| 1 |
use { |
| 2 |
anyhow::Result, |
| 3 |
regex_automata::{ |
| 4 |
hybrid::{ |
| 5 |
dfa::{OverlappingState, DFA}, |
| 6 |
regex::{self, Regex}, |
| 7 |
}, |
| 8 |
nfa::thompson, |
| 9 |
util::{prefilter::Prefilter, syntax}, |
| 10 |
Anchored, Input, PatternSet, |
| 11 |
}, |
| 12 |
regex_test::{ |
| 13 |
CompiledRegex, Match, RegexTest, SearchKind, Span, TestResult, |
| 14 |
TestRunner, |
| 15 |
}, |
| 16 |
}; |
| 17 |
|
| 18 |
use crate::{create_input, suite, untestify_kind}; |
| 19 |
|
| 20 |
const EXPANSIONS: &[&str] = &["is_match", "find", "which"]; |
| 21 |
|
| 22 |
/// Tests the default configuration of the hybrid NFA/DFA. |
| 23 |
#[test] |
| 24 |
fn default() -> Result<()> { |
| 25 |
let builder = Regex::builder(); |
| 26 |
TestRunner::new()? |
| 27 |
.expand(EXPANSIONS, |t| t.compiles()) |
| 28 |
// Without NFA shrinking, this test blows the default cache capacity. |
| 29 |
.blacklist("expensive/regression-many-repeat-no-stack-overflow") |
| 30 |
.test_iter(suite()?.iter(), compiler(builder)) |
| 31 |
.assert(); |
| 32 |
Ok(()) |
| 33 |
} |
| 34 |
|
| 35 |
/// Tests the hybrid NFA/DFA with prefilters enabled. |
| 36 |
#[test] |
| 37 |
fn prefilter() -> Result<()> { |
| 38 |
let my_compiler = |test: &RegexTest, regexes: &[String]| { |
| 39 |
// Parse regexes as HIRs so we can get literals to build a prefilter. |
| 40 |
let mut hirs = vec![]; |
| 41 |
for pattern in regexes.iter() { |
| 42 |
hirs.push(syntax::parse_with(pattern, &config_syntax(test))?); |
| 43 |
} |
| 44 |
let kind = match untestify_kind(test.match_kind()) { |
| 45 |
None => return Ok(CompiledRegex::skip()), |
| 46 |
Some(kind) => kind, |
| 47 |
}; |
| 48 |
let pre = Prefilter::from_hirs_prefix(kind, &hirs); |
| 49 |
let mut builder = Regex::builder(); |
| 50 |
builder.dfa(DFA::config().prefilter(pre)); |
| 51 |
compiler(builder)(test, regexes) |
| 52 |
}; |
| 53 |
TestRunner::new()? |
| 54 |
.expand(EXPANSIONS, |t| t.compiles()) |
| 55 |
// Without NFA shrinking, this test blows the default cache capacity. |
| 56 |
.blacklist("expensive/regression-many-repeat-no-stack-overflow") |
| 57 |
.test_iter(suite()?.iter(), my_compiler) |
| 58 |
.assert(); |
| 59 |
Ok(()) |
| 60 |
} |
| 61 |
|
| 62 |
/// Tests the hybrid NFA/DFA with NFA shrinking enabled. |
| 63 |
/// |
| 64 |
/// This is *usually* not the configuration one wants for a lazy DFA. NFA |
| 65 |
/// shrinking is mostly only advantageous when building a full DFA since it |
| 66 |
/// can sharply decrease the amount of time determinization takes. But NFA |
| 67 |
/// shrinking is itself otherwise fairly expensive currently. Since a lazy DFA |
| 68 |
/// has no compilation time (other than for building the NFA of course) before |
| 69 |
/// executing a search, it's usually worth it to forgo NFA shrinking. |
| 70 |
/// |
| 71 |
/// Nevertheless, we test to make sure everything is OK with NFA shrinking. As |
| 72 |
/// a bonus, there are some tests we don't need to skip because they now fit in |
| 73 |
/// the default cache capacity. |
| 74 |
#[test] |
| 75 |
fn nfa_shrink() -> Result<()> { |
| 76 |
let mut builder = Regex::builder(); |
| 77 |
builder.thompson(thompson::Config::new().shrink(true)); |
| 78 |
TestRunner::new()? |
| 79 |
.expand(EXPANSIONS, |t| t.compiles()) |
| 80 |
.test_iter(suite()?.iter(), compiler(builder)) |
| 81 |
.assert(); |
| 82 |
Ok(()) |
| 83 |
} |
| 84 |
|
| 85 |
/// Tests the hybrid NFA/DFA when 'starts_for_each_pattern' is enabled for all |
| 86 |
/// tests. |
| 87 |
#[test] |
| 88 |
fn starts_for_each_pattern() -> Result<()> { |
| 89 |
let mut builder = Regex::builder(); |
| 90 |
builder.dfa(DFA::config().starts_for_each_pattern(true)); |
| 91 |
