yeptris 0.6.33.2-aarch64-linux → 0.6.34.2-aarch64-linux

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
checksums.yaml CHANGED
@@ -1,7 +1,7 @@
1
1
  ---
2
2
  SHA256:
3
- metadata.gz: 4f539306a31099d38b0040d34986abe9aecaf7f4f7950f78fbd828002d0d186d
4
- data.tar.gz: a325b02fec0989a9d21ee8615c7993d05e9a9b2f7062a64560656a4699160109
3
+ metadata.gz: 35e1c278d17a938cfdf797f05231844362cd5eff838dcc07f7f6626ac6b0615a
4
+ data.tar.gz: f0a05034530c91ccde139ba2c64c1deb88ded39a752e1e281480499b1f6c1052
5
5
  SHA512:
6
- metadata.gz: 0ac10a304643b6fa2dd77979974b0809c6032c8918a9c9a9e192667e479338f0b037a45f1c240e20dccb110f354d46854c3cde5414d14a861a9577b353baba83
7
- data.tar.gz: 1ba4bb2d25f6c2cbca95360336b8a8827533dace201394f8f92c2cea6395743bee00613963aa650ea2e9cafdc4ff647e7a5db4b251b25086a85491bfee6c6349
6
+ metadata.gz: 43f40a6018a4663aeca90017a7bf631dfa4f8acebd324463711a4103a4f9435e884cec06512ab4ca1570bde1efe108097443c5ad4e5b16b94458d0f0fbaead46
7
+ data.tar.gz: 3f117e473a8100ed25ff2253ddd38cabca9c945545d0af67011ff2ae016d22ac72b6f61e5d68e629e21489f29cc169a9cdecdc08a33e44794c39e181a11944ac
data/lib/yeptris/json.rb CHANGED
@@ -45,7 +45,14 @@ module Yeptris
45
45
  # yeptris-ruby#278: with no Ruby-version-matched native build the
46
46
  # pure-Ruby walk measured 0.11x vs JSON.parse on real documents —
47
47
  # the marshal engine keeps the fallback C-side end to end.
48
- return marshal_engine(source, 0) if ::Yeptris::FFI::MARSHAL_STRICT_DUP
48
+ # STRICT_DUPLICATE_KEYS rides the flags: json >= 3's default
49
+ # contract raises on duplicates. The flags were hardcoded to 0
50
+ # here, so every lane without the native ext (windows-latest has
51
+ # none) answered last-wins under json >= 3 — found by canon's
52
+ # dependent CI, whose parity pin now names the lane.
53
+ return marshal_engine(
54
+ source, STRICT_DUPLICATE_KEYS ? ::Yeptris::FFI::MARSHAL_STRICT_DUP_KEYS : 0
55
+ ) if ::Yeptris::FFI::MARSHAL_STRICT_DUP
49
56
  tape_engine(source)
50
57
  end
51
58
 
data/lib/yeptris/yaml.rb CHANGED
@@ -1,5 +1,10 @@
1
1
  # frozen_string_literal: true
2
2
 
3
+ # The BulkBuilder's `when Date` must not depend on the caller loading
4
+ # stdlib's date (issue #289: a clean `require "yeptris"` raised NameError
5
+ # before the DumpError branch could run). Psych requires date the same way.
6
+ require "date"
7
+
3
8
  # Self-sufficient (issue #95): the neutral surface is a valid entry
4
9
  # point on its own — lutaml-model requires exactly this file.
5
10
  require "yeptris"
data/lib/yeptris.rb CHANGED
@@ -3,7 +3,7 @@
3
3
  module Yeptris
4
4
  # The gem's version lives in the parent namespace's file — the last
5
5
  # internal require (yeptris/version) retired with it.
6
- VERSION = "0.6.33.2".freeze
6
+ VERSION = "0.6.34.2".freeze
7
7
 
8
8
  # Informational notices are OPT-IN (yeptris-ruby#217): consumers
9
9
  # embed yeptris inside CLIs that assert clean stderr, and a
data/libyeptris.so CHANGED
Binary file
@@ -3,7 +3,7 @@
3
3
  cmake_minimum_required(VERSION 3.20)
4
4
 
5
5
  project(yeptris
6
- VERSION 0.6.33
6
+ VERSION 0.6.34
7
7
  DESCRIPTION "Ultra-fast YAML 1.2 parser, emitter and streamer in C"
8
8
  LANGUAGES C CXX
9
9
  )
@@ -54,6 +54,7 @@ set(YEPTRIS_SOURCES
54
54
  yeptris/emit/emit.c
55
55
  yeptris/emit/float/print.c
56
56
  yeptris/emit/float/dragon.c
57
+ yeptris/emit/float/fpconv.c
57
58
  )
58
59
 
59
60
  if(YEPTRIS_WITH_CBOR)
@@ -43,6 +43,17 @@ YEPTRIS_API YeptrisStatus yeptris_marshal(const char* data, size_t len, YeptrisS
43
43
  * bulk path behind Node#to_ruby (one call, not one per node). */
44
44
  YEPTRIS_API YeptrisStatus yeptris_marshal_node(YeptrisNode node, char** out, size_t* out_len);
45
45
 
46
+ #define YEPTRIS_MARSHAL_STRICT_DUP_KEYS 1u
47
+
48
+ /* yeptris_marshal_node with flags. STRICT_DUP_KEYS raises
49
+ * YEPTRIS_ERROR_PARSE ("duplicate key in object") when any map carries
50
+ * the same key twice — json 3's default contract; the JSON DOM keeps
51
+ * both pairs, and the non-strict stream stays last-wins (json 2.x).
