Class: Bundler::Spinel::History

Inherits:
Object
  • Object
show all
Defined in:
lib/bundler/spinel/history.rb

Overview

The historical record: how the catalog shifts as the Spinel compiler evolves. Each “run” is a full corpus re-probe at one engine revision; this renders the verdict-mix timeline + the gem-level deltas between consecutive runs — the bug pipeline closing the loop, made visible.

Constant Summary collapse

RUNS =

Ordered runs (oldest→newest). Each: rev, date, the engine commit subject, the full-corpus probe snapshot, and a curated note on what moved.

[
  { rev: "a03bb49", date: "2026-05-28", commit: "the first full-corpus survey (189,742 gems)",
    file: "survey-fresh/compat.jsonl", note: nil },
  { rev: "8d88ebe", date: "2026-05-29", commit: "module/reflection + GC fixes (is_a?, .class, respond_to?, #1052)",
    file: "survey-8d88ebe/compat.jsonl",
    note: "Mixed: the module-object fixes graduated gems, but <code>96b21e6</code> " \
          "(module_function support) <strong>regressed</strong> ~160 gems whose objects " \
          "were built from a variable-held class (the <code>brass</code> cluster) — caught " \
          "by the re-probe and filed as <a href=\"https://github.com/matz/spinel/issues/1062\">matz/spinel#1062</a>." },
  { rev: "f8040f3", date: "2026-05-31", commit: "DCE for synthetic module class methods (#1062 fix), instance_methods const-fold (#1073), Array#transpose, map→array",
    file: "survey-f8040f3/compat.jsonl",
    note: "<strong>Recovery + gains.</strong> matz bisected #1062 to <code>96b21e6</code> and fixed it " \
          "(<code>e2e010c</code>); together with the new <code>instance_methods</code> const-fold " \
          "(<a href=\"https://github.com/matz/spinel/issues/1073\">#1073</a>) and <code>transpose</code>/map " \
          "specializations, the brass cluster and thousands more moved out of <code>rejected</code>." },
  { rev: "95557f5", date: "2026-06-02", commit: "module/class-body side effects + lexical const refs (#1256), Regexp.last_match(n) (#1257), preserve Float-in-Hash (#1258), Struct typing, JSON.generate, alias, +14 more",
    file: "survey-95557f5/compat.jsonl",
    note: "<strong>The biggest single jump yet.</strong> 22 upstream commits — including fixes for three " \
          "issues this harness filed (<a href=\"https://github.com/matz/spinel/issues/1256\">#1256</a> module-body, " \
          "<a href=\"https://github.com/matz/spinel/issues/1257\">#1257</a> <code>Regexp.last_match</code>, " \
          "<a href=\"https://github.com/matz/spinel/issues/1258\">#1258</a> Float-in-Hash) plus Struct typing, " \
          "<code>alias</code>, <code>JSON.generate</code> for records and more — moved <strong>20,175</strong> gems " \
          "out of <code>rejected</code>, among them <code>rspec</code>, <code>globalid</code>, " \
          "<code>mini_portile2</code> and <code>coffee-rails</code>." },
  { rev: "a782696", date: "2026-06-03", commit: "StringScanner unscan/check + Error, Time#to_s + puts-nil, Dir.exist? + alias_method dispatch, missing int-hash keys as nil, RBS extractor heterogeneous-union→poly, subclass-initialize poly unification (13 commits)",
    file: "survey-a782696/compat.jsonl",
    note: "<strong>A consolidation rev.</strong> The base verdict mix is essentially flat after the previous " \
          "jump — 13 upstream commits of correctness fixes (<code>StringScanner</code>, <code>Time#to_s</code>, " \
          "<code>Dir.exist?</code>/<code>alias_method</code>, and the RBS-extractor union→poly change) graduated a " \
          "small set of gems — <code>google-adwords-api</code>, <code>libdatadog</code>, <code>random_user_agent</code>, " \
          "<code>twitter_username_extractor</code> and the <code>redcar-*</code> cluster — while a couple regressed " \
          "and were caught by the re-probe. The bigger story this rev was off the catalog: the harness found " \
          "<code>spinel_analyze</code> consuming 100+ GB on a cluster of auto-generated API-SDK gems (a compiler " \
          "memory blow-up, filed upstream)." },
  { rev: "9c0a5f0", date: "2026-06-04", commit: "79 commits — incl. fixes for 6 harness-filed issues: stdlib-class-in-ivar (#1305), reopen-Object (#1306), lambda/proc branch-local (#1315), &blk+block_given? (#1316), inject(&:sym) (#1317), ignored-require constant (#1273)",