TestRunner::new()? |
| 92 |
.expand(EXPANSIONS, |t| t.compiles()) |
| 93 |
// Without NFA shrinking, this test blows the default cache capacity. |
| 94 |
.blacklist("expensive/regression-many-repeat-no-stack-overflow") |
| 95 |
.test_iter(suite()?.iter(), compiler(builder)) |
| 96 |
.assert(); |
| 97 |
Ok(()) |
| 98 |
} |
| 99 |
|
| 100 |
/// Tests the hybrid NFA/DFA when 'specialize_start_states' is enabled. |
| 101 |
#[test] |
| 102 |
fn specialize_start_states() -> Result<()> { |
| 103 |
let mut builder = Regex::builder(); |
| 104 |
builder.dfa(DFA::config().specialize_start_states(true)); |
| 105 |
TestRunner::new()? |
| 106 |
.expand(EXPANSIONS, |t| t.compiles()) |
| 107 |
// Without NFA shrinking, this test blows the default cache capacity. |
| 108 |
.blacklist("expensive/regression-many-repeat-no-stack-overflow") |
| 109 |
.test_iter(suite()?.iter(), compiler(builder)) |
| 110 |
.assert(); |
| 111 |
Ok(()) |
| 112 |
} |
| 113 |
|
| 114 |
/// Tests the hybrid NFA/DFA when byte classes are disabled. |
| 115 |
/// |
| 116 |
/// N.B. Disabling byte classes doesn't avoid any indirection at search time. |
| 117 |
/// All it does is cause every byte value to be its own distinct equivalence |
| 118 |
/// class. |
| 119 |
#[test] |
| 120 |
fn no_byte_classes() -> Result<()> { |
| 121 |
let mut builder = Regex::builder(); |
| 122 |
builder.dfa(DFA::config().byte_classes(false)); |
| 123 |
TestRunner::new()? |
| 124 |
.expand(EXPANSIONS, |t| t.compiles()) |
| 125 |
// Without NFA shrinking, this test blows the default cache capacity. |
| 126 |
.blacklist("expensive/regression-many-repeat-no-stack-overflow") |
| 127 |
.test_iter(suite()?.iter(), compiler(builder)) |
| 128 |
.assert(); |
| 129 |
Ok(()) |
| 130 |
} |
| 131 |
|
| 132 |
/// Tests that hybrid NFA/DFA never clears its cache for any test with the |
| 133 |
/// default capacity. |
| 134 |
/// |
| 135 |
/// N.B. If a regex suite test is added that causes the cache to be cleared, |
| 136 |
/// then this should just skip that test. (Which can be done by calling the |
| 137 |
/// 'blacklist' method on 'TestRunner'.) |
| 138 |
#[test] |
| 139 |
fn no_cache_clearing() -> Result<()> { |
| 140 |
let mut builder = Regex::builder(); |
| 141 |
builder.dfa(DFA::config().minimum_cache_clear_count(Some(0))); |
| 142 |
TestRunner::new()? |
| 143 |
.expand(EXPANSIONS, |t| t.compiles()) |
| 144 |
// Without NFA shrinking, this test blows the default cache capacity. |
| 145 |
.blacklist("expensive/regression-many-repeat-no-stack-overflow") |
| 146 |
.test_iter(suite()?.iter(), compiler(builder)) |
| 147 |
.assert(); |
| 148 |
Ok(()) |
| 149 |
} |
| 150 |
|
| 151 |
/// Tests the hybrid NFA/DFA when the minimum cache capacity is set. |
| 152 |
#[test] |
| 153 |
fn min_cache_capacity() -> Result<()> { |
| 154 |
let mut builder = Regex::builder(); |
| 155 |
builder |
| 156 |
.dfa(DFA::config().cache_capacity(0).skip_cache_capacity_check(true)); |
| 157 |
TestRunner::new()? |
| 158 |
.expand(EXPANSIONS, |t| t.compiles()) |
| 159 |
.test_iter(suite()?.iter(), compiler(builder)) |
| 160 |
.assert(); |
| 161 |
Ok(()) |
| 162 |
} |
| 163 |
|
| 164 |
fn compiler( |
| 165 |
mut builder: regex::Builder, |
| 166 |
) -> impl FnMut(&RegexTest, &[String]) -> Result<CompiledRegex> { |
| 167 |
move |test, regexes| { |
| 168 |
// Parse regexes as HIRs for some analysis below. |
| 169 |
let mut hirs = vec![]; |
| 170 |
for pattern in regexes.iter() { |
| 171 |
hirs.push(syntax::parse_with(pattern, &config_syntax(test))?); |
| 172 |
} |
| 173 |
|
| 174 |
// Check if our regex contains things that aren't supported by DFAs. |
| 175 |
// That is, Unicode word boundaries when searching non-ASCII text. |
