| 1 |
# These test the UTF-8 modes expose by regex-automata. Namely, when utf8 is |
| 2 |
# true, then we promise that the haystack is valid UTF-8. (Otherwise behavior |
| 3 |
# is unspecified.) This also corresponds to building the regex engine with the |
| 4 |
# following two guarantees: |
| 5 |
# |
| 6 |
# 1) For any non-empty match reported, its span is guaranteed to correspond to |
| 7 |
# valid UTF-8. |
| 8 |
# 2) All empty or zero-width matches reported must never split a UTF-8 |
| 9 |
# encoded codepoint. If the haystack has invalid UTF-8, then this results in |
| 10 |
# unspecified behavior. |
| 11 |
# |
| 12 |
# The (2) is in particular what we focus our testing on since (1) is generally |
| 13 |
# guaranteed by regex-syntax's AST-to-HIR translator and is well tested there. |
| 14 |
# The thing with (2) is that it can't be described in the HIR, so the regex |
| 15 |
# engines have to handle that case. Thus, we test it here. |
| 16 |
# |
| 17 |
# Note that it is possible to build a regex that has property (1) but not |
| 18 |
# (2), and vice versa. This is done by building the HIR with 'utf8=true' but |
| 19 |
# building the Thompson NFA with 'utf8=false'. We don't test that here because |
| 20 |
# the harness doesn't expose a way to enable or disable UTF-8 mode with that |
| 21 |
# granularity. Instead, those combinations are lightly tested via doc examples. |
| 22 |
# That's not to say that (1) without (2) is uncommon. Indeed, ripgrep uses it |
| 23 |
# because it cannot guarantee that its haystack is valid UTF-8. |
| 24 |
|
| 25 |
# This tests that an empty regex doesn't split a codepoint. |
| 26 |
[[test]] |
| 27 |
name = "empty-utf8yes" |
| 28 |
regex = '' |
| 29 |
haystack = '☃' |
| 30 |
matches = [[0, 0], [3, 3]] |
| 31 |
unicode = true |
| 32 |
utf8 = true |
| 33 |
|
| 34 |
# Tests the overlapping case of the above. |
| 35 |
[[test]] |
| 36 |
name = "empty-utf8yes-overlapping" |
| 37 |
regex = '' |
| 38 |
haystack = '☃' |
| 39 |
matches = [[0, 0], [3, 3]] |
| 40 |
unicode = true |
| 41 |
utf8 = true |
| 42 |
match-kind = "all" |
| 43 |
search-kind = "overlapping" |
| 44 |
|
| 45 |
# This tests that an empty regex DOES split a codepoint when utf=false. |
| 46 |
[[test]] |
| 47 |
name = "empty-utf8no" |
| 48 |
regex = '' |
| 49 |
haystack = '☃' |
| 50 |
matches = [[0, 0], [1, 1], [2, 2], [3, 3]] |
| 51 |
unicode = true |
| 52 |
utf8 = false |
| 53 |
|
| 54 |
# Tests the overlapping case of the above. |
| 55 |
[[test]] |
| 56 |
name = "empty-utf8no-overlapping" |
| 57 |
regex = '' |
| 58 |
haystack = '☃' |
| 59 |
matches = [[0, 0], [1, 1], [2, 2], [3, 3]] |
| 60 |
unicode = true |
| 61 |
utf8 = false |
| 62 |
match-kind = "all" |
| 63 |
search-kind = "overlapping" |
| 64 |
|
| 65 |
# This tests that an empty regex doesn't split a codepoint, even if we give |
| 66 |
# it bounds entirely within the codepoint. |
| 67 |
# |
| 68 |
# This is one of the trickier cases and is what motivated the current UTF-8 |
| 69 |
# mode design. In particular, at one point, this test failed the 'is_match' |
| 70 |