52
+ * Memory: *out is malloc'd, free with yeptris_marshal_free exactly
53
+ * once; on failure nothing is allocated. */
54
+ YEPTRIS_API YeptrisStatus yeptris_marshal_node_ex(YeptrisNode node, uint32_t flags, char** out,
55
+ size_t* out_len);
56
+
46
57
  YEPTRIS_API void yeptris_marshal_free(char* out);
47
58
 
48
59
  #ifdef __cplusplus
@@ -31,6 +31,11 @@ int yep_d2fixed(double d, uint32_t precision, char* buf);
31
31
  /* YAML words for non-finite values (finite input returns 0). */
32
32
  int yep_d2s_nonfinite(double d, char* buf);
33
33
 
34
+ /* The json-gem float byte shape (grisu2 digits + fpconv emit
35
+ * windows, vendored in fpconv.c) for JSON-mode canonical scalars
36
+ * (yeptris#651). 32-byte buffer. Returns the length written. */
37
+ int yep_d2s_json(double d, char* buf);
38
+
34
39
  #ifdef __cplusplus
35
40
  }
36
41
  #endif
@@ -0,0 +1,428 @@
1
+ /* fpconv.c — the json-gem float byte shape (yeptris#651).
2
+ *
3
+ * Vendored from night-shift/fpconv in the variant carried by the Ruby
4
+ * json package (ruby/json ext/json/ext/vendor/fpconv.c), which appends
5
+ * ".0" to integral floats. JSON.generate on json >= 2.10 emits floats
6
+ * through exactly this code, so byte parity for the JSON writer
7
+ * (emit_canonical_scalar's JSON route) requires the same digits AND
8
+ * the same emit windows: grisu2 is not always-shortest, so the
9
+ * clean-room shortest printer cannot be reshaped to match.
10
+ *
11
+ * Boost Software License - Version 1.0 - August 17th, 2003.
12
+ * Permission is hereby granted, free of charge, to any person or
13
+ * organization obtaining a copy of this software (see
14
+ * https://www.boost.org/LICENSE_1_0.txt). Attribution preserved per
15
+ * the vendoring discipline (cf. the ryu port plan).
16
+ */
17
+
18
+ #include <stdbool.h>
19
+ #include <stdint.h>
20
+ #include <string.h>
21
+
22
+ #include "emit/float/api.h"
23
+
24
+ #define npowers 87
25
+ #define steppowers 8
26
+ #define firstpower -348 /* 10 ^ -348 */
27
+
28
+ #define expmax -32
29
+ #define expmin -60
30
+
31
+ typedef struct Fp {
32
+ uint64_t frac;
33
+ int exp;
34
+ } Fp;
35
+
36
+ static const Fp powers_ten[] = {
37
+ {18054884314459144840U, -1220}, {13451937075301367670U, -1193}, {10022474136428063862U, -1166},
38
+ {14934650266808366570U, -1140}, {11127181549972568877U, -1113}, {16580792590934885855U, -1087},
39
+ {12353653155963782858U, -1060}, {18408377700990114895U, -1034}, {13715310171984221708U, -1007},
40
+ {10218702384817765436U, -980}, {15227053142812498563U, -954}, {11345038669416679861U, -927},
41
+ {16905424996341287883U, -901}, {12595523146049147757U, -874}, {9384396036005875287U, -847},
42
+ {13983839803942852151U, -821}, {10418772551374772303U, -794}, {15525180923007089351U, -768},
43
+ {11567161174868858868U, -741}, {17236413322193710309U, -715}, {12842128665889583758U, -688},
44
+ {9568131466127621947U, -661}, {14257626930069360058U, -635}, {10622759856335341974U, -608},
45
+ {15829145694278690180U, -582}, {11793632577567316726U, -555}, {17573882009934360870U, -529},
46
+ {13093562431584567480U, -502}, {9755464219737475723U, -475}, {14536774485912137811U, -449},
47
+ {10830740992659433045U, -422}, {16139061738043178685U, -396}, {12024538023802026127U, -369},
48
+ {17917957937422433684U, -343}, {13349918974505688015U, -316}, {9946464728195732843U, -289},
49
