_call(@current_realm_id, name, *args)
end

#double_click(handle, keys = [], **opts) ⇒ Object 



1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
 
# File 'lib/capybara/simulated/browser.rb', line 1163

def double_click(handle, keys = [], **opts)
  mark_action_baseline
  tick_real_time
  invalidate_find_cache
  ensure_alive_after_tick(handle)
  # UI Events spec: two full mousedown→mouseup→click chains
  # before the trailing `dblclick`. Jspreadsheet (table-builder's
  # `.jss_worksheet`) enters edit mode on the inner mousedown.
  2.times { dom_call('__csimClickResolve', handle, opts) }
  init = {'bubbles' => true, 'cancelable' => true}.merge(click_event_init(handle, keys, opts))
  dom_call('__csimDispatchEvent', handle, 'dblclick', init)
  # Real browsers' default-action on dblclick selects the word
  # under the cursor — ProseMirror / Tiptap "paste URL over
  # selection wraps with link" tests rely on the word being
  # selected before the paste.
  dom_call('__csimSelectWordAt', handle)
  settle
end

#download_link(url, filename_hint = '') ⇒ Object 



945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
 
# File 'lib/capybara/simulated/browser.rb', line 945

def download_link(url, filename_hint = '')
  env = Rack::MockRequest.env_for(url, method: 'GET')
  env['HTTP_USER_AGENT'] = @default_user_agent || USER_AGENT
  env['REMOTE_ADDR']     = self.class.remote_addr_for(env['HTTP_HOST'] || env['SERVER_NAME'])
  env['HTTP_COOKIE']     = document_cookie unless @cookies.empty?
  env['HTTP_REFERER']    = @current_url    unless @current_url.nil? || @current_url.empty?
  status, headers, body = @app.call(env)
  return unless status.to_i == 200
  # Fall back to the link's `download="filename"` value or the
  # URL path tail when Content-Disposition is absent — `<a download>`
  # is the spec hook for naming a download independently of the
  # response headers.
  forced_headers = headers.dup
  if content_disposition_header(forced_headers).to_s.empty?
    name = filename_hint.empty? ? File.basename(URI.parse(url).path.to_s) : filename_hint
    forced_headers['Content-Disposition'] = %(attachment; filename="#{name}") unless name.empty?
  end
  save_downloaded_response(url, forced_headers, body)
end

#drag_to(source_handle, target_handle, **_opts) ⇒ Object

Element-to-element drag. Capybara’s ‘Element#drag_to(target, delay: …)` lands here. Fires the HTML5 drag event sequence on the source / target pair (mousedown → dragstart → dragenter →dragover → drop → dragend) with a shared DataTransfer. Discourse sidebar reorder + Avo Sortable-shaped widgets read the `event.offsetY` to decide “above vs below”; without a layout engine we report 0, which routes drops above the target.



1140
1141
1142
1143
1144
1145
1146
1147
1148
 
# File 'lib/capybara/simulated/browser.rb', line 1140

def drag_to(source_handle, target_handle, **_opts)
  mark_action_baseline
  tick_real_time
  invalidate_find_cache
  ensure_alive_after_tick(source_handle)
  ensure_alive_after_tick(target_handle)
  dom_call('__csimDragOnto', source_handle, target_handle)
  drain_after_user_action
end

#drain_after_user_action ⇒ Object

Every user-action entry point (set / send_keys / select / unselect) ends in this trio: drain any pending form submit, drain any pending Element.click anchor activation, then settle the page. Missing one site silently breaks the Stimulus filter pattern that wires ‘link.click()` into input/change/keypress listeners.



1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
 
# File 'lib/capybara/simulated/browser.rb', line 1372

def drain_after_user_action
  consume_pending_form_submit
  consume_pending_navigation
  settle
  # Settle bails on first observable change, but Backburner-style
  # 500 ms debounces park behind a setTimeout that hasn't fired
  # yet. Drain one 600 ms window so input → debounce → parent
  # state propagation completes before the next Capybara call.
  @runtime.drain_timers(USER_ACTION_DRAIN_MS) if @timers_active && @runtime.respond_to?(:drain_timers)
end

#drain_pending_navigation ⇒ Object 



3585
3586
3587
3588
3589
 
# File 'lib/capybara/simulated/browser.rb', line 3585

def drain_pending_navigation
  consume_pending_location
  consume_pending_reload
  consume_pending_history_traverse
end

#drop(handle, args) ⇒ Object

HTML5 drag-and-drop simulation. Capybara routes ‘Element#drop` here with a flat list of paths / Pathnames / Hashes; build a DataTransfer-shaped object and dispatch dragenter / dragover / drop in sequence.



1125
1126
1127
1128
1129
1130
1131
 
# File 'lib/capybara/simulated/browser.rb', line 1125

def drop(handle, args)
  tick_real_time
  invalidate_find_cache
  ensure_alive_after_tick(handle)
  items = args.flat_map {|arg| drop_items(arg) }
  dom_call('__csimDropOnto', handle, items)
end

#drop_items(arg) ⇒ Object 



1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
 
# File 'lib/capybara/simulated/browser.rb', line 1149

def drop_items(arg)
  case arg
  when Hash
    arg.map {|type, value| {'kind' => 'string', 'type' => type.to_s, 'value' => value.to_s} }
  when ->(x) { x.respond_to?(:to_path) }
    path = arg.to_path
    [{'kind' => 'file', 'name' => File.basename(path), 'path' => path}]
  when String
    [{'kind' => 'file', 'name' => File.basename(arg), 'path' => arg}]
  else
    []
  end
end

#drop_pending_transfers ⇒ Object

Release every outstanding transfer token’s backing store. The transfer registry is process-wide (it bridges isolates) and survives isolate teardown, so an unimported token would leak across the whole run; drop them on ‘reset!` via the (still-live) main context — `transferDrop` is idempotent, so dropping already-imported tokens is a harmless no-op.



3199
3200
3201
3202
3203
 
# File 'lib/capybara/simulated/browser.rb', line 3199

def drop_pending_transfers
  toks = @transfer_tokens_lock.synchronize { ts = @transfer_tokens; @transfer_tokens = []; ts }
  return if toks.empty?
  @runtime.call('__csimTransferDropAll', toks) rescue nil
end

#durable_source(url) ⇒ Object

Fetch a source body and report how long it stays safely reusable per its OWN response headers — an absolute freshness deadline (Time), or nil when the response is not durably cacheable (no-store / no-cache / max-age=0 / dynamic with no freshness). This lets a loader persist the body across visits and skip the round-trip next time, driven by the server’s cache directives (RFC 9111 §5.2.2 / §4.2.2 heuristic) — NOT a URL-shape guess. ‘clear_volatile` drops the body from the volatile per-visit asset cache, but a content-hashed asset’s source is content-stable while fresh, so a loader’s own cross-visit cache can hold it for ‘fresh_until`. Used by the external-asset cache (`external_asset_source`, scripts + stylesheets); name is generic.



2294
2295
2296
2297
2298
2299
2300
 
# File 'lib/capybara/simulated/browser.rb', line 2294

def durable_source(url)
  body = rack_fetch_body(url)
  return [nil, nil] unless body
  entry = @@asset_cache.lookup(url)
  fresh_until = entry && entry.fresh? && entry.max_age ? entry.stored_at + entry.max_age : nil
  [body, fresh_until]
end

#empty_find_result?(result) ⇒ Boolean

Returns:

  • (Boolean) 


693
694
695
 
# File 'lib/capybara/simulated/browser.rb', line 693

def empty_find_result?(result)
  result.nil? || (result.respond_to?(:empty?) && result.empty?)
end

#encode_image(pixels_ref, width, height, mime_type = 'image/png', quality = 90) ⇒ Object 



3283
3284
3285
3286
3287
3288
3289
3290
3291
3292
3293
3294
3295
 
# File 'lib/capybara/simulated/browser.rb', line 3283

def encode_image(pixels_ref, width, height, mime_type = 'image/png', quality = 90)
  host_image_op('encode_image') {
    require 'vips' unless defined?(Vips)
    raw = transfer_buffer_fetch(pixels_ref).to_s
    w   = width.to_i
    h   = height.to_i
    next nil if w <= 0 || h <= 0 || raw.bytesize < w * h * 4
    img = Vips::Image.new_from_memory_copy(raw, w, h, 4, :uchar)
    ext = MIME_TO_VIPS_EXT[mime_type.to_s.downcase] || '.png'
    opts = (ext == '.jpg' || ext == '.webp') ? {Q: quality.to_i} : {}
    {'refId' => transfer_buffer_stash(img.write_to_buffer(ext, **opts))}
  }
end

#enqueue_window_message(data, origin, source_handle) ⇒ Object

Queue a cross-window message for delivery into THIS window’s VM (called by the Driver on the target Browser). Delivered as a ‘message` event the next time this window settles / ticks.



3077
3078
3079
 
# File 'lib/capybara/simulated/browser.rb', line 3077

def enqueue_window_message(data, origin, source_handle)
  @window_inbox << {'data' => data, 'origin' => origin.to_s, 'sourceHandle' => source_handle.to_s}
end

#ensure_alive_after_tick(handle) ⇒ Object

‘tick_real_time` may have rebuilt the DOM (Ember route hydration finishing on its first idle tick replaces server-rendered nodes with fresh ones). `Node` ran check_stale before calling here, but that was BEFORE the tick — re-verify after so Capybara catches the now-stale handle and retries the find. Otherwise `__csim*` lookups would return null and the operation would silently no-op (or, in the case of `__csimClickResolve`, dispatch on a detached node whose listeners no longer matter).

Raises:



814
815
816
817
 
# File 'lib/capybara/simulated/browser.rb', line 814

def ensure_alive_after_tick(handle)
  return if dom_call('__csimAlive', handle)
  raise Capybara::Simulated::StaleElement, "Element with handle #{handle} is no longer attached to the document"
end

#eval_esm_module(url, src = nil) ⇒ Object

Native ESM entry point. QuickJS uses its ‘vm.module_loader`; V8 uses `Context#compile_module` + `Module#instantiate` / `#evaluate` + `Context#dynamic_import_resolver=`. Both runtimes expose `eval_esm_module`.



2350
2351
2352
 
# File 'lib/capybara/simulated/browser.rb', line 2350

def eval_esm_module(url, src = nil)
  @runtime.eval_esm_module(url, src)
end

#evaluate_async_script(code, args = []) ⇒ Object 



1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
 
# File 'lib/capybara/simulated/browser.rb', line 1867

def evaluate_async_script(code, args = [])
  tick_real_time
  invalidate_find_cache
  # Runs in the active frame realm inside `within_frame` (Selenium
  # parity), same as evaluate_script; the result slot is realm-local so
  # the poll below must read from the same realm.
  dom_call('__evalAsyncScript', code.to_s, marshal_args(args || []))
  # Pump virtual time so any setTimeout-driven completion lands.
  # Capybara's polling can't help here — we're inside one session
  # call, not a retry loop.
  deadline = Process.clock_gettime(Process::CLOCK_MONOTONIC) +
             Capybara.default_max_wait_time.to_f
  loop do
    result = dom_call('__pollAsyncResult')
    return result['value'] if result.is_a?(Hash) && result.key?('value')
    break if Process.clock_gettime(Process::CLOCK_MONOTONIC) >= deadline
    sleep 0.01
    tick_real_time
  end
  nil
end

#evaluate_script(code, args = []) ⇒ Object 



1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
 
# File 'lib/capybara/simulated/browser.rb', line 1791

def evaluate_script(code, args = [])
  # Drain timers first so ready handlers (jQuery `$(handler)`,
  # framework `DOMContentLoaded` listeners) run before the
  # user's script. Without this, `execute_script` can fire
  # *before* the page's own setup code that the test expects
  # to be active.
  tick_real_time
  invalidate_find_cache
  # Routes to the active frame realm inside `within_frame` (Selenium
  # parity: `evaluate_script` runs in the current browsing context).
  result = dom_call('__csimEvalScript', code.to_s, marshal_args(args || []))
  drain_pending_navigation
  result
end

#event_source_close(id) ⇒ Object 



2385
2386
2387
2388
2389
 
# File 'lib/capybara/simulated/browser.rb', line 2385

def event_source_close(id)
  thread = @event_source_threads.delete(id.to_i)
  thread&.kill
  nil
end

#event_source_open(url) ⇒ Object

── EventSource (SSE) ──────────────────────────────────────────

Mastodon (and any app using Server-Sent Events) opens an ‘EventSource` to a streaming endpoint and expects pushed events to fire `message`/typed listeners on the live instance. Our implementation:

1. JS-side `new EventSource(url)` calls `__csim_eventSourceOpen`
   which returns an integer handle and spawns a Ruby thread.
2. The thread holds a chunked-read HTTP connection open and
   parses the SSE event-stream wire format, pushing each
   `{id:, type:, data:, lastEventId:}` (or `{type: '__open'}`
   / `{type: '__error', message:}` sentinel) onto a
   thread-safe queue.
3. Settle's drain loop calls `deliver_event_source_events`
   which polls the queue and hands the batch to
   `__csim_deliverEventSourceEvents` for dispatch.

rusty_racer / quickjs.rb VMs are single-threaded; only the main thread ever enters the VM. Background threads only touch the Queue. ‘reset!` and per-visit context rebuilds kill all open threads — the new VM gets a fresh handle space.



2374
2375
2376
2377
2378
2379
2380
2381
2382
2383
 
# File 'lib/capybara/simulated/browser.rb', line 2374

def event_source_open(url)
  id     = (@event_source_seq += 1)
  queue  = @event_source_queue
  thread = Thread.new do
    Thread.current.report_on_exception = false
    run_event_source_reader(id, url.to_s, queue)
  end
  @event_source_threads[id] = thread
  id
end

#event_source_pending? ⇒ Boolean

Returns:

  • (Boolean) 


2395
 
# File 'lib/capybara/simulated/browser.rb', line 2395

def event_source_pending? = !@event_source_queue.empty?

#event_source_poll ⇒ Object 



2391
2392
2393
 
# File 'lib/capybara/simulated/browser.rb', line 2391

def event_source_poll
  drain_queue(@event_source_queue)
end

#execute_script(code, args = []) ⇒ Object

Fire-and-forget variant: runs the script but never returns its value to Ruby. Lets execute_script handle scripts whose return is a complex JS object (jQuery chainable, DOM tree, …) that the marshaller would recurse into.