    file: "survey-9c0a5f0/compat.jsonl",
    note: "<strong>The harness loop paying off.</strong> matz landed fixes for <strong>six</strong> issues this " \
          "harness filed the day before — all common idioms: <code>block_given?</code> with a named " \
          "<code>&amp;blk</code> (<a href=\"https://github.com/matz/spinel/issues/1316\">#1316</a>), " \
          "<code>inject(&amp;:+)</code> (<a href=\"https://github.com/matz/spinel/issues/1317\">#1317</a>), " \
          "reopening <code>class Object</code> (<a href=\"https://github.com/matz/spinel/issues/1306\">#1306</a>), " \
          "a stdlib class held in an instance variable " \
          "(<a href=\"https://github.com/matz/spinel/issues/1305\">#1305</a>), and a branch-assigned local inside a " \
          "lambda/proc (<a href=\"https://github.com/matz/spinel/issues/1315\">#1315</a>). <strong>3,487</strong> gems " \
          "moved out of <code>rejected</code> (110.3k→106.8k). The one feature ruled out of scope — aliasing the " \
          "regexp special globals (<a href=\"https://github.com/matz/spinel/issues/1307\">#1307</a>) — now fails with a " \
          "clear diagnostic instead of bad C." },
  { rev: "5c9790c", date: "2026-06-05", commit: "17 commits — fixes for 3 harness-filed typed-collection issues: Hash#fetch on int_int_hash (#1329), Array#join on poly_array (#1332), Class-in-collection→poly (#1337); plus regex line-anchoring/gsub-buffer + first-class string type",
    file: "survey-5c9790c/compat.jsonl",
    note: "<strong>Typed-collection coverage.</strong> matz fixed three issues this harness filed hours earlier — all " \
          "the same shape: a method that exists on the generic path but was missing on a <em>specialized</em> " \
          "collection. <code>Hash#fetch</code> on an int→int hash " \
          "(<a href=\"https://github.com/matz/spinel/issues/1329\">#1329</a>), <code>Array#join</code> on a mixed " \
          "<code>poly_array</code> (<a href=\"https://github.com/matz/spinel/issues/1332\">#1332</a>), and a " \
          "<code>Class</code> value stored in a Hash/Array now typed as poly instead of int " \
          "(<a href=\"https://github.com/matz/spinel/issues/1337\">#1337</a>, which had broken every options-hash " \
          "carrying an exception class). The compile+scan base barely moves on fixes like these — they're " \
          "full-surface/runtime, so the graduation shows in the behaviour-verified tier." },
  { rev: "57af7f9", date: "2026-06-07", commit: "~40 commits — 9 harness-filed issues closed in one wave: the ecosystem-spine front doors (alias→attr_reader #1356, &:sym-after-positional parse #1359), unary operator mangling (#1357), sp_sym_intern link (#1355), plus typed-collection nil steps (#801/#1180) and #line / --emit-symbol-map diagnostics (the #1338 RFC direction)",
    file: "survey-57af7f9/compat.jsonl",
    note: "<strong>The spine-gems wave.</strong> Auditing why <code>bundler</code>/<code>rake</code>/" \
          "<code>minitest</code>/<code>thor</code> reject found two shallow front doors — " \
          "<code>alias</code> to an <code>attr_reader</code>-generated method " \
          "(<a href=\"https://github.com/matz/spinel/issues/1356\">#1356</a>, rake + thor) and " \
          "<code>&amp;:sym</code> after a positional argument mis-parsed as a hash literal " \
          "(<a href=\"https://github.com/matz/spinel/issues/1359\">#1359</a>, bundler + minitest) — and matz closed " \
          "both within a day, alongside 7 more harness filings. All four spine gems now compile past their old " \
          "blockers into distinct second-tier issues " \
          "(<a href=\"https://github.com/matz/spinel/issues/1368\">#1368</a> et al.). C compile errors now map back " \
          "to Ruby source lines via <code>#line</code>, on by default — the " \
          "<a href=\"https://github.com/matz/spinel/issues/1338\">#1338</a> RFC direction. The behaviour-verified " \
          "tier reached 144 mechanical ★ this run." },
  { rev: "cb23cc6", date: "2026-06-09", commit: "~23 commits — 12 harness-filed issues closed in one wave, incl. Mutex/Monitor#synchronize + Thread.new now running and carrying their block value (#1360), class-instance-vars in class methods (#1352), is_a?(IncludedModule)/ancestors (#1350), \\h/\\H regex (#1349), reopened-builtin optional defaults (#1348), bitwise/shift operator mangling (#1358/#1368), the captured-&block value/type family",