+ {14821387422376473014U, -263}, {11042794154864902060U, -236}, {16455045573212060422U, -210},
50
+ {12259964326927110867U, -183}, {18268770466636286478U, -157}, {13611294676837538539U, -130},
51
+ {10141204801825835212U, -103}, {15111572745182864684U, -77}, {11258999068426240000U, -50},
52
+ {16777216000000000000U, -24}, {12500000000000000000U, 3}, {9313225746154785156U, 30},
53
+ {13877787807814456755U, 56}, {10339757656912845936U, 83}, {15407439555097886824U, 109},
54
+ {11479437019748901445U, 136}, {17105694144590052135U, 162}, {12744735289059618216U, 189},
55
+ {9495567745759798747U, 216}, {14149498560666738074U, 242}, {10542197943230523224U, 269},
56
+ {15709099088952724970U, 295}, {11704190886730495818U, 322}, {17440603504673385349U, 348},
57
+ {12994262207056124023U, 375}, {9681479787123295682U, 402}, {14426529090290212157U, 428},
58
+ {10748601772107342003U, 455}, {16016664761464807395U, 481}, {11933345169920330789U, 508},
59
+ {17782069995880619868U, 534}, {13248674568444952270U, 561}, {9871031767461413346U, 588},
60
+ {14708983551653345445U, 614}, {10959046745042015199U, 641}, {16330252207878254650U, 667},
61
+ {12166986024289022870U, 694}, {18130221999122236476U, 720}, {13508068024458167312U, 747},
62
+ {10064294952495520794U, 774}, {14996968138956309548U, 800}, {11173611982879273257U, 827},
63
+ {16649979327439178909U, 853}, {12405201291620119593U, 880}, {9242595204427927429U, 907},
64
+ {13772540099066387757U, 933}, {10261342003245940623U, 960}, {15290591125556738113U, 986},
65
+ {11392378155556871081U, 1013}, {16975966327722178521U, 1039}, {12648080533535911531U, 1066}};
66
+
67
+ static Fp find_cachedpow10(int exp, int* k) {
68
+ const double one_log_ten = 0.30102999566398114;
69
+
70
+ int approx = (int)(-(exp + npowers) * one_log_ten);
71
+ int idx = (approx - firstpower) / steppowers;
72
+
73
+ while (1) {
74
+ int current = exp + powers_ten[idx].exp + 64;
75
+
76
+ if (current < expmin) {
77
+ idx++;
78
+ continue;
79
+ }
80
+
81
+ if (current > expmax) {
82
+ idx--;
83
+ continue;
84
+ }
85
+
86
+ *k = (firstpower + idx * steppowers);
87
+
88
+ return powers_ten[idx];
89
+ }
90
+ }
91
+
92
+ #define fracmask 0x000FFFFFFFFFFFFFU
93
+ #define expmask 0x7FF0000000000000U
94
+ #define hiddenbit 0x0010000000000000U
95
+ #define signmask 0x8000000000000000U
96
+ #define expbias (1023 + 52)
97
+
98
+ #define absv(n) ((n) < 0 ? -(n) : (n))
99
+ #define minv(a, b) ((a) < (b) ? (a) : (b))
100
+
101
+ static const uint64_t tens[] = {10000000000000000000U,
102
+ 1000000000000000000U,
103
+ 100000000000000000U,
104
+ 10000000000000000U,
105
+ 1000000000000000U,
106
+ 100000000000000U,
107
+ 10000000000000U,
108
+ 1000000000000U,
109
+ 100000000000U,
110
+ 10000000000U,
111
+ 1000000000U,
112
+ 100000000U,
113
+ 10000000U,
114
+ 1000000U,
115
+ 100000U,
116
+ 10000U,
117
+ 1000U,
118
+ 100U,
119
+ 10U,
120
+ 1U};
121
+
122
+ static uint64_t get_dbits(double d) {
123
+ union {
124
+ double dbl;
125
+ uint64_t i;
126
+ } dbl_bits = {d};
127
+
128
+ return dbl_bits.i;
129
+ }
130
+
131
+ static Fp build_fp(double d) {
132
+ uint64_t bits = get_dbits(d);
133
+
134
+ Fp fp;
135
+ fp.frac = bits & fracmask;
136
+ fp.exp = (bits & expmask) >> 52;
137
+
138
+ if (fp.exp) {
139
+ fp.frac += hiddenbit;
140
+ fp.exp -= expbias;
141
+
142
+ } else {
143
+ fp.exp = -expbias + 1;
144
+ }
145
+
146
+ return fp;
147
+ }
148
+
149
+ static void normalize(Fp* fp) {
150