1810
1811
1812
1813
1814
1815
1816
 
# File 'lib/capybara/simulated/browser.rb', line 1810

def execute_script(code, args = [])
  tick_real_time
  invalidate_find_cache
  dom_call('__csimExecScript', code.to_s, marshal_args(args || []))
  drain_pending_navigation
  nil
end

#external_asset_source(url) ⇒ Object

Body of an external durably-cacheable asset (classic script or stylesheet), served from the cross-visit cache when still fresh, else fetched (which read-throughs the per-visit asset cache) and cached iff durably cacheable. Returns nil on 4xx / fetch failure so the JS caller skips it exactly as the old ‘__rackFetch` branch did.



2320
2321
2322
2323
2324
2325
2326
2327
2328
2329
2330
2331
2332
2333
2334
2335
2336
2337
2338
2339
2340
2341
2342
2343
2344
 
# File 'lib/capybara/simulated/browser.rb', line 2320

def external_asset_source(url)
  key = resolve_against_current(url.to_s)
  return nil unless key.is_a?(String)
  @@asset_src_lock.synchronize do
    if (e = @@asset_src[key])
      return e[0] if e[1].nil? || Time.now < e[1]
      @@asset_src.delete(key)
    end
  end
  # `durable_source` already does the spec-compliant fetch + header-driven
  # freshness (RFC 9111 max-age → absolute deadline); reuse it instead of
  # re-deriving `fresh_until` here.
  body, fresh_until = durable_source(key)
  return nil unless body
  # Script / stylesheet source is TEXT, but the raw Rack / binread body
  # arrives BINARY-tagged (see `RuntimeShared.utf8_text`).
  body = RuntimeShared.utf8_text(body)
  if fresh_until
    @@asset_src_lock.synchronize do
      @@asset_src.clear if @@asset_src.size >= ASSET_SRC_MAX
      @@asset_src[key] = [body, fresh_until]
    end
  end
  body
end

#file_input?(handle) ⇒ Boolean

Returns:

  • (Boolean) 


759
760
761
 
# File 'lib/capybara/simulated/browser.rb', line 759

def file_input?(handle)
  tag(handle) == 'input' && attr(handle, 'type').to_s.downcase == 'file'
end

#file_picks_for(handle) ⇒ Object 



1083
1084
1085
 
# File 'lib/capybara/simulated/browser.rb', line 1083

def file_picks_for(handle)
  (@file_picks && @file_picks[handle]) || []
end

#find_css(css, context_handle = nil) ⇒ Object 



614
615
616
617
618
619
620
621
622
623
624
625
 
# File 'lib/capybara/simulated/browser.rb', line 614

def find_css(css, context_handle = nil)
  s = css.to_s
  return find_xpath(s, context_handle) if xpath_shaped?(s)
  find_with_timer_fallback(:css, s, context_handle) do
    dom_call('__csimQuery', context_handle || current_document_handle, s).to_a
  rescue StandardError => e
    # Invalid selector → empty result. Callers that genuinely
    # need the throw go through `evaluate_script`.
    raise unless syntax_or_invalid_selector_error?(e)
    []
  end
end

#find_first_css(css, context_handle = nil) ⇒ Object 



627
628
629
630
631
632
633
634
635
636
 
# File 'lib/capybara/simulated/browser.rb', line 627

def find_first_css(css, context_handle = nil)
  s = css.to_s
  find_with_timer_fallback(:css_first, s, context_handle) do
    h = dom_call('__csimQueryOne', context_handle || current_document_handle, s).to_i
    h.zero? ? nil : h
  rescue StandardError => e
    raise unless syntax_or_invalid_selector_error?(e)
    nil
  end
end

#find_with_timer_fallback(kind, arg, ctx) ⇒ Object 



670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
 
# File 'lib/capybara/simulated/browser.rb', line 670

def find_with_timer_fallback(kind, arg, ctx)
  tick_real_time if timer_wait_elapsed?
  result = cached_find(kind, arg, ctx) { yield }
  # An empty result is the wait-for-it case: Capybara is retrying for
  # an element that hasn't appeared yet. Re-tick so the next poll
  # observes anything an active timer OR a background-IO channel
  # (Worker / EventSource / a held long-poll publish) is about to
  # deliver. Gating on `@timers_active` alone misses the held-poll
  # case — a MessageBus subscription waiting on a cross-session
  # publish has NO pending JS timer (the re-poll only schedules after
  # the current poll returns), so `@timers_active` is false while
  # `hijack_fetch_pending?` is true. Without `async_io_pending?` here
  # the delivered message never reaches the DOM during find-polling
  # (only `evaluate_script`, which ticks unconditionally, would see
  # it). Non-empty results keep the fast path — no extra tick.
  return result unless empty_find_result?(result) && (@timers_active || async_io_pending?)

  tick_real_time
  return result unless @find_cache_dirty

  cached_find(kind, arg, ctx) { yield }
end

#find_xpath(xpath, context_handle = nil) ⇒ Object

XPath is evaluated inside V8 against the live JS DOM via the xpathway engine (bundled, installed at snapshot build). One IPC per ‘find_xpath` — no serialise + reparse round-trip.



663
664
665
666
667
668
 
# File 'lib/capybara/simulated/browser.rb', line 663

def find_xpath(xpath, context_handle = nil)
  xpath_str = xpath.to_s
  find_with_timer_fallback(:xpath, xpath_str, context_handle) do
    dom_call('__csimEvaluateXPath', xpath_str, context_handle || 0).to_a
  end
end

#finish_trace_to(path, trace = (@trace || @pending_trace)) ⇒ Object

Persist ‘trace` (defaults to live or pending) to `path` and return the path. Doesn’t clear — ‘clear_trace!` is the explicit follow-up so a caller can inspect after writing if it wants.



1700
1701
1702
1703
 
# File 'lib/capybara/simulated/browser.rb', line 1700

def finish_trace_to(path, trace = (@trace || @pending_trace))
  return nil unless trace
  trace.write_json(path)
end

#format_temporal_value(v, handle) ⇒ Object 



1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
 
# File 'lib/capybara/simulated/browser.rb', line 1070

def format_temporal_value(v, handle)
  type = attr(handle, 'type').to_s.downcase
  case type
  when 'date'           then v.respond_to?(:strftime) ? v.strftime('%Y-%m-%d') : v.to_s
  when 'time'           then v.respond_to?(:strftime) ? v.strftime('%H:%M') : v.to_s
  when 'datetime-local' then v.respond_to?(:strftime) ? v.strftime('%Y-%m-%dT%H:%M') : v.to_s
  when 'month'          then v.respond_to?(:strftime) ? v.strftime('%Y-%m')  : v.to_s
  when 'week'           then v.respond_to?(:strftime) ? v.strftime('%Y-W%V') : v.to_s
  else
    v.is_a?(Date) ? v.strftime('%Y-%m-%d') : v.to_s
  end
end

#frame_self_target?(target) ⇒ Boolean

Does a link/form ‘target` load into the CURRENT frame? Empty or `_self` do; `_top` / `_blank` / `_parent` / a named context do not. `_top` correctly navigates the main page (it falls through to `navigate`). `_parent` from a frame nested ≥2 levels would ideally navigate the intermediate parent frame, not the top page — that ancestor-targeted case isn’t modelled yet (rare); it currently navigates the main page.

Returns:

  • (Boolean) 


609
610
611
612
 
# File 'lib/capybara/simulated/browser.rb', line 609

def frame_self_target?(target)
  t = target.to_s.downcase
  t.empty? || t == '_self'
end

#geolocation_state_json ⇒ Object

Backs the ‘__csimGeolocationState` host fn. Returns the configured geolocation override as a JSON string (or ’null’ when none is set), which the JS geolocation object reads on every getCurrentPosition / watchPosition call.



1849
1850
1851
 
# File 'lib/capybara/simulated/browser.rb', line 1849

def geolocation_state_json
  JSON.generate(@geolocation)
end

#go_back ⇒ Object

Capybara-initiated ‘page.go_back` runs from Ruby, not inside a JS call, so it’s safe to rebuild the Context synchronously. The ‘force:` flag bypasses the deferral that `history_go` uses to avoid terminating the running JS context.



1576
 
# File 'lib/capybara/simulated/browser.rb', line 1576

def go_back        ; history_go(-1, force: true) ; end

#go_forward ⇒ Object 



1577
 
# File 'lib/capybara/simulated/browser.rb', line 1577

def go_forward     ; history_go(+1, force: true) ; end

#handle_modal(type, message, default_value) ⇒ Object

JS-side ‘alert(…)` / `confirm(…)` / `prompt(…)` route here. If no handler is pushed (typical of apps under test), accept the dialog (Rails system-test default) so `data-turbo-confirm` / similar progress without an explicit `accept_confirm` in the test.



3703
3704
3705
3706
3707
3708
3709
3710
3711
3712
3713
3714
 
# File 'lib/capybara/simulated/browser.rb', line 3703

def handle_modal(type, message, default_value)
  handler = @modal_handlers.pop
  if handler
    handler.call(type, message, default_value)
  else
    case type.to_s
    when 'alert'   then nil
    when 'confirm' then true
    when 'prompt'  then default_value.to_s
    end
  end
end

#hijack_fetch_pending? ⇒ Boolean

Returns:

  • (Boolean) 


2894
 
# File 'lib/capybara/simulated/browser.rb', line 2894

def hijack_fetch_pending? = !@hijack_fetch_threads.empty? || !@hijack_fetch_queue.empty?

#history_go(delta, force: false) ⇒ Object

Move through the history stack by ‘delta`. Per HTML spec, a same-document traversal (within a chain of pushState entries rooted at a single navigation) updates `location` and fires `popstate` with the entry’s state — no full reload. A cross- document traversal replays the entry (full navigate / re-POST).



1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
 
# File 'lib/capybara/simulated/browser.rb', line 1584

def history_go(delta, force: false)
  delta = delta.to_i
  return if delta == 0
  target = @history_idx + delta
  return if target < 0 || target >= @history.size
  if same_document_traversal?(@history_idx, target)
    # Pure pushState traversal — no VM rebuild, safe to run
    # inline; the popstate dispatch happens within the current
    # call's JS context.
    @history_idx = target
    entry = @history[target]
    @current_url = entry[:url]
    @runtime.call('__csimUpdateLocation', @current_url)
    @runtime.call('__csimDispatchPopState', entry[:state])
  elsif force
    # Ruby-driven (`page.go_back`) — no live JS call to interrupt,
    # safe to rebuild the Context synchronously.
    perform_history_traverse(target)
  else
    # JS-driven (`history.back()` from a page handler): replaying
    # the history entry synchronously would call `rebuild_ctx`
    # on the still-executing Context and terminate the current
    # call with `ScriptTerminatedError` (terminating the
    # in-flight call on the isolate). Stash the intent
    # and drain after the call returns — mirrors
    # `location_assign` / `location_reload`.
    @pending_history_traverse = target
  end
end

#history_length ⇒ Object

Total history entries (after forward-tail truncation), surfaced to JS ‘history.length` via the `__historyLength` host fn.



3632
3633
3634
 
# File 'lib/capybara/simulated/browser.rb', line 3632

def history_length
  [@history.size, 1].max
end

#history_push(url, state = nil) ⇒ Object

‘history.pushState(state, _, url)` from SPA navigation (Turbo Visit, InstantClick, …) appends a new browser-history entry. Mirror that on the Ruby side so `Capybara#go_back` traverses within the pushState chain (fires `popstate`) and only crosses to a real reload when the back hits a `:visit` boundary.



3623
3624
3625
3626
3627
3628
 
# File 'lib/capybara/simulated/browser.rb', line 3623

def history_push(url, state = nil)
  resolved = resolve_against_current(url.to_s)
  record_url_transition(resolved)
  @current_url = resolved
  record_history({method: :get, url: resolved, state: state, kind: :push_state})
end

#history_state(url, state = nil) ⇒ Object

‘history.pushState(state, ”, ’/path’)‘ ships the URL through `__setCurrentUrl` and lands here. Tab controllers / SPA frameworks pass a relative path; resolve it against the existing absolute `@current_url` so subsequent `resolve_against_current(href)` calls (e.g. click_link to a relative href) don’t hit ‘URI::BadURIError: both URI are relative`. `history.replaceState(state, _, url)` updates the current entry in place rather than appending. Both the state and (when given) the URL are mirrored on Ruby’s slot so a subsequent back to this entry restores the same state.



3605
3606
3607
3608
3609
3610
3611
3612
3613
3614
3615
3616
3617
 
# File 'lib/capybara/simulated/browser.rb', line 3605

def history_state(url, state = nil)
  if url
    resolved = resolve_against_current(url.to_s)
    record_url_transition(resolved)
    @current_url = resolved
  end
  return if @history_idx < 0
  @history[@history_idx] = (@history[@history_idx] || {}).merge(
    url:   @current_url,
    state: state,
    kind:  @history[@history_idx] ? @history[@history_idx][:kind] : :push_state
  )
end

#horizon_fast_forward_step ⇒ Object

This tick’s deterministic virtual-clock advance (ms). Default is the fixed ‘POLL_TICK_STEP_MS` — never wall-derived, so per-poll JS/Ruby/GC cost cannot shift WHEN a timer fires (the wall-sync↔perf coupling this replaces). When the page is observably idle (nothing runnable now, no background IO) but a near-future timer is parked within `FF_HORIZON_MS`, fast-forward straight to it — but only after the transient-guard window so pre-debounce states are still observed across several polls. `FF_HORIZON_MS=0` ⇒ pure fixed-step.