# variant of the test but not 'find'. This is because the 'is_match' code path |
| 71 |
# is specifically optimized for "was a match found" rather than "where is the |
| 72 |
# match." In the former case, you don't really care about the empty-vs-non-empty |
| 73 |
# matches, and thus, the codepoint splitting filtering logic wasn't getting |
| 74 |
# applied. (In multiple ways across multiple regex engines.) In this way, you |
| 75 |
# can wind up with a situation where 'is_match' says "yes," but 'find' says, |
| 76 |
# "I didn't find anything." Which is... not great. |
| 77 |
# |
| 78 |
# I could have decided to say that providing boundaries that themselves split |
| 79 |
# a codepoint would have unspecified behavior. But I couldn't quite convince |
| 80 |
# myself that such boundaries were the only way to get an inconsistency between |
| 81 |
# 'is_match' and 'find'. |
| 82 |
# |
| 83 |
# Note that I also tried to come up with a test like this that fails without |
| 84 |
# using `bounds`. Specifically, a test where 'is_match' and 'find' disagree. |
| 85 |
# But I couldn't do it, and I'm tempted to conclude it is impossible. The |
| 86 |
# fundamental problem is that you need to simultaneously produce an empty match |
| 87 |
# that splits a codepoint while *not* matching before or after the codepoint. |
| 88 |
[[test]] |
| 89 |
name = "empty-utf8yes-bounds" |
| 90 |
regex = '' |
| 91 |
haystack = '𝛃' |
| 92 |
bounds = [1, 3] |
| 93 |
matches = [] |
| 94 |
unicode = true |
| 95 |
utf8 = true |
| 96 |
|
| 97 |
# Tests the overlapping case of the above. |
| 98 |
[[test]] |
| 99 |
name = "empty-utf8yes-bounds-overlapping" |
| 100 |
regex = '' |
| 101 |
haystack = '𝛃' |
| 102 |
bounds = [1, 3] |
| 103 |
matches = [] |
| 104 |
unicode = true |
| 105 |
utf8 = true |
| 106 |
match-kind = "all" |
| 107 |
search-kind = "overlapping" |
| 108 |
|
| 109 |
# This tests that an empty regex splits a codepoint when the bounds are |
| 110 |
# entirely within the codepoint. |
| 111 |
[[test]] |
| 112 |
name = "empty-utf8no-bounds" |
| 113 |
regex = '' |
| 114 |
haystack = '𝛃' |
| 115 |
bounds = [1, 3] |
| 116 |
matches = [[1, 1], [2, 2], [3, 3]] |
| 117 |
unicode = true |
| 118 |
utf8 = false |
| 119 |
|
| 120 |
# Tests the overlapping case of the above. |
| 121 |
[[test]] |
| 122 |
name = "empty-utf8no-bounds-overlapping" |
| 123 |
regex = '' |
| 124 |
haystack = '𝛃' |
| 125 |
bounds = [1, 3] |
| 126 |
matches = [[1, 1], [2, 2], [3, 3]] |
| 127 |
unicode = true |
| 128 |
utf8 = false |
| 129 |
match-kind = "all" |
| 130 |
search-kind = "overlapping" |
| 131 |
|
| 132 |
# In this test, we anchor the search. Since the start position is also a UTF-8 |
| 133 |
# boundary, we get a match. |
| 134 |
[[test]] |
| 135 |
name = "empty-utf8yes-anchored" |
| 136 |
regex = '' |
| 137 |
haystack = '𝛃' |
| 138 |
matches = [[0, 0]] |
| 139 |
anchored = true |
| 140 |
unicode = true |
| 141 |
utf8 = true |
| 142 |
|
| 143 |
# Tests the overlapping case of the above. |
| 144 |
[[test]] |
| 145 |