+ while ((fp->frac & hiddenbit) == 0) {
151
+ fp->frac <<= 1;
152
+ fp->exp--;
153
+ }
154
+
155
+ int shift = 64 - 52 - 1;
156
+ fp->frac <<= shift;
157
+ fp->exp -= shift;
158
+ }
159
+
160
+ static void get_normalized_boundaries(Fp* fp, Fp* lower, Fp* upper) {
161
+ upper->frac = (fp->frac << 1) + 1;
162
+ upper->exp = fp->exp - 1;
163
+
164
+ while ((upper->frac & (hiddenbit << 1)) == 0) {
165
+ upper->frac <<= 1;
166
+ upper->exp--;
167
+ }
168
+
169
+ int u_shift = 64 - 52 - 2;
170
+
171
+ upper->frac <<= u_shift;
172
+ upper->exp = upper->exp - u_shift;
173
+
174
+ int l_shift = fp->frac == hiddenbit ? 2 : 1;
175
+
176
+ lower->frac = (fp->frac << l_shift) - 1;
177
+ lower->exp = fp->exp - l_shift;
178
+
179
+ lower->frac <<= lower->exp - upper->exp;
180
+ lower->exp = upper->exp;
181
+ }
182
+
183
+ static Fp multiply(Fp* a, Fp* b) {
184
+ const uint64_t lomask = 0x00000000FFFFFFFF;
185
+
186
+ uint64_t ah_bl = (a->frac >> 32) * (b->frac & lomask);
187
+ uint64_t al_bh = (a->frac & lomask) * (b->frac >> 32);
188
+ uint64_t al_bl = (a->frac & lomask) * (b->frac & lomask);
189
+ uint64_t ah_bh = (a->frac >> 32) * (b->frac >> 32);
190
+
191
+ uint64_t tmp = (ah_bl & lomask) + (al_bh & lomask) + (al_bl >> 32);
192
+ /* round up */
193
+ tmp += 1U << 31;
194
+
195
+ Fp fp = {ah_bh + (ah_bl >> 32) + (al_bh >> 32) + (tmp >> 32), a->exp + b->exp + 64};
196
+
197
+ return fp;
198
+ }
199
+
200
+ static void round_digit(char* digits, int ndigits, uint64_t delta, uint64_t rem, uint64_t kappa,
201
+ uint64_t frac) {
202
+ while (rem < frac && delta - rem >= kappa &&
203
+ (rem + kappa < frac || frac - rem > rem + kappa - frac)) {
204
+ digits[ndigits - 1]--;
205
+ rem += kappa;
206
+ }
207
+ }
208
+
209
+ static int generate_digits(Fp* fp, Fp* upper, Fp* lower, char* digits, int* K) {
210
+ uint64_t wfrac = upper->frac - fp->frac;
211
+ uint64_t delta = upper->frac - lower->frac;
212
+
213
+ Fp one;
214
+ one.frac = 1ULL << -upper->exp;
215
+ one.exp = upper->exp;
216
+
217
+ uint64_t part1 = upper->frac >> -one.exp;
218
+ uint64_t part2 = upper->frac & (one.frac - 1);
219
+
220
+ int idx = 0, kappa = 10;
221
+ const uint64_t* divp;
222
+ /* 1000000000 */
223
+ for (divp = tens + 10; kappa > 0; divp++) {
224
+ uint64_t div = *divp;
225
+ unsigned digit = (unsigned)(part1 / div);
226
+
227
+ if (digit || idx) {
228
+ digits[idx++] = (char)(digit + '0');
229
+ }
230
+
231
+ part1 -= digit * div;
232
+ kappa--;
233
+
234
+ uint64_t tmp = (part1 << -one.exp) + part2;
235
+ if (tmp <= delta) {
236
+ *K += kappa;
237
+ round_digit(digits, idx, delta, tmp, div << -one.exp, wfrac);
238
+
239
+ return idx;
240
+ }
241
+ }
242
+
243
+ /* 10 */
244
+ const uint64_t* unit = tens + 18;
245
+
246
+ while (true) {
247
+ part2 *= 10;
248
+ delta *= 10;
249
+ kappa--;
250
+
251
+ unsigned digit = (unsigned)(part2 >> -one.exp);
252
+ if (digit || idx) {
253
+ digits[idx++] = (char)(digit + '0');
254
+ }
255
+
256
+ part2 &= one.frac - 1;
257
+ if (part2 < delta) {
258
+ *K += kappa;
259
+ round_digit(digits, idx, delta, part2, one.frac, wfrac * *unit);
260
+
261
+ return idx;
262
+ }
263
+
264
+ unit--;
265
+ }
266
+ }
267
+
268
+ static int grisu2(double d, char* digits, int* K) {
269
+ Fp w = build_fp(d);
270
+
271
+ Fp lower, upper;
272
+ get_normalized_boundaries(&w, &lower, &upper);
273
+
274
+ normalize(&w);
275
+
276
+ int k;
277
+ Fp cp = find_cachedpow10(upper.exp, &k);