2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
2026
2027
2028
2029
2030
2031
2032
2033
2034
2035
2036
2037
2038
2039
2040
2041
2042
2043
2044
2045
2046
2047
2048
2049
2050
2051
2052
2053
2054
 
# File 'lib/capybara/simulated/browser.rb', line 2011

def horizon_fast_forward_step
  # Escape hatch to the legacy wall-sync clock (virtual advance = real
  # wall-elapsed per poll). The deterministic model decouples perf from
  # timing but can't match a real browser's wall-proportional cadence for
  # timing-fragile heavy-JS flows; `CSIM_CLOCK_WALL=1` restores wall-sync.
  if @clock_wall
    now = Process.clock_gettime(Process::CLOCK_MONOTONIC)
    step = ((now - (@wall_clock_last || now)) * 1000).to_i.clamp(0, 1000)
    @wall_clock_last = now
    return step
  end
  # (1) Background async (cheap Ruby-side checks, no V8 crossing) we must let
  #     land before jumping the clock: advance one fixed step, reset the guard.
  if worker_pending? || event_source_pending? || hijack_fetch_pending? || websocket_pending?
    @ff_transient_polls = 0
    return POLL_TICK_STEP_MS
  end
  # No fast-forward support on this runtime (e.g. a worker realm) → fixed step.
  return POLL_TICK_STEP_MS unless @runtime_supports_ff
  # ONE V8 crossing: `delay` = ms until the nearest timer; 0 = runnable now
  # (a rAF or a due-now timer — equivalent to `has_ready_timer?`), -1 = none.
  delay = @runtime.next_timer_delay_ms
  # (2) Runnable now → fixed step, reset guard (not a quiet pre-debounce window).
  if delay.zero?
    @ff_transient_polls = 0
    return POLL_TICK_STEP_MS
  end
  # (3) Nothing parked → nothing to fast-forward to.
  return POLL_TICK_STEP_MS if delay.negative?
  # (4) Beyond the horizon (ahoy 1000 / session-timeout / analytics): leave
  #     parked, advance only at the fixed rate. Not a transient window.
  if delay > FF_HORIZON_MS
    @ff_transient_polls = 0
    return POLL_TICK_STEP_MS
  end
  # (5) Near-future timer, page idle: hold the pre-debounce window for the
  #     guard so transient-catch tests observe the intermediate state.
  @ff_transient_polls = (@ff_transient_polls || 0) + 1
  return POLL_TICK_STEP_MS if @ff_transient_polls < FF_TRANSIENT_GUARD_POLLS
  # (6) Fast-forward: jump exactly to the next timer's due. `runLoopStepLocal`
  #     breaks on strict `nextDue > limit`, so `limit = virtualNow + delay`
  #     (== that timer's due) fires it — and ONLY it, not a timer 1 ms later.
  delay
end

#hover(handle) ⇒ Object 



1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
 
# File 'lib/capybara/simulated/browser.rb', line 1220

def hover(handle)
  mark_action_baseline
  tick_real_time
  invalidate_find_cache
  ensure_alive_after_tick(handle)
  # Set `document._hoverElement` so `:hover` pseudo-class matches
  # resolve against this element (Redmine's gantt tooltips +
  # context-menu submenus rely on CSS `:hover`). The host fn
  # call into `__csimSetHover` does the slot update on the JS
  # side AND fires `mouseover` / `mouseenter` — keeping the
  # state-set and dispatch on the same path avoids the
  # double-eval recursion the inlined `globalThis.document.
  # _hoverElement = ...` triggered (the eval string ran inside
  # a fresh microtask that re-entered the hover listeners).
  dom_call('__csimSetHover', handle)
end

#html ⇒ Object

‘page.html` inside a `within_frame` block returns the frame document’s source (Selenium parity), so route through the active realm.



1497
1498
1499
1500
 
# File 'lib/capybara/simulated/browser.rb', line 1497

def html
  tick_real_time
  dom_call('__csimDocumentHtml').to_s
end

#inner_html(handle) ⇒ Object 



757
 
# File 'lib/capybara/simulated/browser.rb', line 757

def inner_html(handle)  = dom_call('__csimInnerHTML', handle).to_s

#invalidate_find_cache ⇒ Object

Any operation that may have mutated the DOM (click, set, send_keys, navigate, hover, …) must call this so the next find falls through to a fresh V8 query. Timer drains that fire any callbacks also dirty (see ‘tick_real_time`).



750
751
752
 
# File 'lib/capybara/simulated/browser.rb', line 750

def invalidate_find_cache
  @find_cache_dirty = true
end

#location_assign(url) ⇒ Object

Defer the navigation: doing it from inside the running V8 call would dispose the Context mid-call. tick_real_time drains after the call returns. Same pattern as ‘__csimPendingFormSubmit`.



3556
3557
3558
 
# File 'lib/capybara/simulated/browser.rb', line 3556

def location_assign(url)
  @pending_location = resolve_against_current(url.to_s)
end

#location_reload ⇒ Object

Mirror of ‘location_assign`’s deferral for ‘location.reload()`: the JS call lands here from `__locationReload`; running `browser.refresh` directly would `navigate` (rebuilding the Context) while we’re still inside the V8 call, which V8 terminates with a ‘ScriptTerminatedError`. Stash the intent and drain it from `tick_real_time` after the call returns.



3579
 
# File 'lib/capybara/simulated/browser.rb', line 3579

def location_reload   ; @pending_reload = true ; end

#log_console(severity, message) ⇒ Object 



1755
1756
1757
1758
1759
1760
1761
 
# File 'lib/capybara/simulated/browser.rb', line 1755

def log_console(severity, message)
  # Diagnostic mirror: surface page console output on stderr regardless
  # of trace state (engine bring-up / CI triage).
  warn "[console:#{severity}] #{message.to_s[0, 300]}" if CONSOLE_STDERR
  return unless @trace
  @trace.log_console(severity, annotate_console_message(severity, message))
end

#log_network(method, url, status) ⇒ Object 



1776
 
# File 'lib/capybara/simulated/browser.rb', line 1776

def log_network(method, url, status) = @trace&.log_network(method, url, status)

#lookup_node(handle) ⇒ Object 



793
794
795
 
# File 'lib/capybara/simulated/browser.rb', line 793

def lookup_node(handle)
  handle if dom_call('__csimAlive', handle)
end

#mark_action_baseline ⇒ Object

Pin the URL the page is at as a user action BEGINS — the FIRST line of every action entry (click / double_click / right_click / hover / set / send_keys / select / unselect). A Turbo Drive Visit the action triggers is async — its pushState may fire synchronously mid-action or in a LATER find-poll tick (the test’s ‘wait_for_loaded`) — so `record_url_transition` uses this baseline to recognise the pre-action URL as the action’s starting point, not a walkable intermediate, and skip queuing it. Set at action entry (NOT the tail drain, which runs after the pushState); must precede the action’s first ‘tick_real_time` so a deferred prior-page timer firing in that tick is still measured against the pre-action URL. Persists until the next action (so the async case is covered) and is reset by `navigate` so a stale baseline can’t leak across a document boundary.



1363
1364
1365
 
# File 'lib/capybara/simulated/browser.rb', line 1363

def mark_action_baseline
  @action_url_baseline = @current_url
end

#marshal_args(args) ⇒ Object

Capybara passes Node instances directly as script args (‘session.evaluate_script(’arguments.click()‘, some_node)`). the marshaller can’t pass a Ruby Node, so wrap as a sentinel the JS side recognises and rehydrates via the handle registry.



1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
 
# File 'lib/capybara/simulated/browser.rb', line 1857

def marshal_args(args)
  args.map {|a|
    case a
    when Capybara::Simulated::Node then {'__elementHandle' => a.handle_id}
    when Array                       then marshal_args(a)
    when Hash                        then a.transform_values {|v| marshal_args([v]).first }
    else a
    end
  }
end

#mime_type_for_path(path) ⇒ Object 



165
166
167
 
# File 'lib/capybara/simulated/browser.rb', line 165

def mime_type_for_path(path)
  Rack::Mime.mime_type(File.extname(path.to_s), '')
end

#modifier_flags(keys) ⇒ Object 



1192
1193
1194
1195
1196
1197
 
# File 'lib/capybara/simulated/browser.rb', line 1192

def modifier_flags(keys)
  Array(keys).each_with_object({}) {|k, h|
    field = MODIFIER_KEYS[k.is_a?(Symbol) ? k : k.to_sym]
    h[field] = true if field
  }
end

#navigate_post(url, body, content_type, depth: 0, from_history: false) ⇒ Object 



2155
2156
2157
2158
2159
2160
2161
2162
2163
2164
2165
2166
2167
2168
2169
2170
2171
2172
2173
2174
2175
2176
2177
2178
2179
2180
2181
2182
2183
2184
2185
2186
2187
2188
2189
 
# File 'lib/capybara/simulated/browser.rb', line 2155

def navigate_post(url, body, content_type, depth: 0, from_history: false)
  raise 'too many redirects' if depth > 10
  invalidate_find_cache
  record_history({method: :post, url: url, body: body, content_type: content_type}) unless from_history || depth > 0
  env = Rack::MockRequest.env_for(url, method: 'POST', input: body)
  env['CONTENT_TYPE']   = content_type.empty? ? 'application/x-www-form-urlencoded' : content_type
  env['CONTENT_LENGTH'] = body.bytesize.to_s
  apply_default_request_env(env, referer: @current_url)
  status, headers, resp_body = dispatch_rack_or_http(url, env, method: 'POST', body: body)
  merge_set_cookie(headers)
  if (loc = redirect_location(status, headers))
    next_url = resolve_against_current(loc)
    resp_body.close if resp_body.respond_to?(:close)
    # HTTP semantics: 301/302/303 → method becomes GET; 307/308
    # require the method (and body) to be preserved.
    if [307, 308].include?(status)
      return navigate_post(next_url, body, content_type, depth: depth + 1)
    else
      return navigate(next_url, depth: depth + 1)
    end
  end
  if download_response?(headers)
    save_downloaded_response(url, headers, resp_body)
    return
  end
  @current_url = url
  record_response(status, headers)
  html         = read_rack_body(resp_body)
  # Same rebuild-on-full-load contract as `navigate`. POST
  # responses (form submissions that don't redirect, AJAX-less
  # data-remote replies) replace the page; we follow real-browser
  # semantics and bring up a fresh VM rather than papering over
  # the previous one's state.
  boot_response_into_ctx(html)
end

#node_path(handle) ⇒ Object 



791
 
# File 'lib/capybara/simulated/browser.rb', line 791

def node_path(handle)    = dom_call('__csimNodePath', handle).to_s

#open_child_window(url, name) ⇒ Object

‘window.open(url, name)` from JS — returns the new (or reused, by name) window’s handle, or nil. The URL is resolved against THIS document so a relative ‘window.open(’/x’)‘ targets the right origin/path.



3056
3057
3058
3059
 
# File 'lib/capybara/simulated/browser.rb', line 3056

def open_child_window(url, name)
  return nil unless @driver.respond_to?(:open_window_from_js)
  @driver.open_window_from_js(self, url.to_s, name.to_s)
end

#opener_handle ⇒ Object 



3072
 
# File 'lib/capybara/simulated/browser.rb', line 3072

def opener_handle                = @driver.respond_to?(:opener_handle_of)    ? @driver.opener_handle_of(self)                : nil

#option_selected?(h) ⇒ Boolean

HTML spec: ‘<option>.selected` IDL is true when the `selected` attribute is set OR when no sibling option has `selected` and this is the first non-disabled option of a single-select `<select>` (implicit default). Capybara’s ‘have_select(selected: “Choose an option”)` filter calls `selected?` on each option; without the implicit-default branch, a select with no explicit `<option selected>` reports no selected options and the matcher fails even though the first option is the currently chosen one in real browsers.

Returns:

  • (Boolean) 


780
 
# File 'lib/capybara/simulated/browser.rb', line 780

def option_selected?(h)  = !!dom_call('__csimOptionSelected', h)

#outer_html(handle) ⇒ Object 



758
 
# File 'lib/capybara/simulated/browser.rb', line 758

def outer_html(handle)  = dom_call('__csimOuterHTML', handle).to_s

#parse_trace_mode(raw) ⇒ Object 



1687
1688
1689
1690
 
# File 'lib/capybara/simulated/browser.rb', line 1687

def parse_trace_mode(raw)
  return :on_failure if raw.nil? || raw.empty?
  TRACE_MODES[raw] || raise(ArgumentError, "CSIM_TRACE must be one of #{TRACE_MODES.keys.join(', ')}; got #{raw.inspect}")
end

#polling? ⇒ Boolean

Capybara polls find / has_? via ‘synchronize` while `Driver#wait?` is true. We stay true while there’s any scheduled timer (‘@timers_active` is flipped by the JS bridge’s ‘__setTimersActive` callback), plus a sticky grace window after the last timer fires so a `setTimeout` firing mid-loop doesn’t drop us off polling before Capybara’s own retry deadline.

Settle-gen idle gate: a recurring ‘setInterval` from a framework runloop (Ember / Glimmer) keeps `@timers_active` true forever even when nothing observable is changing. Without a second signal, Capybara waits the full `default_max_wait_time` on every `has_css?` / `has_no_css?` that’s destined to fail — which Discourse’s ‘CapybaraTimeoutExtension` reports as a “slow spec” failure. Track `@runtime.settle_gen` across polls: when it hasn’t bumped for ‘IDLE_SETTLE_POLLS` calls, drop polling even though timers are scheduled. `settle_gen` already bumps on every DOM mutation / URL change (see __settleGen wiring), so this only short-circuits genuinely idle loops.

Returns:

  • (Boolean) 


1915
1916
1917
1918
1919
1920
1921
1922
1923
1924
1925
1926
1927
1928
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
 
# File 'lib/capybara/simulated/browser.rb', line 1915

def polling?
  # Background-thread work (workers, EventSource, MessageBus
  # long-poll) keeps the settle loop alive even when settle_gen
  # is otherwise idle.
  return true if worker_pending? || event_source_pending? || hijack_fetch_pending? || window_message_pending? || websocket_pending?
  if @timers_active
    gen = @runtime.settle_gen
    if @last_polled_gen.nil? || gen != @last_polled_gen
      @last_polled_gen = gen
      @idle_settle_polls = 0
      @polling_grace = POLLING_GRACE_POLLS
      return true
    end
    @idle_settle_polls += 1
    return true if @idle_settle_polls < IDLE_SETTLE_POLLS
    # Treat as idle for this poll; if a fresh timer fires later
    # the next poll's settle_gen check will resume polling.
    false
  elsif @polling_grace && @polling_grace > 0
    @polling_grace -= 1
    true
  else
    false
  end
end

#post_message_to_window(target_handle, data, origin) ⇒ Object

‘targetWindow.postMessage(data, origin)` — route to the target window’s inbox, tagged with this window as the source.



