Class: SAS::XPT::Reader

Inherits:
Object
  • Object
show all
Defined in:
lib/hlsv/xpt/reader.rb

Constant Summary collapse

TOP_HEADER =

The first header record consists of the following character string, in ASCII

"HEADER RECORD*******LIBRARY HEADER RECORD!!!!!!!000000000000000000000000000000  ".b
MEMBER_HEADER1 =

Notice the doc gives on page 4: "HEADER RECORDMEMBER HEADER RECORD!!!!!!!00000000000000000160000 0000140" But look at the dump on page 9, it is: "HEADER RECORDMEMBER HEADER RECORD!!!!!!!000000000000000001600000000140"

"HEADER RECORD*******MEMBER  HEADER RECORD!!!!!!!000000000000000001600000000140  ".b
MEMBER_HEADER1_VMS = 
"HEADER RECORD*******MEMBER  HEADER RECORD!!!!!!!000000000000000001600000000136  ".b
MEMBER_HEADER2 =

Note the 0140 that appears in the member header record above. This value specifies the size of the variable descriptor (NAMESTR) record that is described later in this document. On the VAX/VMS operating system, the value will be 0136 instead of 0140. This means that the descriptor will be only 136 bytes instead of 140.

"HEADER RECORD*******DSCRPTR HEADER RECORD!!!!!!!000000000000000000000000000000  ".b
NAMESTR_RECORD_START =

"HEADER RECORD*******NAMESTR HEADER RECORD!!!!!!!000000xxxx00000000000000000000 "

"HEADER RECORD*******NAMESTR HEADER RECORD!!!!!!!000000".b
NAMESTR_RECORD_VARS_RE = 
/^\d{4}$/
NAMESTR_RECORD_END = 
"00000000000000000000  ".b
OBSERVATION_HEADER = 
"HEADER RECORD*******OBS     HEADER RECORD!!!!!!!000000000000000000000000000000  ".b
SPECIAL_MISSING_VALUES = 
'_ABCDEFGHIJKLMNOPQRSTUVWXYZ'.b

Instance Attribute Summary collapse

Instance Method Summary collapse

Constructor Details

#initialize(xpt_path, input_encoding: 'binary', output_encoding: nil) ⇒ Reader

Returns a new instance of Reader.



57
58
59
60
61
62
 
# File 'lib/hlsv/xpt/reader.rb', line 57

def initialize(xpt_path, input_encoding: 'binary', output_encoding: nil)
  @xpt_path = xpt_path
  @input_encoding = input_encoding
  @output_encoding = output_encoding || input_encoding
  read_file
end

Instance Attribute Details

#input_encoding ⇒ Object (readonly)

input encoding of string values



49
50
51
 
# File 'lib/hlsv/xpt/reader.rb', line 49

def input_encoding
  @input_encoding
end

#library ⇒ Object (readonly)

output Library instance



55
56
57
 
# File 'lib/hlsv/xpt/reader.rb', line 55

def library
  @library
end

#output_encoding ⇒ Object (readonly)

output encoding for string values (default = input_encoding)



52
53
54
 
# File 'lib/hlsv/xpt/reader.rb', line 52

def output_encoding
  @output_encoding
end

#xpt_path ⇒ Object (readonly)

input file path



46
47
48
 
# File 'lib/hlsv/xpt/reader.rb', line 46

def xpt_path
  @xpt_path
end

Instance Method Details

#change_encoding(string) ⇒ Object 



343
344
345
346
347
348
 
# File 'lib/hlsv/xpt/reader.rb', line 343

def change_encoding(string)
  string.force_encoding(input_encoding)
  unless string.valid_encoding?
    issue "invalid input encoding #{input_encoding} for #{string.inspect}"
  end
end

#error(message) ⇒ Object 



356
357
358
359
 
# File 'lib/hlsv/xpt/reader.rb', line 356

def error(message)
  warn "ERROR: #{message}"
  raise "cannot continue"
end

#ibm_to_ieee(ibm_bytes) ⇒ Object

Convert IBM-format floating point (bytes) to IEEE 754 64-bit (float).