278
+
279
+ w = multiply(&w, &cp);
280
+ upper = multiply(&upper, &cp);
281
+ lower = multiply(&lower, &cp);
282
+
283
+ lower.frac++;
284
+ upper.frac--;
285
+
286
+ *K = -k;
287
+
288
+ return generate_digits(&w, &upper, &lower, digits, K);
289
+ }
290
+
291
+ static int emit_digits(char* digits, int ndigits, char* dest, int K, bool neg) {
292
+ int exp = absv(K + ndigits - 1);
293
+
294
+ if (K >= 0 && exp < 15) {
295
+ memcpy(dest, digits, (size_t)ndigits);
296
+ memset(dest + ndigits, '0', (size_t)K);
297
+
298
+ /* add a .0 to mark this as a float. */
299
+ dest[ndigits + K] = '.';
300
+ dest[ndigits + K + 1] = '0';
301
+
302
+ return ndigits + K + 2;
303
+ }
304
+
305
+ /* write decimal w/o scientific notation */
306
+ if (K < 0 && (K > -7 || exp < 10)) {
307
+ int offset = ndigits - absv(K);
308
+ /* fp < 1.0 -> write leading zero */
309
+ if (offset <= 0) {
310
+ offset = -offset;
311
+ dest[0] = '0';
312
+ dest[1] = '.';
313
+ memset(dest + 2, '0', (size_t)offset);
314
+ memcpy(dest + offset + 2, digits, (size_t)ndigits);
315
+
316
+ return ndigits + 2 + offset;
317
+
318
+ /* fp > 1.0 */
319
+ } else {
320
+ memcpy(dest, digits, (size_t)offset);
321
+ dest[offset] = '.';
322
+ memcpy(dest + offset + 1, digits + offset, (size_t)(ndigits - offset));
323
+
324
+ return ndigits + 1;
325
+ }
326
+ }
327
+
328
+ /* write decimal w/ scientific notation */
329
+ ndigits = minv(ndigits, 18 - neg);
330
+
331
+ int idx = 0;
332
+ dest[idx++] = digits[0];
333
+
334
+ if (ndigits > 1) {
335
+ dest[idx++] = '.';
336
+ memcpy(dest + idx, digits + 1, (size_t)(ndigits - 1));
337
+ idx += ndigits - 1;
338
+ }
339
+
340
+ dest[idx++] = 'e';
341
+
342
+ char sign = K + ndigits - 1 < 0 ? '-' : '+';
343
+ dest[idx++] = sign;
344
+
345
+ int cent = 0;
346
+
347
+ if (exp > 99) {
348
+ cent = exp / 100;
349
+ dest[idx++] = (char)(cent + '0');
350
+ exp -= cent * 100;
351
+ }
352
+ if (exp > 9) {
353
+ int dec = exp / 10;
354
+ dest[idx++] = (char)(dec + '0');
355
+ exp -= dec * 10;
356
+
357
+ } else if (cent) {
358
+ dest[idx++] = '0';
359
+ }
360
+
361
+ dest[idx++] = (char)(exp % 10 + '0');
362
+
363
+ return idx;
364
+ }
365
+
366
+ static int filter_special(double fp, char* dest) {
367
+ if (fp == 0.0) {
368
+ dest[0] = '0';
369
+ dest[1] = '.';
370
+ dest[2] = '0';
371
+ return 3;
372
+ }
373
+
374
+ uint64_t bits = get_dbits(fp);
375
+
376
+ bool nan = (bits & expmask) == expmask;
377
+
378
+ if (!nan) {
379
+ return 0;
380
+ }
381
+
382
+ if (bits & fracmask) {
383
+ dest[0] = 'n';
384
+ dest[1] = 'a';
385
+ dest[2] = 'n';
386
+
387
+ } else {
388
+ dest[0] = 'i';
389
+ dest[1] = 'n';
390
+ dest[2] = 'f';
391
+ }
392
+
393
+ return 3;
394
+ }
395
+
396
+ /* Fast and accurate double to string conversion based on Florian
397
+ * Loitsch's Grisu-algorithm. The generated string is never longer
398
+ * than 32 characters and is not null terminated. Returns the number
399
+ * of written characters. */
400
+ static int fpconv_dtoa(double d, char dest[32]) {
401
+ char digits[18];
402
+
403
+ int str_len = 0;
404
+ bool neg = false;
405
+
406
+ if (get_dbits(d) & signmask) {
407
+ dest[0] = '-';
408
+ str_len++;
409
+ neg = true;
410
+ }
411
+
412
+ int spec = filter_special(d, dest + str_len);
413
+
414
+ if (spec) {
415
+ return str_len + spec;
416
+ }
417
+
418
+ int K = 0;
419
+ int ndigits = grisu2(d, digits, &K);
420
+
421