| 176 |
if !test.haystack().is_ascii() { |
| 177 |
for hir in hirs.iter() { |
| 178 |
if hir.properties().look_set().contains_word_unicode() { |
| 179 |
return Ok(CompiledRegex::skip()); |
| 180 |
} |
| 181 |
} |
| 182 |
} |
| 183 |
if !configure_regex_builder(test, &mut builder) { |
| 184 |
return Ok(CompiledRegex::skip()); |
| 185 |
} |
| 186 |
let re = builder.build_many(®exes)?; |
| 187 |
let mut cache = re.create_cache(); |
| 188 |
Ok(CompiledRegex::compiled(move |test| -> TestResult { |
| 189 |
run_test(&re, &mut cache, test) |
| 190 |
})) |
| 191 |
} |
| 192 |
} |
| 193 |
|
| 194 |
fn run_test( |
| 195 |
re: &Regex, |
| 196 |
cache: &mut regex::Cache, |
| 197 |
test: &RegexTest, |
| 198 |
) -> TestResult { |
| 199 |
let input = create_input(test); |
| 200 |
match test.additional_name() { |
| 201 |
"is_match" => { |
| 202 |
TestResult::matched(re.is_match(cache, input.earliest(true))) |
| 203 |
} |
| 204 |
"find" => match test.search_kind() { |
| 205 |
SearchKind::Earliest | SearchKind::Leftmost => { |
| 206 |
let input = |
| 207 |
input.earliest(test.search_kind() == SearchKind::Earliest); |
| 208 |
TestResult::matches( |
| 209 |
re.find_iter(cache, input) |
| 210 |
.take(test.match_limit().unwrap_or(std::usize::MAX)) |
| 211 |
.map(|m| Match { |
| 212 |
id: m.pattern().as_usize(), |
| 213 |
span: Span { start: m.start(), end: m.end() }, |
| 214 |
}), |
| 215 |
) |
| 216 |
} |
| 217 |
SearchKind::Overlapping => { |
| 218 |
try_search_overlapping(re, cache, &input).unwrap() |
| 219 |
} |
| 220 |
}, |
| 221 |
"which" => match test.search_kind() { |
| 222 |
SearchKind::Earliest | SearchKind::Leftmost => { |
| 223 |
// There are no "which" APIs for standard searches. |
| 224 |
TestResult::skip() |
| 225 |
} |
| 226 |
SearchKind::Overlapping => { |
| 227 |
let dfa = re.forward(); |
| 228 |
let cache = cache.as_parts_mut().0; |
| 229 |
let mut patset = PatternSet::new(dfa.pattern_len()); |
| 230 |
dfa.try_which_overlapping_matches(cache, &input, &mut patset) |
| 231 |
.unwrap(); |
| 232 |
TestResult::which(patset.iter().map(|p| p.as_usize())) |
| 233 |
} |
| 234 |
}, |
| 235 |
name => TestResult::fail(&format!("unrecognized test name: {}", name)), |
| 236 |
} |
| 237 |
} |
| 238 |
|
| 239 |
/// Configures the given regex builder with all relevant settings on the given |
| 240 |
/// regex test. |
| 241 |
/// |
| 242 |
/// If the regex test has a setting that is unsupported, then this returns |
| 243 |
/// false (implying the test should be skipped). |
| 244 |
fn configure_regex_builder( |
| 245 |
test: &RegexTest, |
| 246 |
builder: &mut regex::Builder, |
| 247 |
) -> bool { |
| 248 |
let match_kind = match untestify_kind(test.match_kind()) { |
| 249 |
None => return false, |
| 250 |
Some(k) => k, |
| 251 |
}; |
| 252 |
|
| 253 |
let mut dfa_config = |
| 254 |
DFA::config().match_kind(match_kind).unicode_word_boundary(true); |
| 255 |
// When doing an overlapping search, we might try to find the start of each |
| 256 |
// match with a custom search routine. In that case, we need to tell the |
| 257 |
// reverse search (for the start offset) which pattern to look for. The |
| 258 |
// only way that API works is when anchored starting states are compiled |
| 259 |
// for each pattern. This does technically also enable it for the forward |
| 260 |
// DFA, but we're okay with that. |
| 261 |
if test.search_kind() == SearchKind::Overlapping { |
| 262 |
dfa_config = dfa_config.starts_for_each_pattern(true); |
| 263 |
} |
| 264 |
builder |
| 265 |
.syntax(config_syntax(test)) |
| 266 |
.thompson(config_thompson(test)) |
| 267 |
.dfa(dfa_config); |
| 268 |
true |
| 269 |
} |
| 270 |
|
| 271 |