3063
3064
3065
3066
 
# File 'lib/capybara/simulated/browser.rb', line 3063

def post_message_to_window(target_handle, data, origin)
  return unless @driver.respond_to?(:window_post_message)
  @driver.window_post_message(self, target_handle.to_s, data, origin.to_s)
end

#pure_fragment_navigation?(url) ⇒ Boolean

Returns:

  • (Boolean) 


965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
 
# File 'lib/capybara/simulated/browser.rb', line 965

def pure_fragment_navigation?(url)
  return true  if url.start_with?('#')
  doc_url = current_browsing_context_url
  return false if doc_url.nil?
  target = resolve_against_current(url)
  a = URI.parse(target)
  b = URI.parse(doc_url)
  # Same-document iff everything but the fragment matches AND the
  # fragment actually changes — `a.fragment != b.fragment` covers
  # both adding/changing a fragment and *clearing* one (target has
  # no fragment while the current URL does, e.g. `location.hash =
  # ''`). The old `!a.fragment.nil?` missed the clearing case, so a
  # hash-reset turned into a full document reload.
  a.scheme == b.scheme && a.host == b.host && a.port == b.port &&
    a.path == b.path && a.query == b.query && a.fragment != b.fragment
rescue URI::InvalidURIError
  false
end

#push_user_agent_to_js ⇒ Object 



1540
1541
1542
1543
1544
 
# File 'lib/capybara/simulated/browser.rb', line 1540

def push_user_agent_to_js
  ua = @default_user_agent or return
  return unless @runtime
  @runtime.eval("try { Object.defineProperty(navigator, 'userAgent', { value: #{ua.to_json}, configurable: true }); } catch (_) {}")
end

#rack_fetch(method, url, body, headers, redirect_mode, env_extras: nil) ⇒ Object

URLs we won’t even try to route through Rack: anything that isn’t http(s) (data: / mailto: / about:) plus pseudo-tokens like V8’s ‘<snapshot>` that sourcemap libraries pull out of error stacks and feed straight to `fetch()` / `xhr.open()`.



3410
3411
3412
3413
3414
3415
3416
3417
3418
3419
3420
3421
3422
3423
3424
3425
3426
3427
3428
3429
3430
3431
3432
3433
3434
3435
3436
3437
3438
3439
3440
3441
3442
3443
3444
3445
3446
3447
3448
3449
3450
3451
3452
3453
3454
3455
3456
3457
3458
3459
3460
3461
3462
3463
3464
 
# File 'lib/capybara/simulated/browser.rb', line 3410

def rack_fetch(method, url, body, headers, redirect_mode, env_extras: nil)
  target = resolve_against_current(url.to_s)
  return nil unless target.is_a?(String) && target.match?(%r{\Ahttps?://}i)
  method = (method || 'GET').to_s.upcase
  redirected = false
  # JS-side base64-encodes Blob/File bodies (raw bytes survive
  # the engine's UTF-8 string boundary that way); decode before
  # handing to Rack so the upload PUT lands intact.
  if headers.is_a?(Hash) && headers['X-Csim-Body-B64'].to_s == '1'
    body = Base64.decode64(body.to_s)
    headers = headers.reject {|k, _| k == 'X-Csim-Body-B64' }
  end
  MAX_FETCH_REDIRECTS.times do
    # GET-only cache shortcut (RFC 9111). Fresh hit → skip @app.call
    # entirely; stale-but-revalidatable → fall through with conditional
    # headers added so the server can return 304.
    cache_entry = method == 'GET' ? @@asset_cache.lookup(target) : nil
    if cache_entry&.fresh?
      log_network(method, target, cache_entry.status)
      return response_hash(cache_entry.status, cache_entry.headers, cache_entry.body, target, redirected)
    end

    env = Rack::MockRequest.env_for(target, method: method, input: body || '')
    apply_request_headers(env, headers) if headers
    apply_request_headers(env, @@asset_cache.revalidation_headers(cache_entry)) if cache_entry
    apply_default_request_env(env, referer: @current_url, force: false)
    env.merge!(env_extras) if env_extras
    status, resp_headers, resp_body = dispatch_rack_or_http(target, env, method: method, body: body)
    merge_set_cookie(resp_headers)
    log_network(method, target, status)
    if status == 304 && cache_entry
      resp_body.close if resp_body.respond_to?(:close)
      @@asset_cache.refresh(cache_entry, resp_headers)
      return response_hash(cache_entry.status, cache_entry.headers, cache_entry.body, target, redirected)
    end
    if redirect_mode != 'manual' && (loc = redirect_location(status, resp_headers))
      raise StandardError, '[capybara-simulated] fetch: redirect blocked by redirect=error mode' if redirect_mode == 'error'
      redirected = true
      preserve = [307, 308].include?(status)
      next_url = resolve_against(loc, target)
      target = carry_fragment(target, next_url)
      method = 'GET' unless preserve
      body = nil unless preserve
      resp_body.close if resp_body.respond_to?(:close)
      next
    end
    body_str = read_rack_body(resp_body)
    @@asset_cache.store(target, status, resp_headers, body_str) if method == 'GET'
    return response_hash(status, resp_headers, body_str, target, redirected)
  end
  raise StandardError, "[capybara-simulated] fetch exceeded #{MAX_FETCH_REDIRECTS} redirects"
rescue StandardError => e
  warn "[capybara-simulated] rack_fetch failed: #{e.class}: #{e.message[0, 200]}"
  nil
end

#rack_fetch_async(method, url, body, headers_json) ⇒ Object

── Hijack-aware async XHR ─────────────────────────────────────

Real browsers’ long-poll keeps the request socket open across the entire user-interactive session, so a server-side ‘MessageBus.publish` (or any other middleware writing through `rack.hijack`) lands on the open connection and the client gets the response when the server is ready. Our default `__rackFetch` is purely sync — the middleware’s hijack path never engaged, so MessageBus’s ‘subscribe(channel, -1)` + `__status` reset chain dropped any publish that landed between two scheduled polls.

‘rack_fetch_async` runs the Rack call with a `rack.hijack` lambda installed. The lambda is invoked iff the middleware actually hijacks; we detect that and spawn a background thread to read from the pipe until the middleware closes its end (a publish landed via `notify_clients`, or `cleanup_timer` fired the empty-`[]` close after `long_polling_interval`). Non-hijacking responses queue immediately on the same thread — no thread spawn, no backpressure beyond the existing sync `__rackFetch` cost.

The contract is generic: any middleware that follows the Rack hijack protocol works, not just ‘message_bus`. JS-side XHR’s async path routes every request here; sync XHRs (‘xhr.open(_, _, false)`, deprecated) stay on `__rackFetch` because the hijack contract can’t satisfy a synchronous XHR response anyway. Returns either a response hash (immediate — middleware didn’t hijack) or a ‘=> N‘ token (deferred — middleware hijacked the connection and a background thread is reading the pipe). The JS-side XHR checks the return shape to pick between inline processing and waiting for `_csim deliverHijackedFetches`.



2836
2837
2838
2839
2840
2841
2842
2843
2844
2845
2846
2847
2848
2849
2850
2851
2852
2853
2854
2855
2856
2857
2858
2859
2860
2861
2862
2863
2864
2865
2866
2867
2868
2869
2870
2871
2872
2873
2874
2875
2876
2877
 
# File 'lib/capybara/simulated/browser.rb', line 2836

def rack_fetch_async(method, url, body, headers_json)
  headers = begin
    JSON.parse(headers_json.to_s)
  rescue JSON::ParserError
    {}
  end
  # `rack_fetch` already handles redirects, cookie merge, the
  # asset cache shortcut, and download detection — keep async
  # XHRs on that single source of truth. The new behaviour is
  # the hijack hook for long-poll-shaped requests: install
  # `rack.hijack` so the middleware can hold the connection
  # open until something publishes through it.
  #
  # We can't unconditionally install the hijack env keys: some
  # downstream Discourse middleware paths take a different
  # streaming branch when `rack.hijack?` is truthy (even
  # without ever invoking the lambda) and the response then
  # re-renders the page in a slightly different order, racing
  # subsequent Capybara `find`s into StaleElement. Restrict
  # the hook to URLs that look like the long-poll endpoints
  # we actually need it for (`/message-bus/{id}/poll` today;
  # extend as new patterns surface).
  read_io = nil
  env_extras =
    if HIJACK_AWARE_URL_PATTERNS.any? {|re| re.match?(url.to_s) }
      {
        'rack.hijack?' => true,
        'rack.hijack'  => lambda {
          read_io, write_io = IO.pipe
          write_io
        }
      }
    end
  resp = rack_fetch(method, url, body, headers, 'follow', env_extras: env_extras)
  return resp || {'status' => 0, 'headers' => {}, 'body' => ''} unless read_io
  id = (@hijack_fetch_seq += 1)
  @hijack_fetch_threads[id] = Thread.new do
    Thread.current.report_on_exception = false
    run_hijacked_pipe_read(id, read_io, @hijack_fetch_queue)
  end
  {'handle' => id}
end

#rack_fetch_async_abort(id) ⇒ Object 



2888
2889
2890
2891
2892
 
# File 'lib/capybara/simulated/browser.rb', line 2888

def rack_fetch_async_abort(id)
  thread = @hijack_fetch_threads.delete(id.to_i)
  thread&.kill
  nil
end

#rack_fetch_body(url) ⇒ Object

── Host-fn callbacks invoked by bridge.js ──────────────────



2277
2278
2279
2280
2281
 
# File 'lib/capybara/simulated/browser.rb', line 2277

def rack_fetch_body(url)
  result = rack_fetch('GET', url, '', {}, 'follow')
  return nil unless result && result['status'].to_i < 400
  result['body'].to_s
end

#read_file_pick(handle, index, start = nil, finish = nil) ⇒ Object

JS-side ‘__HostBackedFile.text()` / `arrayBuffer()` route through this to read attached file bytes on demand — ActiveStorage’s ‘DirectUpload` MD5-chunks the file via FileReader before POSTing to `/rails/active_storage/direct_uploads`. Returns the requested byte range as base64 so binary content survives the engine string boundary (same approach as `__csimReadBlobBase64`).



1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
 
# File 'lib/capybara/simulated/browser.rb', line 1094

def read_file_pick(handle, index, start = nil, finish = nil)
  paths = file_picks_for(handle.to_i)
  path = paths && paths[index.to_i]
  return nil unless path && File.exist?(path)
  size = File.size(path)
  s = [start.to_i, 0].max
  e = finish.nil? ? size : [finish.to_i, size].min
  return Base64.strict_encode64('') if e <= s
  bytes = File.open(path, 'rb') do |f|
    f.seek(s)
    f.read(e - s)
  end
  Base64.strict_encode64(bytes || '')
end

#read_rack_body(body) ⇒ Object

Rack response bodies must respond to ‘each` (or be an Array of strings). `to_s` on a streaming body returns the inspect form, not the bytes — which silently shipped 43-byte `#<Rack::Files…>` strings to the JS engine for big assets like jquery.js.



3546
3547
3548
3549
3550
3551
 
# File 'lib/capybara/simulated/browser.rb', line 3546

def read_rack_body(body)
  buf = +''
  body.each {|chunk| buf << chunk.to_s } if body.respond_to?(:each)
  body.close if body.respond_to?(:close)
  buf
end

#record_action(kind, description) ⇒ Object

Wraps a driver action so the trace records description, urls, console / network activity, and (on action error / full mode) a post-action DOM snapshot. Re-entrant: nested recorded actions (label-click → click, session send_keys → send_keys) let the outer step own the boundary and the inner just yields.

‘description` is a String or Proc — Procs are lazy-evaluated only when a step is actually being recorded, so the off-path doesn’t pay ‘describe_node_handle`’s V8 round-trip.



1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
 
# File 'lib/capybara/simulated/browser.rb', line 1720

def record_action(kind, description)
  # Off-mode: no autostart, only proceed if a trace was started
  # explicitly via `driver.start_tracing`. Hot path for users
  # who set CSIM_TRACE=off.
  return yield if @trace.nil? && @trace_mode == :off
  if @trace.nil?
    @trace = Trace.new(metadata: {auto_started_at: Time.now.utc.iso8601(3)})
    @runtime.call('__csimSetTraceActive', true)
  end
  return yield if @recording_action
  @recording_action = true
  desc = description.is_a?(Proc) ? description.call : description
  @trace.begin_step(kind, description: desc, url_before: @current_url)
  error = nil
  begin
    yield
  rescue => e
    error = {class: e.class.name, message: e.message}
    raise
  ensure
    # `full` mode serializes the document after every action; the
    # default `on_failure` mode only snapshots when an action
    # errored. The V8 round-trip + DOM serialize is the
    # expensive part of trace recording, so skipping it on the
    # happy path is the whole point of the default.
    dom = (error || @trace_mode == :full) ? html : nil
    @trace.finish_step(url_after: @current_url, dom_after: dom, error: error)
    @recording_action = false
  end
end

#record_history(entry) ⇒ Object 



1634
1635
1636
1637
1638
1639
1640
 
# File 'lib/capybara/simulated/browser.rb', line 1634

def record_history(entry)
  # Discard any forward-history tail (a real browser drops the
  # redo stack the moment you navigate after a `go_back`).
  @history = @history[0..@history_idx] if @history_idx + 1 < @history.size
  @history << entry.merge(kind: entry[:kind] || :visit)
  @history_idx = @history.size - 1
end

#record_response(status, headers) ⇒ Object 



1509
1510
1511
1512
 
# File 'lib/capybara/simulated/browser.rb', line 1509

def record_response(status, headers)
  @last_response_status  = status
  @last_response_headers = headers.to_h
end

#record_url_transition(new_url) ⇒ Object

Called whenever ‘@current_url` is about to be set to a new value during a page-load drain or a settle tick driven by a user action; queues the prior URL for surface-via- `current_url` so a polling matcher walks the intermediate chain. Out-of-band JS-driven pushStates (`execute_script(“history.pushState(…)”)`) bypass the queue —they have no chain of microtask-driven transitions to walk, and the caller expects to read the new URL one-shot. Bounded to size 8 to guard against runaway chains; `current_url`’s staleness check drops the rest on any read past the polling- cadence window. Without the queue the finish_installation wizard chain’s intermediate ‘/wizard` would be invisible: the JS-side `replaceWith` to `/wizard/steps/setup` lands during a tick, so by the time Capybara polls `@current_url` is already the final URL.