+ str_len += emit_digits(digits, ndigits, dest + str_len, K, neg);
422
+
423
+ return str_len;
424
+ }
425
+
426
+ int yep_d2s_json(double d, char* buf) {
427
+ return fpconv_dtoa(d, buf);
428
+ }
@@ -377,7 +377,10 @@ static void emit_canonical_scalar(yep_writer* w, const yep_dnode* n) {
377
377
  tmp[tl] = '\0';
378
378
  double d = strtod(tmp, &end);
379
379
  if (end != NULL && *end == '\0' && end != tmp) {
380
- int wl = yep_d2s_shortest(d, buf);
380
+ /* JSON floats carry the json-gem byte shape so
381
+ * JSON.generate round trips stay byte-identical
382
+ * (yeptris#651); YAML keeps the shortest shape. */
383
+ int wl = w->json ? yep_d2s_json(d, buf) : yep_d2s_shortest(d, buf);
381
384
  wr_put(w, buf, (uint32_t)wl);
382
385
  return;
383
386
  }
@@ -656,6 +656,171 @@ static YeptrisStatus map_fail(int rc) {
656
656
  return rc == -1 ? YEPTRIS_ERROR_MEMORY : YEPTRIS_ERROR_UNSUPPORTED;
657
657
  }
658
658
 
659
+ /* ---- strict duplicate-key pre-scan (yeptris-ruby#278) ----------------
660
+ *
661
+ * json 3 makes duplicate keys an error by DEFAULT; the marshal route
662
+ * must raise before any bytes are written, exactly like the native
663
+ * materializer's rb_hash_aref probe. One linear pass over the same
664
+ * records the writer consumes — the writer itself stays untouched
665
+ * (the dup law lives here, once). The JSON DOM keeps both pairs
666
+ * ("{\"a\":1,\"a\":2}" map_count=2), so the records see them; the
667
+ * non-strict Marshal stream stays last-wins, which is json 2.x's
668
+ * contract.
669
+ *
670
+ * Key identity is the record's typed payload: span kinds compare the
671
+ * arena bytes, the rest compare (kind, b, payload). One growable
672
+ * open-addressing table serves the INNERMOST open map (cleared on
673
+ * close) — duplicates are per-map by definition.
674
+ */
675
+
676
+ typedef struct {
677
+ uint64_t p; /* 0 = empty slot (payload 0 with kind 0 cannot be a key) */
678
+ uint32_t off, len;
679
+ uint8_t kind, b;
680
+ } dup_slot;
681
+
682
+ typedef struct {
683
+ dup_slot* slots;
684
+ size_t cap; /* power of two */
685
+ } dup_set;
686
+
687
+ static uint64_t dup_hash(const dup_slot* k, const char* arena) {
688
+ uint64_t h = 1469598103934665603ull;
689
+ if (k->len != UINT32_MAX && (k->kind == YEP_V_STR || k->kind == YEP_V_TIMESTAMP ||
690
+ k->kind == YEP_V_ALIAS || k->kind == YEP_V_ANCHOR)) {
691
+ for (uint32_t i = 0; i < k->len; i++) {
692
+ h ^= (unsigned char)arena[k->off + i];
693
+ h *= 1099511628211ull;
694
+ }
695
+ }
696
+ h ^= k->kind;
697
+ h *= 1099511628211ull;
698
+ h ^= k->b;
699
+ h *= 1099511628211ull;
700
+ h ^= k->p;
701
+ h *= 1099511628211ull;
702
+ return h;
703
+ }
704
+
705
+ static int dup_eq(const dup_slot* a, const dup_slot* b, const char* arena) {
706
+ if (a->kind != b->kind || a->b != b->b || a->p != b->p) {
707
+ return 0;
708
+ }
709
+ int spanned = (a->kind == YEP_V_STR || a->kind == YEP_V_TIMESTAMP || a->kind == YEP_V_ALIAS ||
710
+ a->kind == YEP_V_ANCHOR);
711
+ if (!spanned) {
712
+ return 1;
713
+ }
714
+ if (a->len != b->len) {
715
+ return 0;
716
+ }
717
+ return a->len == 0 || memcmp(arena + a->off, arena + b->off, a->len) == 0;
718
+ }
719
+
720
+ static int dup_set_grow(dup_set* s) {
721
+ size_t ncap = s->cap ? s->cap * 2 : 16;
722
+ dup_slot* ns = calloc(ncap, sizeof(*ns));
723
+ if (ns == NULL) {
724
+ return -1;
725
+ }
726
+ if (s->slots != NULL) {
727