    file: "survey-cb23cc6/compat.jsonl",
    note: "<strong>Mutex/Thread come in from the cold.</strong> With matz/spinel#1360 landing " \
          "(<code>Mutex#synchronize</code>/<code>Thread.new</code> now run single-threaded and carry their " \
          "block's value), the static pre-filter no longer hard-rejects them — they're flagged " \
          "<code>risky</code> (correct for defensive use, degenerate only for true concurrency). Re-probing " \
          "the 8,833 gems that had been rejected on sight for <code>Mutex.new</code>/<code>Thread.new</code>, " \
          "<strong>3,493 (39.5%) moved out of <code>rejected</code></strong> — they compile now; the static " \
          "wall had been hiding it. The four spine gems (<code>bundler</code>/<code>rake</code>/" \
          "<code>minitest</code>/<code>thor</code>) now reject for their <em>real</em> reason — the deep " \
          "metaprogramming surface (<code>send</code>/<code>method_missing</code>/<code>cattr_accessor</code>) — " \
          "not a misleading <code>hard:Mutex.new</code>. The differential re-audit also caught one regression in " \
          "the wave: returning <code>self</code> from a reopened-builtin method " \
          "(<a href=\"https://github.com/matz/spinel/issues/1386\">#1386</a>, <code>to-bool</code>), filed same day." },
  { rev: "b60fbd7", date: "2026-06-15", commit: "636 commits — the largest wave yet. A type-inference + codegen rewrite (typed-array/poly ~222, inference/cast/box ~168, string ~124, block/yield ~101, module/singleton ~94) closing many harness-filed issues at once: reopened-builtin self (#1386), hash-block String type (#1382/#1394), circular require_relative (#1373), computed require (#1383), class-method yield (#1387), Scheduled-server boot + SIGTERM (#1369/#1384). New surfaces: value-type objects, --rbs/--emit-rbs/--emit-symbol-map.",
    file: "survey-b60fbd7/compat.jsonl",
    note: "<strong>The largest movement in the catalog's history.</strong> 636 upstream commits — a type-inference " \
          "rewrite that resolved the entire <code>unresolved:&lt;method&gt;</code> reject family (~60k records → ~0): " \
          "<strong>~49,000 gems moved out of <code>rejected</code></strong> (−46%), +22k to <code>clean</code>. The " \
          "wave also changed unresolvable <code>require</code> to hard-fail the compile (post-#1383); because " \
          "<code>require \"gem/version\"</code> is near-universal that spuriously rejected thousands, so the probe " \
          "now classifies a require-only failure as the no-load-path limit (<code>risky [load-path:require]</code>, " \
          "9,716 gems corrected). Cost side, caught by re-verifying every ★ at this rev: <strong>~34 of 175 " \
          "behaviour-verified gems regressed</strong> (12 miscompiles incl. <code>after_commit_action</code>, 22 " \
          "codegen incl. bare-<code>super</code>/<code>ForwardingSuperNode</code>) — the trust tier dropped 175→136. " \
          "Fresh-verify overrode stickiness, so the ★ count is honest, not carried." },
].freeze
ORDER = 
%w[clean risky rejected].freeze

Instance Method Summary collapse

Constructor Details

#initialize(base = ".") ⇒ History

Returns a new instance of History.



116
117
118
 
# File 'lib/bundler/spinel/history.rb', line 116

def initialize(base = ".")
  @base = base
end

Instance Method Details

#build_html(out) ⇒ Object

Render the full history page to ‘out`.



121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
 
# File 'lib/bundler/spinel/history.rb', line 121

def build_html(out)
  runs = RUNS.map { |r| r.merge(tally: tally(File.join(@base, r[:file]))) }
            .select { |r| r[:tally] }
  body = +""
  body << "<h1>How the catalog has changed</h1>\n"
  body << %(<p class="lede">Each row is a full re-probe of the ~190k-gem corpus at one )
  body << %(<a href="https://github.com/matz/spinel">Spinel</a> revision. The harness's )
  body << %(real product is a stream of focused compiler bugs; this is the loop closing — )
  body << %(fixes landing upstream, gems graduating out of <code>rejected</code>.</p>\n)

  body << timeline_table(runs)
  runs.each_cons(2) { |a, b| body << delta_card(a, b) }

  File.write(out, page("History — SpinelGems", body))
  out
end