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
 
# File 'lib/hlsv/xpt/reader.rb', line 261

def ibm_to_ieee(ibm_bytes)
  # IBM mainframe:    sign * 0.mantissa * 16 ** (exponent - 64)
  # Python uses IEEE: sign * 1.mantissa * 2 ** (exponent - 1023)

  # Pad-out to 8 bytes if necessary. We expect 2 to 8 bytes, but
  # there's no need to check; bizarre sizes will cause a struct
  # module unpack error.
  if ibm_bytes.length < 8
   ibm_bytes = ibm_bytes.append_as_bytes("\x00\x00\x00\x00\x00\x00\x00\x00")[...8]
  end

  # parse the 64 bits of IBM float as one 8-byte unsigned long long
  ulong = ibm_bytes.unpack1('Q>')
  # puts "ulong = #{ulong}"

  # IBM: 1-bit sign, 7-bits exponent, 56-bits mantissa
  sign = ulong & 0x8000000000000000
  exponent = (ulong & 0x7f00000000000000) >> 56
  mantissa = ulong & 0x00ffffffffffffff
  # puts "sign = #{sign}"
  # puts "exponent = #{exponent}"
  # puts "mantissa = #{mantissa}"

  if mantissa == 0
    if ibm_bytes[0] == "\x00".b
      return 0.0
    elsif ibm_bytes[0] == "\x80".b
      return -0.0
    elsif ibm_bytes[0] == '.'.b
      return nil
    elsif SPECIAL_MISSING_VALUES.include?(ibm_bytes[0])
      return :"#{ibm_bytes[0]}"
    else
      raise "Neither 'true' zero nor NaN: #{ibm_bytes.inspect}"
    end
  end

  # IBM-format exponent is base 16, so the mantissa can have up to 3
  # leading zero-bits in the binary mantissa. IEEE format exponent
  # is base 2, so we don't need any leading zero-bits and will shift
  # accordingly. This is one of the criticisms of IBM-format, its
  # wobbling precision.
  if (ulong & 0x0080000000000000) != 0
    shift = 3
  elsif (ulong & 0x0040000000000000) != 0
    shift = 2
  elsif (ulong & 0x0020000000000000) != 0
    shift = 1
  else
    shift = 0
  end
  mantissa >>= shift
  # puts "shift = #{shift}"
  # puts "mantissa = #{mantissa}"

  # clear the 1 bit to the left of the binary point
  # this is implicit in IEEE specification
  mantissa &= 0xffefffffffffffff
  # puts "mantissa = #{mantissa}"

  # IBM exponent is excess 64, but we subtract 65, because of the
  # implicit 1 left of the radix point for the IEEE mantissa
  exponent -= 65
  # puts "exponent = #{exponent}"
  # IBM exponent is base 16, IEEE is base 2, so we multiply by 4
  exponent <<= 2
  # puts "exponent = #{exponent}"
  # IEEE exponent is excess 1023, but we also increment for each
  # right-shift when aligning the mantissa's first 1-bit
  exponent += shift + 1023
  # puts "exponent = #{exponent}"

  # IEEE: 1-bit sign, 11-bits exponent, 52-bits mantissa
  # We didn't shift the sign bit, so it's already in the right spot
  ieee = sign | (exponent << 52) | mantissa
  # puts "ieee = #{ieee}"
  result = [ieee].pack('Q>').unpack1('G')
  # puts "result = #{result}"

  result
end

#issue(message, actual = nil, expected = nil) ⇒ Object 



350
351
352
353
354
 
# File 'lib/hlsv/xpt/reader.rb', line 350

def issue(message, actual = nil, expected = nil)
  warn message
  warn "- expected: #{expected.inspect}" if expected
  warn "- actual:   #{actual.inspect}" if actual
end

#read_file ⇒ Object 



64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
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
 
# File 'lib/hlsv/xpt/reader.rb', line 64

def read_file

  File.open(xpt_path, 'rb') do |io|

    # All transport data set records are 80 bytes in length.
    # If there is not sufficient data to reach 80 bytes,
    # then a record is padded with ASCII blanks to 80 bytes.