+ free(s->slots);
728
+ }
729
+ s->slots = ns;
730
+ s->cap = ncap;
731
+ return 0;
732
+ }
733
+
734
+ /* 1 = duplicate found (TLS error set), 0 = clean, -1 = oom */
735
+ static int strict_dup_scan(const yep_value_ctx* c) {
736
+ dup_set s = {NULL, 0};
737
+ int rc = 0;
738
+ const YeptrisValue* vals = c->vals;
739
+ for (size_t i = 0; i < c->n && rc == 0; i++) {
740
+ const YeptrisValue* v = &vals[i];
741
+ if (v->kind == YEP_V_MAP_OPEN) {
742
+ if (s.slots != NULL) {
743
+ memset(s.slots, 0, s.cap * sizeof(*s.slots)); /* innermost map only */
744
+ }
745
+ continue;
746
+ }
747
+ if (v->kind == YEP_V_CLOSE || v->is_key != 1) {
748
+ continue;
749
+ }
750
+ dup_slot k = {v->p, v->off, v->len, v->kind, v->b};
751
+ if (k.p == 0) {
752
+ k.p = 1; /* never aliases the empty-slot sentinel */
753
+ }
754
+ if (s.cap == 0 && dup_set_grow(&s) != 0) {
755
+ rc = -1;
756
+ break;
757
+ }
758
+ uint64_t h = dup_hash(&k, c->arena);
759
+ for (;;) {
760
+ dup_slot* slot = &s.slots[h & (s.cap - 1)];
761
+ if (slot->p == 0) {
762
+ *slot = k;
763
+ break;
764
+ }
765
+ if (dup_eq(slot, &k, c->arena)) {
766
+ yep_error_set(yep_error_tls(), YEP_ERR_UNEXPECTED, 0, 0, 0,
767
+ "duplicate key in object");
768
+ rc = 1;
769
+ break;
770
+ }
771
+ h++;
772
+ }
773
+ }
774
+ free(s.slots);
775
+ return rc;
776
+ }
777
+
778
+ static int dom_subtree_tagged(const yep_dom* d, uint32_t id);
779
+ static YeptrisStatus marshal_records(const yep_value_ctx* c, int compat, YeptrisMarshalMode mode,
780
+ int node_mode, char** out, size_t* out_len);
781
+
782
+ YEPTRIS_API YeptrisStatus yeptris_marshal_node_ex(YeptrisNode node, uint32_t flags, char** out,
783
+ size_t* out_len) {
784
+ if (node == NULL || out == NULL || out_len == NULL ||
785
+ (flags & ~YEPTRIS_MARSHAL_STRICT_DUP_KEYS) != 0) {
786
+ return YEPTRIS_ERROR_ARG;
787
+ }
788
+ *out = NULL;
789
+ *out_len = 0;
790
+ yeptris_node* h = (yeptris_node*)node;
791
+ yep_dom* dom = yep_doc_dom(h->doc); /* #378: a lazy parse carries the
792
+ * tape — the tree materializes here */
793
+ if (dom == NULL) {
794
+ return YEPTRIS_ERROR_MEMORY; /* the lazy route's retry contract */
795
+ }
796
+ if (dom_subtree_tagged(dom, h->id) != 0) {
797
+ yep_error_set(yep_error_tls(), YEP_ERR_UNEXPECTED, 0, 0, 0,
798
+ "marshal: explicitly tagged node not expressible; "
799
+ "fall back to the value walk");
800
+ return YEPTRIS_ERROR_UNSUPPORTED;
801
+ }
802
+ yep_value_ctx* c = NULL;
803
+ if (yep_values_from_dom(dom, h->id, 0, &c) != 0) {
804
+ return YEPTRIS_ERROR_MEMORY;
805
+ }
806
+ if ((flags & YEPTRIS_MARSHAL_STRICT_DUP_KEYS) != 0) {
807
+ int drc = strict_dup_scan(c);
808
+ if (drc != 0) {
809
+ yep_value_ctx_free(c);
810
+ return drc < 0 ? YEPTRIS_ERROR_MEMORY : YEPTRIS_ERROR_PARSE;
811
+ }
812
+ }
813
+ YeptrisStatus st = marshal_records(c, h->doc->schema == YEPTRIS_SCHEMA_11_COMPAT,
814
+ YEPTRIS_MARSHAL_ALL_DOCS, 1, out, out_len);
815
+ if (st == YEPTRIS_ERROR_UNSUPPORTED) {
816
+ yep_error_set(yep_error_tls(), YEP_ERR_UNEXPECTED, 0, 0, 0,
817
+ "marshal: merge key, timestamp or forward alias not expressible; "
818
+ "fall back to the value walk");
819
+ }
820
+ yep_value_ctx_free(c);
821
+ return st;
822
+ }
823
+
659
824
  static YeptrisStatus marshal_records(const yep_value_ctx* c, int compat, YeptrisMarshalMode mode,
660
825
  int node_mode, char** out, size_t* out_len) {
661
826
  *out = NULL;