481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
 
# File 'lib/capybara/simulated/browser.rb', line 481

def record_url_transition(new_url)
  return unless @ticking || @navigating
  old = @current_url
  return if old.nil? || old.to_s.empty?
  return if old.to_s == new_url.to_s
  # The URL the action started from is the starting point, not an
  # intermediate it walked through — don't surface it to a polling
  # (or one-shot) `current_url` as a step. Genuine mid-action
  # intermediates (a load to /wizard, *then* a replaceState to
  # /wizard/steps/setup) differ from the baseline and still queue.
  return if @action_url_baseline && old.to_s == @action_url_baseline.to_s
  @recent_urls << old.to_s
  @recent_urls.shift while @recent_urls.size > 8
  @recent_urls_last_push_at = Process.clock_gettime(Process::CLOCK_MONOTONIC, :millisecond)
end

#refresh ⇒ Object

is just a re-GET. Replay the current history entry.



3592
3593
3594
 
# File 'lib/capybara/simulated/browser.rb', line 3592

def refresh
  replay_history_entry(@history[@history_idx])
end

#replay_history_entry(entry) ⇒ Object 



1641
1642
1643
1644
1645
1646
1647
1648
 
# File 'lib/capybara/simulated/browser.rb', line 1641

def replay_history_entry(entry)
  return unless entry
  if entry[:method] == :post
    navigate_post(entry[:url], entry[:body], entry[:content_type], from_history: true)
  else
    navigate(entry[:url], from_history: true)
  end
end

#reset! ⇒ Object 



2191
2192
2193
2194
2195
2196
2197
2198
2199
2200
2201
2202
2203
2204
2205
2206
2207
2208
2209
2210
2211
2212
2213
2214
2215
2216
2217
2218
2219
2220
2221
2222
2223
2224
2225
2226
2227
2228
2229
2230
2231
2232
2233
2234
2235
2236
2237
2238
2239
2240
2241
2242
2243
2244
2245
2246
2247
2248
2249
2250
2251
2252
2253
2254
2255
2256
 
# File 'lib/capybara/simulated/browser.rb', line 2191

def reset!
  @cookies.clear
  @local_storage.clear
  @session_storage.clear
  @sticky_headers.clear
  # The driver-side resize buffer has to clear too — without
  # this the previous test's `driver.resize(425, …)` leaks into
  # the next test's default viewport and any cascade rule that
  # gates on `(min-width: …)` reports the wrong answer for the
  # whole new test (Forem's comment-actions dropdown is
  # mobile-collapsed-by-default). The exception is the
  # `default_viewport` channel — drivers built for a mobile
  # session (Discourse's `playwright_mobile_chrome`) need to
  # stay mobile across resets, not snap back to desktop on the
  # next mobile-tagged test.
  if @default_viewport
    @viewport_width  = @default_viewport[0]
    @viewport_height = @default_viewport[1]
  else
    @viewport_width  = nil
    @viewport_height = nil
  end
  @current_url     = nil
  @document_handle = 0
  # A test may leave a frame switched-to without switching back
  # (Capybara's reset_session spec covers exactly this); start the
  # next test back on the main document.
  reset_frame_scope
  @history.clear
  @history_idx     = -1
  @file_picks      = {} if @file_picks
  # Hand the live trace off to `@pending_trace` so an after-hook
  # running after `reset_session!` (Capybara's per-test teardown
  # order) still finds it. Anything stuck in `@pending_trace`
  # from a prior test is dropped — better than fusing two
  # tests' actions into one record.
  @pending_trace    = @trace
  @trace            = nil
  @recording_action = false
  # Kill any open SSE reader threads — the new VM has no JS-side
  # EventSource instances to dispatch into, and the old handles
  # would collide on the fresh handle counter the bridge starts
  # from after `reset_page`. Same shape for worker threads.
  reset_event_sources
  reset_hijacked_fetches
  reset_workers
  reset_websockets
  @window_inbox.clear
  # Free any zero-copy transfer backing stores that went unimported
  # (worker killed before draining its inbox, etc.) before the rebuild.
  drop_pending_transfers
  @blob_registry_lock.synchronize { @blob_registry.clear }
  # Drop volatile entries from the class-level HTTP asset cache
  # so test-local DB state (TranslationOverride, etc.) reaches
  # the app on subsequent visits. Fingerprinted assets
  # (`Cache-Control: immutable`) survive: their URLs are content-
  # addressable so a stale entry can't shadow a later test.
  @@asset_cache.clear_volatile if @@asset_cache.respond_to?(:clear_volatile)
  @runtime.reset_page
  # Per-visit ctx rebuild drops the JS-side trace-active flag,
  # so re-flip it if we're carrying a pending trace into the
  # next visit.
  @runtime.call('__csimSetTraceActive', false)
  reset_timer_state
  invalidate_find_cache
end

#reset_event_sources ⇒ Object 



2548
2549
2550
2551
2552
 
# File 'lib/capybara/simulated/browser.rb', line 2548

def reset_event_sources
  @event_source_threads.each_value(&:kill)
  @event_source_threads.clear
  @event_source_queue.clear
end

#reset_frame_scope ⇒ Object

Return DOM-op routing to the main document and drop any frame stack. Called by ‘switch_to_frame(:top)`, per-test `reset!`, and every full page (re)build (which disposes all frame realms) — anything that invalidates the active `within_frame` scope.



580
581
582
583
 
# File 'lib/capybara/simulated/browser.rb', line 580

def reset_frame_scope
  @current_realm_id = nil
  @frame_stack.clear
end

#reset_hijacked_fetches ⇒ Object 



2904
2905
2906
2907
2908
 
# File 'lib/capybara/simulated/browser.rb', line 2904

def reset_hijacked_fetches
  @hijack_fetch_threads.each_value(&:kill)
  @hijack_fetch_threads.clear
  @hijack_fetch_queue.clear
end

#reset_timer_state ⇒ Object

Re-sync the Ruby-side timer mirror with a freshly-rebuilt JS context. Clear ‘@timers_active` and the `@polling_grace` grace window so the previous page’s pending-timer state doesn’t leak into the next test, leaving ‘Driver#wait?` true and dragging every failing matcher through the full `default_max_wait_time` retry loop.



2067
2068
2069
2070
2071
2072
2073
2074
2075
2076
 
# File 'lib/capybara/simulated/browser.rb', line 2067

def reset_timer_state
  @last_tick_ts       = Process.clock_gettime(Process::CLOCK_MONOTONIC)
  @wall_clock_last    = @last_tick_ts   # CSIM_CLOCK_WALL escape hatch: don't replay the prev page's gap
  @timers_active      = false
  @polling_grace      = nil
  @last_polled_gen    = nil
  @idle_settle_polls  = 0
  @ff_transient_polls = 0
  @context_gen       += 1
end

#reset_websockets ⇒ Object 



2648
2649
2650
2651
2652
2653
2654
2655
2656
2657
2658
2659
 
# File 'lib/capybara/simulated/browser.rb', line 2648

def reset_websockets
  @websocket_threads.each_value(&:kill)
  @websocket_threads.clear
  # Close BOTH pair ends: csim's read/write end and the app's hijack end
  # (the app may abandon its end without closing it — e.g. its connection
  # thread was just killed), so neither leaks across tests.
  @websocket_sockets.each_value     {|s| s.close rescue nil }
  @websocket_app_sockets.each_value {|s| s.close rescue nil }
  @websocket_sockets.clear
  @websocket_app_sockets.clear
  @websocket_queue.clear
end

#reset_workers ⇒ Object 



3137
3138
3139
3140
3141
3142
3143
3144
3145
3146
3147
3148
3149
 
# File 'lib/capybara/simulated/browser.rb', line 3137

def reset_workers
  @workers.each_value do |w|
    w[:inbox] << :terminate
    w[:thread].kill
  end
  @workers.clear
  @worker_outbox.clear
  @worker_in_flight = 0
  @transfer_buffer_lock.synchronize {
    @transfer_buffers.clear
    @transfer_buffer_seq = 0
  }
end

#resolve_against(url, base) ⇒ Object 



3383
3384
3385
3386
3387
3388
3389
3390
3391
3392
3393
3394
3395
3396
3397
3398
3399
3400
3401
3402
3403
 
# File 'lib/capybara/simulated/browser.rb', line 3383

def resolve_against(url, base)
  return url if url =~ %r{\A[a-z]+://}i
  # quickjs.rb's module_loader passes the importer for nested
  # relative imports; if the importer was an inline-script
  # pseudo-name (no scheme), fall through to the page URL.
  base = nil unless base.is_a?(String) && base =~ %r{\A[a-z]+://}i
  eff = base || @current_url || @default_host
  # Memo of `URI.join(eff, url)` — a pure function of (effective base, url).
  # A heavy ESM app re-resolves the same ~80 module specifiers against the
  # same base on every visit (a fresh VM re-instantiates the whole module
  # graph); Ruby's URI parser was a measured ~11% of per-visit wall. The
  # Browser persists across a suite's visits, so this instance-level memo
  # (same scope/threading assumptions as @importmap / @current_url) turns
  # all but the first visit's resolves into hash hits.
  cache = (@resolve_against_cache ||= {})
  cache[[eff, url]] ||= begin
    URI.join(eff, url).to_s
  rescue URI::InvalidURIError, URI::BadURIError
    url
  end
end

#resolve_document_url(url) ⇒ Object

Public entry for the Driver to resolve a ‘window.open` / cross-window `location` URL against THIS window’s document (the internal resolver is private). Honours ‘<base href>` like the page’s own links do.



599
600
601
 
# File 'lib/capybara/simulated/browser.rb', line 599

def resolve_document_url(url)
  resolve_against_current(url, use_base: true)
end

#resolve_module_specifier(specifier, base_url) ⇒ Object 



3372
3373
3374
3375
3376
3377
3378
3379
3380
3381
 
# File 'lib/capybara/simulated/browser.rb', line 3372

def resolve_module_specifier(specifier, base_url)
  @importmap ||= {'imports' => {}, 'scopes' => {}}
  if (mapped = @importmap['imports'][specifier])
    return resolve_against(mapped, base_url)
  end
  if specifier.start_with?('/', './', '../') || specifier.match?(%r{\A[a-z]+://}i)
    return resolve_against(specifier, base_url)
  end
  specifier
end

#resolve_visit_url(url) ⇒ Object 



397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
 
# File 'lib/capybara/simulated/browser.rb', line 397

def resolve_visit_url(url)
  s = url.to_s
  # `about:blank` (and other authority-less schemes) have no `//`, so the
  # `scheme://` test below would treat them as relative paths and prepend
  # the host root. `navigate` handles `about:blank` specially — pass it
  # through untouched (open_new_window opens an about:blank tab).
  return s if s.match?(/\Aabout:/i)
  unless s =~ %r{\A[a-z]+://}i
    # Strip path/query/fragment off the current URL to get the origin
    # root. An opaque or host-less current URL (e.g. `about:blank` in a
    # freshly-opened window) can't yield an origin — fall back to the
    # default host so a subsequent relative `visit` still resolves.
    host_root =
      begin
        u = URI.parse(@current_url.to_s)
        if u.opaque || u.host.nil?
          @default_host
        else
          u.path = ''; u.query = nil; u.fragment = nil
          u.to_s
        end
      rescue URI::InvalidURIError
        @default_host
      end
    host_root = host_root.sub(/\/+$/, '')
    s = "/#{s}" unless s.start_with?('/')
    s = "#{host_root}#{s}"
  end
  # Real browsers percent-encode characters that aren't legal in their
  # URL position before issuing the request. Skip the escape pass when
  # the input is already clean (the common case).
  s.match?(URL_UNSAFE_CHARS) ? URI::DEFAULT_PARSER.escape(s, URL_UNSAFE_CHARS) : s
end

#response_hash(status, headers, body, url, redirected) ⇒ Object 



3493
3494
3495
3496
3497
3498
3499
3500
3501
3502
3503
3504
3505
3506
3507
3508
3509
3510
3511
3512
3513
 
# File 'lib/capybara/simulated/browser.rb', line 3493

def response_hash(status, headers, body, url, redirected)
  raw     = body.to_s
  hdrs    = stringify(headers)
  is_text = text_response?(hdrs)
  # `body` crosses as TEXT — `responseText` semantics: the bytes decoded
  # as UTF-8 with invalid sequences replaced (a leading BOM selects the
  # encoding per the HTML "decode" algorithm and is removed). The real
  # bytes for binary consumers ride `body_b64`; the Rack body arrives
  # BINARY-tagged (see `RuntimeShared.utf8_text`).
  text = RuntimeShared.utf8_text(is_text ? decode_response_bom(raw) : raw)
  out = {
    'status'     => status,
    'headers'    => hdrs,
    'body'       => text,
    'url'        => url,
    'redirected' => redirected,
    'type'       => 'basic'
  }
  out['body_b64'] = Base64.strict_encode64(raw) unless is_text
  out
end

#response_headers ⇒ Object

Rack 3 lowercases header names; Capybara tests do ‘[’Content-Type’]‘.



1504
1505
1506
1507
1508
 
# File 'lib/capybara/simulated/browser.rb', line 1504

def response_headers
  (@last_response_headers || {}).each_with_object({}) {|(k, v), h|
    h[k.to_s.split('-').map(&:capitalize).join('-')] = v
  }
end

#right_click(handle, keys = [], **opts) ⇒ Object 



1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
 
# File 'lib/capybara/simulated/browser.rb', line 1109

def right_click(handle, keys = [], **opts)
  mark_action_baseline
  tick_real_time
  invalidate_find_cache
  ensure_alive_after_tick(handle)
  init = {'bubbles' => true, 'cancelable' => true, 'button' => 2, 'which' => 3}.merge(click_event_init(handle, keys, opts))
  dom_call('__csimDispatchEvent', handle, 'mousedown', init)
  sleep opts[:delay].to_f if opts[:delay].to_f > 0
  dom_call('__csimDispatchEvent', handle, 'mouseup',     init)
  dom_call('__csimDispatchEvent', handle, 'contextmenu', init)
end

#same_document_traversal?(from, to) ⇒ Boolean

Same-document = every entry between ‘from` and `to` (inclusive) is a `:push_state` entry (or the boundary just changed state on the current URL). A `:visit` entry between them means we’d cross a real navigation, which needs a fresh document.