    # top header
    top_header = io.read(TOP_HEADER.length)
    top_header == TOP_HEADER or issue "invalid top header"

    # first_header
    _sas_sas_saslib = io.read(24)   # "SAS     SAS     SASLIB  "
    file_sas_version = io.read(8)   # "vvvvvvvv"
    file_sas_os = io.read(8)        # "oooooooo"
    _blanks = io.read(24)
    file_create_date = io.read(16)  # "ddMMMyy:hh:mm:ss"

    # second header
    file_modify_date = io.read(16)  # "ddMMMyy:hh:mm:ss"
    _blanks = io.read(64)

    @library = Library.new(xpt_path, file_create_date, file_modify_date, file_sas_version, file_sas_os)

    until io.eof?

      # member header record 1
      member_header1 = io.read(80)
      case member_header1
      when MEMBER_HEADER1
        namestr_record_length = 140
      when MEMBER_HEADER1_VMS
        namestr_record_length = 136
      else
        issue "invalid header record 1"
        namestr_record_length = 140
      end

      # member header record 2, constant
      member_header2 = io.read(80)
      member_header2 == MEMBER_HEADER2 or issue "invalid member header record 2"

      # member header data
      # "SAS     dsname  SASDATA version>os>>>>>> (24 blanks) ddMMMyy:hh:mm:ss"
      _sas            = io.read(8)    # "SAS     "
      ds_name         = io.read(8)    # "DM      "
      _sasdata        = io.read(8)    # "SASDATA "
      ds_sas_version  = io.read(8)    # "9.4     "
      ds_sas_os       = io.read(8)    # "X64_SRV1"
      _blanks         = io.read(24)   # "                        "
      ds_create_date  = io.read(16)   # "26MAY21:15:57:45"

      # second member header data
      # "ddMMMyy:hh:mm:ss (16 padding) (40 label) (8 dstype)"

      ds_modify_date = io.read(16)  # "26MAY21:15:57:45"
      _blanks        = io.read(16)  # "                "
      ds_label       = io.read(40)  # "Demographics                            "
      ds_type        = io.read(8)   # "        "

      input_encoding != 'binary' and change_encoding ds_label
      output_encoding != input_encoding and ds_label.encode!(output_encoding)

      dataset = Dataset.new(ds_name.rstrip, ds_label.rstrip, ds_type.rstrip, ds_create_date, ds_modify_date, ds_sas_version.rstrip, ds_sas_os.rstrip)
      @library.datasets << dataset

      # Namestr header record

      # In this header record, xxxx is the number of variables in the data set,
      # displayed with blank-padded numeric characters. For example, for 2 variables, xxxx=0002.
      # xxxx occurs at offset 54 (base 0 as in C language use).
      namestr_header = io.read(80)
      namestr_record_start = namestr_header[...NAMESTR_RECORD_START.length]
      variable_count = namestr_header[NAMESTR_RECORD_START.length, 4]
      namestr_record_end = namestr_header[-NAMESTR_RECORD_END.length..]
      namestr_record_start == NAMESTR_RECORD_START or issue "invalid namestr record start", namestr_record_start, NAMESTR_RECORD_START
      namestr_record_end == NAMESTR_RECORD_END or issue "invalid namestr record end", namestr_record_end, NAMESTR_RECORD_END
      variable_count =~ NAMESTR_RECORD_VARS_RE or error "invalid namestr variable count #{variable_count.inspect}"
      variable_count = variable_count.to_i