name = "empty-utf8yes-anchored-overlapping" |
| 146 |
regex = '' |
| 147 |
haystack = '𝛃' |
| 148 |
matches = [[0, 0]] |
| 149 |
anchored = true |
| 150 |
unicode = true |
| 151 |
utf8 = true |
| 152 |
match-kind = "all" |
| 153 |
search-kind = "overlapping" |
| 154 |
|
| 155 |
# Same as above, except with UTF-8 mode disabled. It almost doesn't change the |
| 156 |
# result, except for the fact that since this is an anchored search and we |
| 157 |
# always find all matches, the test harness will keep reporting matches until |
| 158 |
# none are found. Because it's anchored, matches will be reported so long as |
| 159 |
# they are directly adjacent. Since with UTF-8 mode the next anchored search |
| 160 |
# after the match at [0, 0] fails, iteration stops (and doesn't find the last |
| 161 |
# match at [4, 4]). |
| 162 |
[[test]] |
| 163 |
name = "empty-utf8no-anchored" |
| 164 |
regex = '' |
| 165 |
haystack = '𝛃' |
| 166 |
matches = [[0, 0], [1, 1], [2, 2], [3, 3], [4, 4]] |
| 167 |
anchored = true |
| 168 |
unicode = true |
| 169 |
utf8 = false |
| 170 |
|
| 171 |
# Tests the overlapping case of the above. |
| 172 |
# |
| 173 |
# Note that overlapping anchored searches are a little weird, and it's not |
| 174 |
# totally clear what their semantics ought to be. For now, we just test the |
| 175 |
# current behavior of our test shim that implements overlapping search. (This |
| 176 |
# is one of the reasons why we don't really expose regex-level overlapping |
| 177 |
# searches.) |
| 178 |
[[test]] |
| 179 |
name = "empty-utf8no-anchored-overlapping" |
| 180 |
regex = '' |
| 181 |
haystack = '𝛃' |
| 182 |
matches = [[0, 0]] |
| 183 |
anchored = true |
| 184 |
unicode = true |
| 185 |
utf8 = false |
| 186 |
match-kind = "all" |
| 187 |
search-kind = "overlapping" |
| 188 |
|
| 189 |
# In this test, we anchor the search, but also set bounds. The bounds start the |
| 190 |
# search in the middle of a codepoint, so there should never be a match. |
| 191 |
[[test]] |
| 192 |
name = "empty-utf8yes-anchored-bounds" |
| 193 |
regex = '' |
| 194 |
haystack = '𝛃' |
| 195 |
matches = [] |
| 196 |
bounds = [1, 3] |
| 197 |
anchored = true |
| 198 |
unicode = true |
| 199 |
utf8 = true |
| 200 |
|
| 201 |
# Tests the overlapping case of the above. |
| 202 |
[[test]] |
| 203 |
name = "empty-utf8yes-anchored-bounds-overlapping" |
| 204 |
regex = '' |
| 205 |
haystack = '𝛃' |
| 206 |
matches = [] |
| 207 |
bounds = [1, 3] |
| 208 |
anchored = true |
| 209 |
unicode = true |
| 210 |
utf8 = true |
| 211 |
match-kind = "all" |
| 212 |
search-kind = "overlapping" |
| 213 |
|
| 214 |
# Same as above, except with UTF-8 mode disabled. Without UTF-8 mode enabled, |
| 215 |
# matching within a codepoint is allowed. And remember, as in the anchored test |
| 216 |
# above with UTF-8 mode disabled, iteration will report all adjacent matches. |
| 217 |
# The matches at [0, 0] and [4, 4] are not included because of the bounds of |
| 218 |
# the search. |
| 219 |
[[test]] |
| 220 |
name = "empty-utf8no-anchored-bounds" |