@@ -750,36 +915,7 @@ static int dom_subtree_tagged(const yep_dom* d, uint32_t id) {
750
915
  }
751
916
 
752
917
  YEPTRIS_API YeptrisStatus yeptris_marshal_node(YeptrisNode node, char** out, size_t* out_len) {
753
- if (node == NULL || out == NULL || out_len == NULL) {
754
- return YEPTRIS_ERROR_ARG;
755
- }
756
- *out = NULL;
757
- *out_len = 0;
758
- yeptris_node* h = (yeptris_node*)node;
759
- yep_dom* dom = yep_doc_dom(h->doc); /* #378: a lazy parse carries the
760
- * tape — the tree materializes here */
761
- if (dom == NULL) {
762
- return YEPTRIS_ERROR_MEMORY; /* the lazy route's retry contract */
763
- }
764
- if (dom_subtree_tagged(dom, h->id) != 0) {
765
- yep_error_set(yep_error_tls(), YEP_ERR_UNEXPECTED, 0, 0, 0,
766
- "marshal: explicitly tagged node not expressible; "
767
- "fall back to the value walk");
768
- return YEPTRIS_ERROR_UNSUPPORTED;
769
- }
770
- yep_value_ctx* c = NULL;
771
- if (yep_values_from_dom(dom, h->id, 0, &c) != 0) {
772
- return YEPTRIS_ERROR_MEMORY;
773
- }
774
- YeptrisStatus st = marshal_records(c, h->doc->schema == YEPTRIS_SCHEMA_11_COMPAT,
775
- YEPTRIS_MARSHAL_ALL_DOCS, 1, out, out_len);
776
- if (st == YEPTRIS_ERROR_UNSUPPORTED) {
777
- yep_error_set(yep_error_tls(), YEP_ERR_UNEXPECTED, 0, 0, 0,
778
- "marshal: merge key, timestamp or forward alias not expressible; "
779
- "fall back to the value walk");
780
- }
781
- yep_value_ctx_free(c);
782
- return st;
918
+ return yeptris_marshal_node_ex(node, 0, out, out_len);
783
919
  }
784
920
 
785
921
  YEPTRIS_API void yeptris_marshal_free(char* out) {
@@ -3684,9 +3684,15 @@ static int e_parse_value(yep_engine* e, yep_ctx ctx, uint16_t floor_col) {
3684
3684
  continue;
3685
3685
  }
3686
3686
  int pc_comment = (pc == '#' && (prev == ' ' || prev == '\t'));
3687
+ /* an opening quote is node CONTENT, not a prop: the walk
3688
+ * must stop there ('only_props' stays 0) — walking on put
3689
+ * the line's ' #' inside the quoted span, read the quoted
3690
+ * key of the nightly's run 39223056891 ("x:\n!t \"a #b\": y")
3691
+ * as a comment and rejected the keyed pair */
3687
3692
  if (pc == '&' || pc == '!' || pc == '<' || pc == ',' || pc == '>' ||
3688
3693
  (pc != '\n' && pc != '\r' && !pc_comment &&
3689
- !yep_ct_is((unsigned char)pc, YEP_CT_FLOW_IND) && pc != ':' && pc != ' ')) {
3694
+ !yep_ct_is((unsigned char)pc, YEP_CT_FLOW_IND) && pc != ':' && pc != ' ' &&
3695
+ pc != '"' && pc != '\'')) {
3690
3696
  prev = pc;
3691
3697
  t++;
3692
3698
  continue;
metadata CHANGED
@@ -1,14 +1,14 @@
1
1
  --- !ruby/object:Gem::Specification
2
2
  name: yeptris
3
3
  version: !ruby/object:Gem::Version
4
- version: 0.6.33.2
4
+ version: 0.6.34.2
5
5
  platform: aarch64-linux
6
6
  authors:
7
7
  - Ribose Inc.
8
8
  autorequire:
9
9
  bindir: bin
10
10
  cert_chain: []
11
- date: 2026-10-20 00:00:00.000000000 Z
11
+ date: 2026-10-21 00:00:00.000000000 Z
12
12
  dependencies:
13
13
  - !ruby/object:Gem::Dependency
14
14
  name: ffi
@@ -126,6 +126,7 @@ files:
126
126
  - vendor/libyeptris/src/yeptris/emit/float/api.h
127
127
  - vendor/libyeptris/src/yeptris/emit/float/dragon.c
128
128
  - vendor/libyeptris/src/yeptris/emit/float/floatint.h
129
+ - vendor/libyeptris/src/yeptris/emit/float/fpconv.c
129
130
  - vendor/libyeptris/src/yeptris/emit/float/print.c
130
131
  - vendor/libyeptris/src/yeptris/emit/style.c
131
132
  - vendor/libyeptris/src/yeptris/emit/style.h