Returns:

  • (Boolean) 


1629
1630
1631
1632
 
# File 'lib/capybara/simulated/browser.rb', line 1629

def same_document_traversal?(from, to)
  lo, hi = [from, to].sort
  ((lo + 1)..hi).all? {|i| @history[i] && @history[i][:kind] == :push_state }
end

#select_option(handle) ⇒ Object 



1311
1312
1313
1314
1315
1316
1317
1318
 
# File 'lib/capybara/simulated/browser.rb', line 1311

def select_option(handle)
  mark_action_baseline
  tick_real_time
  invalidate_find_cache
  dom_call('__csimSelectOption', handle)
  tick_real_time
  drain_after_user_action
end

#send_keys(handle, keys) ⇒ Object

Capybara’s ‘send_keys` accepts Strings and Symbols (special keys: `:enter`, `:tab`, `:backspace`, …) and Array combos (modifier + key). We hand each item to JS as a tagged atom so the bridge can fire proper KeyboardEvents with `key` / `code` / `ctrlKey` / `metaKey` / `shiftKey` filled in — required by libraries that gate behaviour on the modifier flags (Redmine’s jstoolbar reads ‘event.ctrlKey || event.metaKey` for Ctrl+B / Cmd+B; quote-reply Stimulus controllers read `event.key`). An Array combo is the canonical “modifier + key” pattern: everything but the last entry is a modifier; the last entry is the key being pressed (String char or Symbol special).



1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
 
# File 'lib/capybara/simulated/browser.rb', line 1255

def send_keys(handle, keys)
  mark_action_baseline
  tick_real_time
  invalidate_find_cache
  ensure_alive_after_tick(handle)
  # Selenium's contract: a bare modifier symbol (`:shift`) at the
  # top level "holds" the modifier from that point on. `:null`
  # releases all modifiers. We rewrite the atom stream so each
  # following character / key carries the accumulated modifiers.
  held = []
  atoms = keys.flat_map {|k|
    case k
    when Symbol
      if k == :null
        held = []; nil
      elsif MODIFIER_KEY_NAMES.include?(k)
        held = (held + [k.to_s]).uniq; nil
      else
        held.empty? ? {'kind' => 'key',  'name'  => k.to_s}
                    : {'kind' => 'combo', 'parts' => held + [k.to_s]}
      end
    when String
      held.empty? ? {'kind' => 'text', 'value' => k}
                  : {'kind' => 'combo', 'parts' => held + [k]}
    when Array
      parts = k.map {|x| x.is_a?(Symbol) ? x.to_s : x.to_s }
      {'kind' => 'combo', 'parts' => parts}
    end
  }.compact
  # Contenteditable hosts (ProseMirror, Trix, Tiptap) reconcile
  # their view between chars; a batched `__csimSendKeys` queues
  # all `beforeinput` events on the same microtask round and PM
  # nukes the editor wrapper when its reconciler can't apply
  # them in order. Split multi-char text atoms into per-char
  # calls with a `settle` between so PM commits each transaction
  # before the next char arrives. Plain `<input>` / `<textarea>`
  # don't need this — keep the single batched call there.
  has_multichar_text = atoms.any? {|a| a['kind'] == 'text' && a['value'].to_s.length > 1 }
  if has_multichar_text && dom_call('__csimIsContentEditable', handle)
    per_char = atoms.flat_map {|a|
      next a unless a['kind'] == 'text' && a['value'].to_s.length > 1
      a['value'].to_s.each_char.map {|c| {'kind' => 'text', 'value' => c} }
    }
    head, *tail = per_char
    dom_call('__csimSendKeys', handle, [head])
    tail.each {|atom|
      tick_real_time
      dom_call('__csimSendKeys', handle, [atom])
      settle
    }
  else
    dom_call('__csimSendKeys', handle, atoms)
  end
  drain_after_user_action
end

#send_session_key(key) ⇒ Object 



1681
 
# File 'lib/capybara/simulated/browser.rb', line 1681

def send_session_key(key) = send_session_keys([key])

#send_session_keys(keys) ⇒ Object

Session-level keystroke. Tab / shift-tab cycle focus; everything else is routed to the currently focused element (if any) as a plain keydown/keyup pair.



1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
 
# File 'lib/capybara/simulated/browser.rb', line 1657

def send_session_keys(keys)
  # Walk the key list with running modifier state so a Selenium-
  # style `(:shift, :enter)` invocation reaches `Browser#send_keys`
  # as one combo atom (shift held over enter), while independent
  # non-modifier keys stay separate calls — each one settles
  # between dispatches so a dropdown highlight (Avo Tags input's
  # arrow navigation) commits before the next key fires. Tab /
  # backtab are focus-advance, dispatched out of band.
  held = []
  Array(keys).each do |k|
    sym = k.is_a?(Symbol) ? k : (k.respond_to?(:to_sym) ? k.to_sym : nil)
    if sym == :tab || sym == :backtab
      dom_call('__csimAdvanceFocus', sym == :backtab)
    elsif sym && MODIFIER_KEY_NAMES.include?(sym)
      held << sym
    else
      handle = active_element_handle
      handle = current_document_handle if handle.nil? || handle.zero?
      atom = held.empty? ? k : (held + [k])
      send_keys(handle, [atom])
    end
  end
end

#set_geolocation(latitude: nil, longitude: nil, accuracy: 10, denied: false, **rest) ⇒ Object

CDP-ish shim: override navigator.geolocation (like CDP’s ‘Emulation.setGeolocationOverride`). State is Ruby-backed on `@geolocation`; the JS geolocation object reads it on every call via the `__csimGeolocationState` host fn, so it survives the per-call VM rebuilds (the same model web storage uses).

set_geolocation(latitude: 35.6, longitude: 139.7)
set_geolocation(latitude: 1, longitude: 2, accuracy: 5, altitude: 10)
set_geolocation(denied: true)  # report PERMISSION_DENIED
set_geolocation                # clear -> report POSITION_UNAVAILABLE



1828
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
 
# File 'lib/capybara/simulated/browser.rb', line 1828

def set_geolocation(latitude: nil, longitude: nil, accuracy: 10, denied: false, **rest)
  @geolocation =
    if denied
      {'denied' => true}
    elsif latitude.nil? || longitude.nil?
      nil
    else
      {'coords' => {'latitude' => latitude, 'longitude' => longitude, 'accuracy' => accuracy}.merge(rest.transform_keys(&:to_s))}
    end

  # Re-deliver to any active watchPosition watchers, mirroring a real
  # browser firing the watch again when the location updates. The JS
  # side reads the fresh @geolocation via the host fn.
  execute_script('if (typeof globalThis.__csimGeoRefireWatches === "function") globalThis.__csimGeoRefireWatches();')
  nil
end

#set_header(name, value) ⇒ Object 



1514
 
# File 'lib/capybara/simulated/browser.rb', line 1514

def set_header(name, value)         ; @sticky_headers[name.to_s] = value.to_s ; end

#set_importmap(json) ⇒ Object

JS-side ‘ingestImportmaps` calls this through the host fn so Ruby-side `resolve_module_specifier` agrees with the bare- specifier map shipped by `<script type=“importmap”>`.



3366
3367
3368
3369
3370
 
# File 'lib/capybara/simulated/browser.rb', line 3366

def set_importmap(json)
  @importmap = JSON.parse(json.to_s)
rescue JSON::ParserError
  @importmap = {'imports' => {}, 'scopes' => {}}
end

#set_value_with_events(handle, value) ⇒ Object 



999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
 
# File 'lib/capybara/simulated/browser.rb', line 999

def set_value_with_events(handle, value)
  mark_action_baseline
  tick_real_time
  invalidate_find_cache
  ensure_alive_after_tick(handle)
  # `attach_file` hands us a Pathname (or Array of Pathnames);
  # the marshaller rejects non-primitive types. Coerce to a path-list
  # form V8 can hold — the actual multipart upload happens later
  # in `build_multipart_body` during form submission.
  coerced = coerce_set_value(value)
  # For date/time-shaped inputs we need the type-specific
  # string. Probe the handle's `type` and re-format Date / Time
  # accordingly — `Date.today` → `2026-05-13` (date input) is
  # already right via to_s, but `Time` needs the input-type-
  # specific format.
  coerced = format_temporal_value(value, handle) if value.is_a?(Date) || value.is_a?(Time)
  @file_picks ||= {}
  # Capybara `attach_file` calls `Node#set` with a Pathname; some
  # callers pass a String path through directly. When the target
  # IS a file input, promote either form into the file-list path
  # so `.files` / `@file_picks` reflect the chosen file.
  coerced = [coerced.to_s] if value.is_a?(Pathname)
  if !coerced.is_a?(Array) && coerced.is_a?(String) && file_input?(handle)
    coerced = [coerced]
  end
  if coerced.is_a?(Array)
    paths = coerced.reject(&:empty?)
    @file_picks[handle] = paths
    # Expose File-list metadata to the JS side BEFORE setting the
    # value: __csimSetValue fires input + change synchronously,
    # and Redmine's onchange="addInputFiles(this)" reads
    # `inputEl.files` — if we set files after, the handler sees
    # an empty FileList and tears down the input.
    file_infos = paths.map {|p|
      stat = (File.stat(p) rescue nil)
      {
        'name'         => File.basename(p),
        'size'         => stat ? stat.size : 0,
        # Real browsers tag the File with the MIME type they
        # sniffed from the path / disk header. Uppy's image-type
        # filter rejects files whose `type` is empty, so without
        # this even a `logo.png` `attach_file` finishes uploading
        # 0 bytes through the validator and the composer's
        # `#file-uploading` flag stays set forever. Use the OS's
        # extension-based guess (matches what selenium / Chromium
        # do on these paths) and fall back to empty when the
        # extension is unknown.
        'type'         => mime_type_for_path(p),
        'lastModified' => stat ? (stat.mtime.to_f * 1000).to_i : 0
      }
    }
    dom_call('__csimSetFiles', handle, file_infos)
    # Mirror real browser: <input type=file>.value reflects only
    # the filename of the first chosen file (security-faked path).
    # __csimSetValue dispatches input + change synchronously.
    js_value = paths.first ? File.basename(paths.first) : ''
    dom_call('__csimSetValue', handle, js_value)
  else
    dom_call('__csimSetValue', handle, coerced)
  end
  drain_after_user_action
end

#set_viewport(w, h) ⇒ Object 



1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
 
# File 'lib/capybara/simulated/browser.rb', line 1546

def set_viewport(w, h)
  @viewport_width  = w.to_i
  @viewport_height = h.to_i
  invalidate_find_cache
  @runtime.eval("globalThis.innerWidth = #{@viewport_width}; globalThis.innerHeight = #{@viewport_height};")
  # Recompute the cascade `@media` rules against the new
  # viewport so visibility checks (Capybara `visible?`,
  # `getComputedStyle().display`) re-reflect mobile-breakpoint
  # `display: none` / `display: block` flips. Without this the
  # cascade keeps the pre-resize hide-rule set.
  @runtime.call('__csimRebuildCascade') if @document_handle.to_i > 0
  # Fire `change` events on every live MediaQueryList whose
  # match state flipped, so libraries that hold `matchMedia(...)`
  # listeners (Discourse's `TrackedMediaQuery` powering the
  # viewport-based mobile/desktop class swap) reactively
  # re-render. The JS-side function iterates `_activeQueries`
  # and dispatches only on transitions — cheap no-op when no
  # query is open.
  @runtime.call('__csimViewportChanged') if @document_handle.to_i > 0
  # Re-fire a `resize` event so libraries that re-layout on
  # resize (responsive nav, sidebar collapse) see the new size.
  @runtime.eval("try { (globalThis.dispatchEvent || function(){})(new Event('resize')); } catch (_) {}")
  nil
end

#set_window_location(handle, url) ⇒ Object 



3069
 
# File 'lib/capybara/simulated/browser.rb', line 3069

def set_window_location(handle, url) = (@driver.window_set_location(handle.to_s, url.to_s) if @driver.respond_to?(:window_set_location))

#settle ⇒ Object

Yield on the first observable change. Each iter (a) drains the chained-await/‘.then` microtask queue a few rounds, (b) checks the JS-side `__settleGen` counter — bumped on every DOM mutation / URL change — and bails if it ticked, otherwise © advances the virtual clock to fire rAF / setTimeout that the chain is parked on. Capybara’s outer polling loop drives the next iter on the next find / has_? — matching real browsers’ “one paint = one observable moment” semantics.

This makes a user-action settle as cheap as ~4 evals when the click immediately mutates DOM, and lets ‘wait_for_*` helpers catch transient states like “modal removed before the redirect_to Visit’s render rebuilds it” — exactly the window real browsers paint at.



1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
 
# File 'lib/capybara/simulated/browser.rb', line 1397

def settle
  start_gen = @runtime.settle_gen
  prev_gen  = start_gen
  SETTLE_MAX_ITER.times do
    deliver_event_source_events
    deliver_worker_messages
    deliver_hijacked_fetches
    deliver_window_messages
    deliver_websocket_events
    break if @runtime.settle_gen > start_gen
    break unless @timers_active || event_source_pending? || worker_pending? || hijack_fetch_pending? || window_message_pending? || websocket_pending?
    # ONE event-loop step replaces the old drain_microtasks(4)+drain_timers(32)
    # pair: it fires due timers, runs a per-task microtask checkpoint (so
    # chained .then / MutationObserver delivery interleave spec-correctly),
    # and runs the render phase — bailing INTERNALLY on the first settleGen
    # bump (yield_on_gen), which preserves the one-observable-boundary-per-poll
    # contract. maxMs 0 when no timer is active just flushes microtasks +
    # render for the work the deliveries above queued.
    @runtime.run_loop_step(@timers_active ? SETTLE_DRAIN_MS : 0, SETTLE_MAX_ITER_TASKS, yield_on_gen: true)
    deliver_event_source_events
    deliver_worker_messages
    deliver_hijacked_fetches
    deliver_window_messages
    deliver_websocket_events
    break if @runtime.settle_gen > start_gen
    # No progress this iter (no DOM/URL change observed) — the
    # remaining timers are queued for the future; bail and let
    # Capybara's wall-clock-driven poll loop drive the next tick
    # via `tick_real_time`. SSE / Worker channels keep us in
    # the loop as long as background threads have data queued.
    break if @runtime.settle_gen == prev_gen && !@runtime.has_ready_timer? && !event_source_pending? && !worker_pending? && !hijack_fetch_pending? && !window_message_pending? && !websocket_pending?
    prev_gen = @runtime.settle_gen
  end
  @find_cache_dirty = true
end

#shadow_root_handle(handle) ⇒ Object 



781
782
783
784
 
# File 'lib/capybara/simulated/browser.rb', line 781

def shadow_root_handle(handle)
  h = dom_call('__csimShadowRoot', handle).to_i
  h.zero? ? nil : h
end

#stack_resolver ⇒ Object 



1772
1773
1774
 
# File 'lib/capybara/simulated/browser.rb', line 1772

def stack_resolver
  @stack_resolver ||= StackResolver.new(self)
end

#start_trace(metadata = {}) ⇒ Object 



1692
1693
1694
1695
 
# File 'lib/capybara/simulated/browser.rb', line 1692

def start_trace(metadata = {})
  @trace = Trace.new(metadata: metadata)
  @runtime.call('__csimSetTraceActive', true)
end

#status_code ⇒ Object 



1502
1503
 
# File 'lib/capybara/simulated/browser.rb', line 1502

def status_code      = (@last_response_status || 200)
# Rack 3 lowercases header names; Capybara tests do `['Content-Type']`.