      # Namestr records

      # Each namestr field is 140 bytes long, but the fields are streamed together and broken in 80-byte pieces.
      # If the last byte of the last namestr field does not fall in the last byte of the 80-byte record, the record is padded with ASCII blanks to 80 bytes.
      # Here is the C structure definition for the namestr record:
      #
      # struct NAMESTR {
      #   short ntype; /* VARIABLE TYPE: 1=NUMERIC, 2=CHAR */
      #   short nhfun; /* HASH OF NNAME (always 0) */
      #   short nlng; /* LENGTH OF VARIABLE IN OBSERVATION */
      #   short nvar0; /* VARNUM */
      #   char8 nname; /* NAME OF VARIABLE */
      #   char40 nlabel; /* LABEL OF VARIABLE */
      #   char8 nform; /* NAME OF FORMAT */
      #   short nfl; /* FORMAT FIELD LENGTH OR 0 */
      #   short nfd; /* FORMAT NUMBER OF DECIMALS */
      #   short nfj; /* 0=LEFT JUSTIFICATION, 1=RIGHT JUST */
      #   char nfill[2]; /* (UNUSED, FOR ALIGNMENT AND FUTURE) */
      #   char8 niform; /* NAME OF INPUT FORMAT */
      #   short nifl; /* INFORMAT LENGTH ATTRIBUTE */
      #   short nifd; /* INFORMAT NUMBER OF DECIMALS */
      #   long npos; /* POSITION OF VALUE IN OBSERVATION */
      #   char rest[52]; /* remaining fields are irrelevant */
      # };
      #
      # Note that the length given in the last 4 bytes of the member header record
      # indicates the actual number of bytes for the NAMESTR structure. The size of
      # the structure listed above is 140 bytes. Under VAX/VMS, the size will be 136
      # bytes, meaning that the 'rest' variable may be truncated.

      variable_count.times do
        dataset.variables << Variable.new(io.read(namestr_record_length), input_encoding: input_encoding, output_encoding: output_encoding)
      end

      padding = 80 - (namestr_record_length * variable_count) % 80
      if padding < 80
        _padding = io.read(padding)
      end

      # Observation header
      observation_header = io.read(80)
      observation_header == OBSERVATION_HEADER or issue "invalid observation header"

      # Data records
      # Data records are streamed in the same way that namestrs are.
      # There is ASCII blank padding at the end of the last record if necessary.
      # There is no special trailing record.

      # Missing Values
      # Missing values are written out with the first byte (the exponent) indicating the proper missing values.
      # All subsequent bytes are 0x00.
      # The first byte is:
      # type  byte
      # ._    0x5f
      # .     0x2e
      # .A    0x41
      # .B    0x42
      # ...
      # .Z    0x5a

      obs_record_length = dataset.obs_record_length
      record_count = 0
      # by construction, the 1st obs record starts on a new 80-byte record
      obs_start_pos = io.pos
      # puts "obs_record_length = #{obs_record_length} obs_start_pos = #{obs_start_pos}"

      blank_record = ' '.b * obs_record_length

      loop do
        break if io.eof?
        # read_pos = io.pos
        buffer = io.read(obs_record_length)
        # puts "reading at #{read_pos} -> #{buffer.length} bytes #{to_hex(buffer)}"
        if buffer.length != obs_record_length
          # puts "#{buffer.length} != #{obs_record_length} -> #{record_count} records"
          break
        end
        if buffer == blank_record
          read_so_far = io.pos - obs_start_pos
          _records_read, current_record_position = read_so_far.divmod(80)
          until_end_of_record = 80 - current_record_position
          if until_end_of_record > 0
            buffer = io.read(until_end_of_record)
            if !buffer.strip.empty?
              error "non-blank bytes at end of records: #{buffer.inspect}"
            end
            break
          end
        end
        record_count += 1
        obs = []
        dataset.variables.each do |var|
          value = buffer[var.position, var.length]
          if var.type == :char
            # puts "char: #{value.inspect}"
            value = value.rstrip
            input_encoding != 'binary' and change_encoding value
            output_encoding != input_encoding and value.encode!(output_encoding)
            obs << value
          else
            # display = value.bytes.map { |b| "%02x" % b }.join
            # puts "num: #{display}"
            obs << ibm_to_ieee(value)
            # TODO? option: ._, .A -> .Z to :_, :A, :Z or nil
            # TODO? option: to_i == to_f => to_i
          end
        end
        dataset.observations << obs
        # exit
      end
      # puts "final pos: #{io.pos} eof: #{io.eof?}"
    end
  end
end

#to_hex(byte_string) ⇒ Object 



361
362
363
 
# File 'lib/hlsv/xpt/reader.rb', line 361

def to_hex(byte_string)
  byte_string.bytes.map { |b| "%02x" % b }.join(' ')
end