| 221 |
regex = '' |
| 222 |
haystack = '𝛃' |
| 223 |
bounds = [1, 3] |
| 224 |
matches = [[1, 1], [2, 2], [3, 3]] |
| 225 |
anchored = true |
| 226 |
unicode = true |
| 227 |
utf8 = false |
| 228 |
|
| 229 |
# Tests the overlapping case of the above. |
| 230 |
# |
| 231 |
# Note that overlapping anchored searches are a little weird, and it's not |
| 232 |
# totally clear what their semantics ought to be. For now, we just test the |
| 233 |
# current behavior of our test shim that implements overlapping search. (This |
| 234 |
# is one of the reasons why we don't really expose regex-level overlapping |
| 235 |
# searches.) |
| 236 |
[[test]] |
| 237 |
name = "empty-utf8no-anchored-bounds-overlapping" |
| 238 |
regex = '' |
| 239 |
haystack = '𝛃' |
| 240 |
bounds = [1, 3] |
| 241 |
matches = [[1, 1]] |
| 242 |
anchored = true |
| 243 |
unicode = true |
| 244 |
utf8 = false |
| 245 |
match-kind = "all" |
| 246 |
search-kind = "overlapping" |
| 247 |
|
| 248 |
# This tests that we find the match at the end of the string when the bounds |
| 249 |
# exclude the first match. |
| 250 |
[[test]] |
| 251 |
name = "empty-utf8yes-startbound" |
| 252 |
regex = '' |
| 253 |
haystack = '𝛃' |
| 254 |
bounds = [1, 4] |
| 255 |
matches = [[4, 4]] |
| 256 |
unicode = true |
| 257 |
utf8 = true |
| 258 |
|
| 259 |
# Tests the overlapping case of the above. |
| 260 |
[[test]] |
| 261 |
name = "empty-utf8yes-startbound-overlapping" |
| 262 |
regex = '' |
| 263 |
haystack = '𝛃' |
| 264 |
bounds = [1, 4] |
| 265 |
matches = [[4, 4]] |
| 266 |
unicode = true |
| 267 |
utf8 = true |
| 268 |
match-kind = "all" |
| 269 |
search-kind = "overlapping" |
| 270 |
|
| 271 |
# Same as above, except since UTF-8 mode is disabled, we also find the matches |
| 272 |
# inbetween that split the codepoint. |
| 273 |
[[test]] |
| 274 |
name = "empty-utf8no-startbound" |
| 275 |
regex = '' |
| 276 |
haystack = '𝛃' |
| 277 |
bounds = [1, 4] |
| 278 |
matches = [[1, 1], [2, 2], [3, 3], [4, 4]] |
| 279 |
unicode = true |
| 280 |
utf8 = false |
| 281 |
|
| 282 |
# Tests the overlapping case of the above. |
| 283 |
[[test]] |
| 284 |
name = "empty-utf8no-startbound-overlapping" |
| 285 |
regex = '' |
| 286 |
haystack = '𝛃' |
| 287 |
bounds = [1, 4] |
| 288 |
matches = [[1, 1], [2, 2], [3, 3], [4, 4]] |
| 289 |
unicode = true |
| 290 |
utf8 = false |
| 291 |
match-kind = "all" |
| 292 |
search-kind = "overlapping" |
| 293 |
|
| 294 |
# This tests that we don't find any matches in an anchored search, even when |
| 295 |
# the bounds include a match (at the end). |
| 296 |
[[test]] |
| 297 |
name = "empty-utf8yes-anchored-startbound" |
| 298 |
regex = '' |
| 299 |
haystack = '𝛃' |
| 300 |
bounds = [1, 4] |
| 301 |
matches = [] |
| 302 |
anchored = true |
| 303 |
unicode = true |
| 304 |
utf8 = true |
| 305 |
|
| 306 |
# Tests the overlapping case of the above. |
| 307 |
[[test]] |
| 308 |
name = "empty-utf8yes-anchored-startbound-overlapping" |
| 309 |
regex = '' |
| 310 |
haystack = '𝛃' |
| 311 |
bounds = [1, 4] |
| 312 |