#storage_clear(kind) ⇒ Object 



3675
3676
3677
3678
 
# File 'lib/capybara/simulated/browser.rb', line 3675

def storage_clear(kind)
  store(kind).clear
  nil
end

#storage_get(kind, key) ⇒ Object

Web Storage host-fn shims. The Ruby-side hashes survive ‘rebuild_ctx` between visits, so apps that cache user data in `localStorage` on page A (Forem’s ‘browserStoreCache(’set’)‘ inside fetchBaseData) see it on page B — without this, every visit boots into a JS-side Map that starts empty and the first-call branches that hinge on cached user data (the onboarding task-card render, `initializeLocalStorageRender`, etc.) silently skip.



3664
3665
3666
 
# File 'lib/capybara/simulated/browser.rb', line 3664

def storage_get(kind, key)
  store(kind)[key.to_s]
end

#storage_key(kind, index) ⇒ Object 



3679
3680
3681
 
# File 'lib/capybara/simulated/browser.rb', line 3679

def storage_key(kind, index)
  store(kind).keys[index.to_i]
end

#storage_length(kind) ⇒ Object 



3682
3683
3684
 
# File 'lib/capybara/simulated/browser.rb', line 3682

def storage_length(kind)
  store(kind).size
end

#storage_remove(kind, key) ⇒ Object 



3671
3672
3673
3674
 
# File 'lib/capybara/simulated/browser.rb', line 3671

def storage_remove(kind, key)
  store(kind).delete(key.to_s)
  nil
end

#storage_set(kind, key, value) ⇒ Object 



3667
3668
3669
3670
 
# File 'lib/capybara/simulated/browser.rb', line 3667

def storage_set(kind, key, value)
  store(kind)[key.to_s] = value.to_s
  nil
end

#submit_form(handle) ⇒ Object

‘Node#submit(*)` (Capybara DSL) hits here. Find the enclosing form, serialise, post.



1482
1483
1484
1485
1486
1487
1488
 
# File 'lib/capybara/simulated/browser.rb', line 1482

def submit_form(handle)
  tick_real_time
  invalidate_find_cache
  form_handle = dom_call('__csimAncestorForm', handle).to_i
  return if form_handle.zero?
  submit_form_handle(form_handle, nil)
end

#submit_form_handle(form_handle, submitter_handle) ⇒ Object

Pulls the serialised form-state out of JS, encodes it, and drives the Rack app via ‘navigate` (for GET) or a POST.



2082
2083
2084
2085
2086
2087
2088
2089
2090
2091
2092
2093
2094
2095
2096
2097
2098
2099
2100
2101
2102
2103
2104
2105
2106
2107
2108
2109
2110
2111
2112
2113
2114
2115
2116
2117
2118
2119
2120
 
# File 'lib/capybara/simulated/browser.rb', line 2082

def submit_form_handle(form_handle, submitter_handle)
  invalidate_find_cache
  spec = dom_call('__csimFormSerialize', form_handle, submitter_handle || 0)
  return unless spec.is_a?(Hash)
  action  = spec['action'].to_s
  method  = spec['method'].to_s.upcase
  method  = 'GET' if method.empty?
  fields  = (spec['fields'] || []).map {|pair| [pair[0].to_s, pair[1].to_s] }
  file_inputs = spec['fileInputs'] || []
  enctype = spec['enctype'].to_s
  multipart = enctype.start_with?('multipart/form-data')
  content_type = nil
  body =
    if multipart
      built = build_multipart_body(fields, file_inputs)
      content_type = built[:content_type]
      built[:body]
    else
      # Non-multipart: file inputs contribute the filename only.
      file_inputs.each do |fi|
        picks = @file_picks && @file_picks[fi['handle'].to_i] || []
        fields << [fi['name'].to_s, picks.first ? File.basename(picks.first) : '']
      end
      URI.encode_www_form(fields)
    end
  action_url = action.empty? ? (current_browsing_context_url || @default_host) : resolve_against_current(action)
  # A form submitted inside a frame whose target is the frame itself
  # navigates the FRAME, not the top page.
  in_frame = !!@current_realm_id && frame_self_target?(spec['target'])
  if method == 'GET'
    uri = URI.parse(action_url)
    uri.query = body unless body.empty?
    in_frame ? navigate_frame(uri.to_s) : navigate(uri.to_s)
  elsif in_frame
    navigate_frame_post(action_url, body, content_type || enctype)
  else
    navigate_post(action_url, body, content_type || enctype)
  end
end

#switch_to_frame(target) ⇒ Object

Capybara ‘switch_to_frame`. `target` is an `<iframe>` handle in the CURRENT realm, or `:parent` / `:top`. Entering builds (or reuses) the frame’s V8 realm and routes subsequent DOM ops there; ‘:parent` pops one level, `:top` returns to the main document. Frame switches invalidate the find cache (its keys aren’t realm-qualified, and a switch is rare).

Scope: finds, reads, interactions (click/fill_in/…), evaluate_script, and a self-targeted navigation (a link / form submit whose default action loads a new document) all route into the frame — the frame’s realm is rebuilt from the fetched document, leaving the top page untouched (see ‘navigate_frame`). Out of scope: `_top` navigates the main page (correct), but a `_parent` target from a frame nested ≥2 levels navigates the main page rather than the intermediate frame, and cross-origin frame locality is resolved against the main page’s origin.



541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
 
# File 'lib/capybara/simulated/browser.rb', line 541

def switch_to_frame(target)
  invalidate_find_cache
  case target
  when :parent
    @frame_stack.pop
    @current_realm_id = @frame_stack.last && @frame_stack.last[:realm_id]
  when :top
    reset_frame_scope
  else
    # Per-frame realms are a V8-engine feature; QuickJS has no nested
    # browsing context to route into. Distinguish that (unsupported
    # engine) from a frame that simply failed to build (below), so the
    # error doesn't misattribute a load failure to the engine.
    unless @runtime.respond_to?(:realm_call)
      raise Capybara::Simulated::FrameNotSupported,
        'within_frame needs a per-frame browsing context, which only the ' \
        'V8 (rusty_racer) engine provides; QuickJS keeps a same-realm fallback.'
    end
    parent_realm = @current_realm_id
    tick_real_time
    rid = dom_call('__csimEnsureFrameRealm', target.to_i).to_i
    if rid.zero?
      raise Capybara::Simulated::StaleElement,
        "could not enter frame ##{target} (not a frame element, or its document failed to load)"
    end
    # Record the iframe handle + the realm it lives in so a frame-scoped
    # navigation can rebuild this exact frame (`reload_current_frame_realm`).
    @frame_stack.push({realm_id: rid, iframe_handle: target.to_i, parent_realm_id: parent_realm})
    @current_realm_id = rid
    # Let the freshly built realm's inline scripts / load handlers settle
    # so a find immediately inside the block sees the loaded document.
    settle
  end
end

#syntax_or_invalid_selector_error?(e) ⇒ Boolean

JS-side selector parser throws a ‘DOMException(’csim: …‘, ’SyntaxError’)‘. The JS engine surfaces it as a `…::SyntaxError` (QuickJS via dynamic-named class) or, under V8, a `RustyRacer::RuntimeError` whose message is `“SyntaxError: csim: …”`. Match the `csim: ` marker anywhere in the message (it’s no longer at the start once the DOMException name is prefixed) or the class suffix, so neither gem becomes a hard dependency.

Returns:

  • (Boolean) 


645
646
647
648
 
# File 'lib/capybara/simulated/browser.rb', line 645

def syntax_or_invalid_selector_error?(e)
  e.class.name.to_s.end_with?('::SyntaxError') ||
    e.message.to_s.include?('csim: ')
end

#tag(handle) ⇒ Object 



755
 
# File 'lib/capybara/simulated/browser.rb', line 755

def tag(handle)         = dom_call('__csimTag', handle).to_s

#tag_name(handle) ⇒ Object 



768
 
# File 'lib/capybara/simulated/browser.rb', line 768

def tag_name(handle)     = tag(handle)

#text(handle) ⇒ Object 



754
 
# File 'lib/capybara/simulated/browser.rb', line 754

def text(handle)        = dom_call('__csimText', handle).to_s

#text_response?(headers) ⇒ Boolean

Returns:

  • (Boolean) 


3536
3537
3538
3539
3540
 
# File 'lib/capybara/simulated/browser.rb', line 3536

def text_response?(headers)
  ct = (headers['content-type'] || headers['Content-Type']).to_s.downcase
  return false if ct.empty?
  TEXT_CONTENT_TYPE_PREFIXES.any? {|p| ct.start_with?(p) }
end

#tick_real_time(step_ms: nil) ⇒ Object

Advance the virtual JS clock and fire timers that came due. When ‘step_ms` is omitted, advance by `horizon_fast_forward_step` — a DETERMINISTIC step (never wall-derived, so per-poll JS/Ruby/GC cost can’t shift when a timer fires): a fixed ‘POLL_TICK_STEP_MS` per poll, fast- forwarding straight to a near-future timer when the page is otherwise idle. Explicit `step_ms` is used by `SleepHook#advance_virtual_clock_ms` (from `Kernel#sleep`) and by `Playwright::Page#wait_for_timeout` to step a precise virtual duration.



1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
 
# File 'lib/capybara/simulated/browser.rb', line 1949

def tick_real_time(step_ms: nil)
  return unless @timers_active || worker_pending? || event_source_pending? || hijack_fetch_pending? || window_message_pending? || websocket_pending?
  # Re-entrancy guard. Capybara's `Result#each` triggers nested
  # finds (visible? per element); the outermost tick has already
  # advanced the clock, the inner calls would only re-drain
  # already-fired timers.
  return if @ticking
  @ticking = true
  begin
    now = Process.clock_gettime(Process::CLOCK_MONOTONIC)
    # Kept wall-anchored ONLY for `timer_wait_elapsed?` / FIND_PRE_TICK_MIN_S
    # (gates tick FREQUENCY for the smoke first-find-no-fire contract); the
    # step SIZE below is deterministic.
    @last_tick_ts = now
    effective_step = step_ms || horizon_fast_forward_step
    if @timers_active && effective_step > 0
      r = @runtime.run_loop_step(effective_step)
      # `dirtied` (settleGen changed) catches a render-phase rAF / microtask-
      # delivered MutationObserver that mutated the DOM without firing a timer
      # (fired == 0) — a fired-count-only test would leave a stale find cache.
      @find_cache_dirty = true if r['dirtied'] || r['fired'].to_i > 0
    end
    # Pull any pending Worker / EventSource messages into JS
    # state. Without this, `evaluate_script` after kicking off
    # a worker round-trip would see stale state — the inbox
    # outbox only drains during `settle`, which doesn't run
    # for direct `execute_script` / `evaluate_script` calls.
    @find_cache_dirty = true if deliver_worker_messages > 0
    @find_cache_dirty = true if deliver_event_source_events > 0
    @find_cache_dirty = true if deliver_hijacked_fetches > 0
    @find_cache_dirty = true if deliver_window_messages > 0
    @find_cache_dirty = true if deliver_websocket_events > 0
  ensure
    @ticking = false
  end
  # Drain navigation intents queued by JS-side handlers that fired
  # during the drain (e.g. `setTimeout(() => location.pathname = X)`).
  # Outside the @ticking guard so the navigate's rebuild_ctx is
  # well-clear of the V8 call we just made.
  drain_pending_navigation
  # Same shape for `form.submit()` queued by a timer callback —
  # Forem's comment-edit form has an `onsubmit` handler that
  # `preventDefault`s, polls for the CSRF meta tag inside
  # `setInterval(…, 1)`, then calls `form.submit()` once the
  # meta is present. The click that originally fired the submit
  # event has already returned by the time the interval triggers,
  # so without this drain the intent sits on the slot forever
  # and the form never posts.
  consume_pending_form_submit
  # And for `<a download>` clicks (Avo's action-download chain
  # goes via file-saver's `saveAs` → synthetic dispatchEvent
  # on a freshly-created anchor with `download` + blob URL).
  consume_pending_download
end

#timer_wait_elapsed? ⇒ Boolean

Returns:

  • (Boolean) 


705
706
707
708
 
# File 'lib/capybara/simulated/browser.rb', line 705

def timer_wait_elapsed?
  @timers_active &&
    (Process.clock_gettime(Process::CLOCK_MONOTONIC) - @last_tick_ts) >= FIND_PRE_TICK_MIN_S
end

#title ⇒ Object 



1490
1491
1492
1493
 
# File 'lib/capybara/simulated/browser.rb', line 1490

def title
  tick_real_time
  @runtime.call('__csimDocumentTitle').to_s
end

#transfer_buffer_fetch(id) ⇒ Object 



3182
3183
3184
 
# File 'lib/capybara/simulated/browser.rb', line 3182

def transfer_buffer_fetch(id)
  @transfer_buffer_lock.synchronize { @transfer_buffers.delete(id.to_i) }
end

#transfer_buffer_fetch_for_js(id) ⇒ Object

Wraps the raw bytes in whatever binary shape the ACTIVE runtime can marshal to a JS Uint8Array (V8: the BINARY-tagged string itself —tag-driven marshalling crosses it as a Uint8Array; QuickJS: base64 that the JS shim’s ‘fetchedToBytes` atob’s — it has no binary marshaller). Asked of the runtime so each engine picks its shape.