/// Configuration of a Thompson NFA compiler from a regex test. |
| 272 |
fn config_thompson(test: &RegexTest) -> thompson::Config { |
| 273 |
let mut lookm = regex_automata::util::look::LookMatcher::new(); |
| 274 |
lookm.set_line_terminator(test.line_terminator()); |
| 275 |
thompson::Config::new().utf8(test.utf8()).look_matcher(lookm) |
| 276 |
} |
| 277 |
|
| 278 |
/// Configuration of the regex parser from a regex test. |
| 279 |
fn config_syntax(test: &RegexTest) -> syntax::Config { |
| 280 |
syntax::Config::new() |
| 281 |
.case_insensitive(test.case_insensitive()) |
| 282 |
.unicode(test.unicode()) |
| 283 |
.utf8(test.utf8()) |
| 284 |
.line_terminator(test.line_terminator()) |
| 285 |
} |
| 286 |
|
| 287 |
/// Execute an overlapping search, and for each match found, also find its |
| 288 |
/// overlapping starting positions. |
| 289 |
/// |
| 290 |
/// N.B. This routine used to be part of the crate API, but 1) it wasn't clear |
| 291 |
/// to me how useful it was and 2) it wasn't clear to me what its semantics |
| 292 |
/// should be. In particular, a potentially surprising footgun of this routine |
| 293 |
/// that it is worst case *quadratic* in the size of the haystack. Namely, it's |
| 294 |
/// possible to report a match at every position, and for every such position, |
| 295 |
/// scan all the way to the beginning of the haystack to find the starting |
| 296 |
/// position. Typical leftmost non-overlapping searches don't suffer from this |
| 297 |
/// because, well, matches can't overlap. So subsequent searches after a match |
| 298 |
/// is found don't revisit previously scanned parts of the haystack. |
| 299 |
/// |
| 300 |
/// Its semantics can be strange for other reasons too. For example, given |
| 301 |
/// the regex '.*' and the haystack 'zz', the full set of overlapping matches |
| 302 |
/// is: [0, 0], [1, 1], [0, 1], [2, 2], [1, 2], [0, 2]. The ordering of |
| 303 |
/// those matches is quite strange, but makes sense when you think about the |
| 304 |
/// implementation: an end offset is found left-to-right, and then one or more |
| 305 |
/// starting offsets are found right-to-left. |
| 306 |
/// |
| 307 |
/// Nevertheless, we provide this routine in our test suite because it's |
| 308 |
/// useful to test the low level DFA overlapping search and our test suite |
| 309 |
/// is written in a way that requires starting offsets. |
| 310 |
fn try_search_overlapping( |
| 311 |
re: &Regex, |
| 312 |
cache: &mut regex::Cache, |
| 313 |
input: &Input<'_>, |
| 314 |
) -> Result<TestResult> { |
| 315 |
let mut matches = vec![]; |
| 316 |
let mut fwd_state = OverlappingState::start(); |
| 317 |
let (fwd_dfa, rev_dfa) = (re.forward(), re.reverse()); |
| 318 |
let (fwd_cache, rev_cache) = cache.as_parts_mut(); |
| 319 |
while let Some(end) = { |
| 320 |
fwd_dfa.try_search_overlapping_fwd( |
| 321 |
fwd_cache, |
| 322 |
input, |
| 323 |
&mut fwd_state, |
| 324 |
)?; |
| 325 |
fwd_state.get_match() |
| 326 |
} { |
| 327 |
let revsearch = input |
| 328 |
.clone() |
| 329 |
.range(input.start()..end.offset()) |
| 330 |
.anchored(Anchored::Pattern(end.pattern())) |
| 331 |
.earliest(false); |
| 332 |
let mut rev_state = OverlappingState::start(); |
| 333 |
while let Some(start) = { |
| 334 |
rev_dfa.try_search_overlapping_rev( |
| 335 |
rev_cache, |
| 336 |
&revsearch, |
| 337 |
&mut rev_state, |
| 338 |
)?; |
| 339 |
rev_state.get_match() |
| 340 |
} { |
| 341 |
let span = Span { start: start.offset(), end: end.offset() }; |
| 342 |
let mat = Match { id: end.pattern().as_usize(), span }; |
| 343 |
matches.push(mat); |
| 344 |
} |
| 345 |
} |
| 346 |
Ok(TestResult::matches(matches)) |
| 347 |
} |