matches = [] |
| 313 |
anchored = true |
| 314 |
unicode = true |
| 315 |
utf8 = true |
| 316 |
match-kind = "all" |
| 317 |
search-kind = "overlapping" |
| 318 |
|
| 319 |
# Same as above, except since UTF-8 mode is disabled, we also find the matches |
| 320 |
# inbetween that split the codepoint. Even though this is an anchored search, |
| 321 |
# since the matches are adjacent, we find all of them. |
| 322 |
[[test]] |
| 323 |
name = "empty-utf8no-anchored-startbound" |
| 324 |
regex = '' |
| 325 |
haystack = '𝛃' |
| 326 |
bounds = [1, 4] |
| 327 |
matches = [[1, 1], [2, 2], [3, 3], [4, 4]] |
| 328 |
anchored = true |
| 329 |
unicode = true |
| 330 |
utf8 = false |
| 331 |
|
| 332 |
# Tests the overlapping case of the above. |
| 333 |
# |
| 334 |
# Note that overlapping anchored searches are a little weird, and it's not |
| 335 |
# totally clear what their semantics ought to be. For now, we just test the |
| 336 |
# current behavior of our test shim that implements overlapping search. (This |
| 337 |
# is one of the reasons why we don't really expose regex-level overlapping |
| 338 |
# searches.) |
| 339 |
[[test]] |
| 340 |
name = "empty-utf8no-anchored-startbound-overlapping" |
| 341 |
regex = '' |
| 342 |
haystack = '𝛃' |
| 343 |
bounds = [1, 4] |
| 344 |
matches = [[1, 1]] |
| 345 |
anchored = true |
| 346 |
unicode = true |
| 347 |
utf8 = false |
| 348 |
match-kind = "all" |
| 349 |
search-kind = "overlapping" |
| 350 |
|
| 351 |
# This tests that we find the match at the end of the haystack in UTF-8 mode |
| 352 |
# when our bounds only include the empty string at the end of the haystack. |
| 353 |
[[test]] |
| 354 |
name = "empty-utf8yes-anchored-endbound" |
| 355 |
regex = '' |
| 356 |
haystack = '𝛃' |
| 357 |
bounds = [4, 4] |
| 358 |
matches = [[4, 4]] |
| 359 |
anchored = true |
| 360 |
unicode = true |
| 361 |
utf8 = true |
| 362 |
|
| 363 |
# Tests the overlapping case of the above. |
| 364 |
[[test]] |
| 365 |
name = "empty-utf8yes-anchored-endbound-overlapping" |
| 366 |
regex = '' |
| 367 |
haystack = '𝛃' |
| 368 |
bounds = [4, 4] |
| 369 |
matches = [[4, 4]] |
| 370 |
anchored = true |
| 371 |
unicode = true |
| 372 |
utf8 = true |
| 373 |
match-kind = "all" |
| 374 |
search-kind = "overlapping" |
| 375 |
|
| 376 |
# Same as above, but with UTF-8 mode disabled. Results remain the same since |
| 377 |
# the only possible match does not split a codepoint. |
| 378 |
[[test]] |
| 379 |
name = "empty-utf8no-anchored-endbound" |
| 380 |
regex = '' |
| 381 |
haystack = '𝛃' |
| 382 |
bounds = [4, 4] |
| 383 |
matches = [[4, 4]] |
| 384 |
anchored = true |
| 385 |
unicode = true |
| 386 |
utf8 = false |
| 387 |
|
| 388 |
# Tests the overlapping case of the above. |
| 389 |
[[test]] |
| 390 |
name = "empty-utf8no-anchored-endbound-overlapping" |
| 391 |
regex = '' |
| 392 |
haystack = '𝛃' |
| 393 |
bounds = [4, 4] |
| 394 |
matches = [[4, 4]] |
| 395 |
anchored = true |
| 396 |
unicode = true |
| 397 |
utf8 = false |
| 398 |
match-kind = "all" |
| 399 |
search-kind = "overlapping" |