3210
3211
3212
3213
3214
 
# File 'lib/capybara/simulated/browser.rb', line 3210

def transfer_buffer_fetch_for_js(id)
  bytes = transfer_buffer_fetch(id)
  return nil unless bytes
  @runtime.wrap_binary(bytes)
end

#transfer_buffer_stash(bytes) ⇒ Object

── postMessage transferable-buffer registry ───────────────────

Large Uint8Array / ArrayBuffer payloads cross isolates by ID; rusty_racer marshals typed arrays as ASCII-8BIT Strings so no JS-side latin-1 / base64 intermediate is built. Without this the 317 MB raw frames in Discourse’s media-optimization-worker peak >4 GB of JS strings before the worker even sees them.



3172
3173
3174
3175
3176
3177
3178
3179
3180
 
# File 'lib/capybara/simulated/browser.rb', line 3172

def transfer_buffer_stash(bytes)
  s = bytes.to_s
  s = s.dup.force_encoding(Encoding::ASCII_8BIT) unless s.encoding == Encoding::ASCII_8BIT
  @transfer_buffer_lock.synchronize {
    id = (@transfer_buffer_seq += 1)
    @transfer_buffers[id] = s
    id
  }
end

#transfer_token_issued(token) ⇒ Object

JS reports each zero-copy transfer token it mints (‘RustyRacer.transferOut`) so we can release any that go unimported. Callable from a worker thread.



3188
3189
3190
3191
3192
 
# File 'lib/capybara/simulated/browser.rb', line 3188

def transfer_token_issued(token)
  t = token.to_i
  @transfer_tokens_lock.synchronize { @transfer_tokens << t } if t > 0
  nil
end

#unselect_option(handle) ⇒ Object 



1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
 
# File 'lib/capybara/simulated/browser.rb', line 1320

def unselect_option(handle)
  mark_action_baseline
  tick_real_time
  invalidate_find_cache
  # Single-select <select>s can't have a selection cleared per
  # HTML — Capybara surfaces this as `UnselectNotAllowed`. Ask
  # the JS side whether the option's parent select is `multiple`
  # before issuing the unselect; the answer doubles as the
  # "found the right ancestor" check.
  info = dom_call('__csimOptionContext', handle)
  if info.is_a?(Hash) && info['hasSelect'] && !info['multiple']
    raise Capybara::UnselectNotAllowed, 'Cannot unselect option from single select box.'
  end
  dom_call('__csimUnselectOption', handle)
  tick_real_time
  drain_after_user_action
end

#update_current_hash(url) ⇒ Object 



984
985
986
987
988
989
990
991
992
993
994
995
996
997
 
# File 'lib/capybara/simulated/browser.rb', line 984

def update_current_hash(url)
  return if @current_url.nil?
  new_url = resolve_against_current(url)
  @current_url = new_url
  # JS-driven same-document fragment navigations (anchor clicks AND
  # `location.hash`/`href`/`assign` sets) are now handled entirely in
  # the VM by `tryFragmentNavigate` — they update the JS location and
  # fire `hashchange` there and never round-trip through here. This
  # path remains only as a defensive fallback for a fragment URL that
  # reaches the Ruby navigate/pending drain by some other route; keep
  # the VM's location object in sync so its `location.href` getter
  # doesn't read stale.
  @runtime.call('__csimUpdateLocation', new_url) if @runtime.respond_to?(:call)
end

#value(handle) ⇒ Object 



769
 
# File 'lib/capybara/simulated/browser.rb', line 769

def value(handle)        = dom_call('__csimValue', handle)

#viewport_height ⇒ Object 



1571
 
# File 'lib/capybara/simulated/browser.rb', line 1571

def viewport_height                 ; @viewport_height || 768  ; end

#viewport_width ⇒ Object 



1570
 
# File 'lib/capybara/simulated/browser.rb', line 1570

def viewport_width                  ; @viewport_width  || 1024 ; end

#visible?(handle) ⇒ Boolean

Returns:

  • (Boolean) 


762
 
# File 'lib/capybara/simulated/browser.rb', line 762

def visible?(handle)    = dom_call('__csimVisible', handle) ? true : false

#visible_text(handle) ⇒ Object 



767
 
# File 'lib/capybara/simulated/browser.rb', line 767

def visible_text(handle) = dom_call('__csimVisibleText', handle).to_s

#visit(url) ⇒ Object

Address-bar navigation: no Referer, and relative paths resolve against the host root (not the current page’s directory).



390
391
392
 
# File 'lib/capybara/simulated/browser.rb', line 390

def visit(url)
  navigate(resolve_visit_url(url), referer: nil)
end

#webauthn ⇒ Object 



3297
 
# File 'lib/capybara/simulated/browser.rb', line 3297

def webauthn = (@webauthn ||= WebauthnState.new)

#websocket_pending? ⇒ Boolean

Returns:

  • (Boolean) 


2638
 
# File 'lib/capybara/simulated/browser.rb', line 2638

def websocket_pending? = !@websocket_queue.empty?

#window_closed?(handle) ⇒ Boolean

Returns:

  • (Boolean) 


3070
 
# File 'lib/capybara/simulated/browser.rb', line 3070

def window_closed?(handle)       = @driver.respond_to?(:window_closed?)      ? @driver.window_closed?(handle.to_s)           : true

#window_location_of(handle) ⇒ Object 



3068
 
# File 'lib/capybara/simulated/browser.rb', line 3068

def window_location_of(handle)   = @driver.respond_to?(:window_location)     ? @driver.window_location(handle.to_s).to_s     : ''

#window_message_pending? ⇒ Boolean

Returns:

  • (Boolean) 


3081
 
# File 'lib/capybara/simulated/browser.rb', line 3081

def window_message_pending? = !@window_inbox.empty?

#with_modal(handler) ⇒ Object

Push a one-shot handler onto the modal-dialog stack — the next modal that fires consumes the topmost handler. Block exit pops in case the dialog never fired.



3691
3692
3693
3694
3695
3696
 
# File 'lib/capybara/simulated/browser.rb', line 3691

def with_modal(handler)
  @modal_handlers.push(handler)
  yield if block_given?
ensure
  @modal_handlers.delete(handler)
end

#worker_pending? ⇒ Boolean

Returns:

  • (Boolean) 


3045
 
# File 'lib/capybara/simulated/browser.rb', line 3045

def worker_pending? = !@worker_outbox.empty? || @worker_in_flight > 0

#worker_post_to_worker(handle, data) ⇒ Object 



3013
3014
3015
3016
3017
3018
 
# File 'lib/capybara/simulated/browser.rb', line 3013

def worker_post_to_worker(handle, data)
  w = @workers[handle.to_i]
  return unless w
  @worker_in_flight += 1
  w[:inbox] << data.to_s
end

#worker_spawn(url) ⇒ Object

── Web Workers ────────────────────────────────────────────────

‘new Worker(url)` in JS lands in `worker_spawn`. The Ruby thread it spawns owns a fresh V8 Context / QuickJS VM (true isolate, separate microtask queue and timer table), evals the worker script there, and runs an event loop draining timers, microtasks, and the inbox queue from the main thread. Each worker’s ‘__csim_workerPostMessage` host fn closes over its handle and routes outgoing messages onto a shared outbox the main settle drains.



2993
2994
2995
2996
2997
2998
2999
3000
3001
3002
3003
3004
3005
3006
3007
3008
3009
3010
3011
 
# File 'lib/capybara/simulated/browser.rb', line 2993

def worker_spawn(url)
  handle       = (@worker_seq += 1)
  inbox        = Thread::Queue.new
  outbox       = @worker_outbox
  engine_class = @runtime.class
  target       = resolve_against_current(url.to_s)
  # Resolve the worker script body on the main thread before
  # handing off to the worker. `blob:` URLs need the main VM's
  # blob registry; calling into the main runtime from a
  # non-owning thread SEGVs (V8 isolates are thread-
  # bound; quickjs.rb's VM is similarly per-thread).
  body = fetch_worker_script(target)
  thread = Thread.new do
    Thread.current.report_on_exception = false
    run_worker(handle, target, body, inbox, outbox, engine_class)
  end
  @workers[handle] = {thread: thread, inbox: inbox}
  handle
end

#worker_terminate(handle) ⇒ Object 



3020
3021
3022
3023
3024
3025
3026
3027
3028
3029
3030
3031
3032
 
# File 'lib/capybara/simulated/browser.rb', line 3020

def worker_terminate(handle)
  w = @workers.delete(handle.to_i)
  return unless w
  w[:inbox] << :terminate
  # Most clean shutdowns are <10 ms; the kill is the fallback
  # for blocked workers.
  w[:thread].join(WORKER_TERMINATE_GRACE)
  w[:thread].kill if w[:thread].alive?
  # A blocked worker that never returned messages leaves
  # `@worker_in_flight` permanently > 0; reset when no workers
  # remain so `polling?` can short-circuit again.
  @worker_in_flight = 0 if @workers.empty?
end

#write_document_cookie(s) ⇒ Object 



3643
3644
3645
3646
3647
3648
3649
3650
3651
3652
3653
3654
 
# File 'lib/capybara/simulated/browser.rb', line 3643

def write_document_cookie(s)
  return if s.nil? || s.empty?
  name, rest = s.split('=', 2)
  return if name.nil? || name.empty?
  parts = (rest || '').split(';').map(&:strip)
  value = parts.shift.to_s
  if cookie_deletion?(parts)
    @cookies.delete(name.strip)
  else
    @cookies[name.strip] = value
  end
end

#ws_close(id, code = 1000, reason = '') ⇒ Object 



2627
2628
2629
2630
2631
2632
2633
2634
2635
2636
 
# File 'lib/capybara/simulated/browser.rb', line 2627

def ws_close(id, code = 1000, reason = '')
  sock = @websocket_sockets[id.to_i] or return
  # Send the close frame and let the close HANDSHAKE complete: the server
  # replies with its own close frame, which the reader thread surfaces as
  # the `__close` event (carrying the agreed code) before tearing the
  # socket down in its `ensure`. Force teardown is `reset_websockets`'s job.
  payload = [code.to_i].pack('n') + reason.to_s.b
  ws_write_frame(sock, 0x8, payload) rescue nil
  nil
end

#ws_open(url, protocols = nil) ⇒ Object 



2570
2571
2572
2573
2574
2575
2576
2577
2578
2579
2580
2581
2582
2583
2584
2585
2586
2587
2588
2589
2590
2591
2592
2593
2594
2595
2596
2597
2598
2599
2600
2601
2602
2603
2604
2605
2606
2607
2608
2609
2610
2611
2612
2613
2614
 
# File 'lib/capybara/simulated/browser.rb', line 2570

def ws_open(url, protocols = nil)
  id       = (@websocket_seq += 1)
  # ws:// → http://, wss:// → https:// for the Rack env; resolve relative
  # against the current document (Action Cable's consumer builds an
  # absolute ws URL, but be tolerant).
  http_url = url.to_s.sub(/\Awss/i, 'https').sub(/\Aws/i, 'http')
  target   = resolve_against_current(http_url)
  key      = SecureRandom.base64(16)
  csim_io, app_io = Socket.pair(:UNIX, :STREAM, 0)
  env = Rack::MockRequest.env_for(target, method: 'GET')
  apply_default_request_env(env, referer: @current_url)
  env['HTTP_UPGRADE']               = 'websocket'
  env['HTTP_CONNECTION']            = 'Upgrade'
  env['HTTP_SEC_WEBSOCKET_KEY']     = key
  env['HTTP_SEC_WEBSOCKET_VERSION'] = '13'
  list = Array(protocols).map(&:to_s).reject(&:empty?)
  env['HTTP_SEC_WEBSOCKET_PROTOCOL'] = list.join(', ') unless list.empty?
  env['rack.hijack?']  = true
  env['rack.hijack']   = -> { app_io }
  env['rack.hijack_io'] = app_io
  # The app hijacks + writes the 101 (synchronously, or on its own event
  # loop thread — Action Cable handles the upgrade on a separate thread, so
  # `@app.call` may return before the handshake bytes appear; the reader
  # blocks until they do). Run it on the main thread like the long-poll
  # hijack so we don't race a second concurrent `@app.call`. (No handshake
  # timeout: a server that never writes the 101 leaks the reader+socket
  # until `reset_websockets` — acceptable, real servers always respond.)
  @app.call(env)
  @websocket_sockets[id]     = csim_io
  @websocket_app_sockets[id] = app_io
  accept = Digest::SHA1.base64digest(key + WS_GUID)
  queue  = @websocket_queue
  @websocket_threads[id] = Thread.new do
    Thread.current.report_on_exception = false
    run_websocket_reader(id, csim_io, accept, queue)
  end
  id
rescue StandardError => e
  # Nothing was registered for cleanup yet — close both pair ends here so
  # a failed upgrade (mis-routed URL, app error) doesn't leak fds.
  csim_io.close rescue nil
  app_io.close  rescue nil
  @websocket_queue << {id: id, type: '__error', message: "#{e.class}: #{e.message}"}
  id
end

#ws_send(id, data, binary = false) ⇒ Object

‘binary` is set by the JS side (it knows whether `send` was given a string or an ArrayBuffer/view) → opcode 0x2 vs the text 0x1. Action Cable is text-only (JSON). The payload’s bytes are written as-is.



2619
2620
2621
2622
2623
2624
2625
 
# File 'lib/capybara/simulated/browser.rb', line 2619

def ws_send(id, data, binary = false)
  sock = @websocket_sockets[id.to_i] or return
  ws_write_frame(sock, binary ? 0x2 : 0x1, data.to_s.b)
  nil
rescue StandardError
  nil
end

#xpath_shaped?(s) ⇒ Boolean

Returns:

  • (Boolean) 


650
651
652
653
654
655
656
657
658
 
# File 'lib/capybara/simulated/browser.rb', line 650

def xpath_shaped?(s)
  # Cheap probe: anything starting with `/` (absolute or relative
  # XPath), `(` (grouped XPath like `(//a)[1]`), or `./` /
  # `..` (XPath current-node + step) is XPath. We can't treat a
  # bare leading `.` as XPath because CSS class selectors look
  # exactly like that (`.contextual`); only the `./` form is
  # unambiguous.
  !!(s =~ %r{\A\s*(?:/|\(\s*/|\./|\.\.)})
end