| 1 | /*! |
| 2 | A module dedicated to plucking inner literals out of a regex pattern, and |
| 3 | then constructing a prefilter for them. We also include a regex pattern |
| 4 | "prefix" that corresponds to the bits of the regex that need to match before |
| 5 | the literals do. The reverse inner optimization then proceeds by looking for |
| 6 | matches of the inner literal(s), and then doing a reverse search of the prefix |
| 7 | from the start of the literal match to find the overall start position of the |
| 8 | match. |
| 9 | |
| 10 | The essential invariant we want to uphold here is that the literals we return |
| 11 | reflect a set where *at least* one of them must match in order for the overall |
| 12 | regex to match. We also need to maintain the invariant that the regex prefix |
| 13 | returned corresponds to the entirety of the regex up until the literals we |
| 14 | return. |
| 15 | |
| 16 | This somewhat limits what we can do. That is, if we a regex like |
| 17 | `\w+(@!|%%)\w+`, then we can pluck the `{@!, %%}` out and build a prefilter |
| 18 | from it. Then we just need to compile `\w+` in reverse. No fuss no muss. But if |
| 19 | we have a regex like \d+@!|\w+%%`, then we get kind of stymied. Technically, |
| 20 | we could still extract `{@!, %%}`, and it is true that at least of them must |
| 21 | match. But then, what is our regex prefix? Again, in theory, that could be |
| 22 | `\d+|\w+`, but that's not quite right, because the `\d+` only matches when `@!` |
| 23 | matches, and `\w+` only matches when `%%` matches. |
| 24 | |
| 25 | All of that is technically possible to do, but it seemingly requires a lot of |
| 26 | sophistication and machinery. Probably the way to tackle that is with some kind |
| 27 | of formalism and approach this problem more generally. |
| 28 | |
| 29 | For now, the code below basically just looks for a top-level concatenation. |
| 30 | And if it can find one, it looks for literals in each of the direct child |
| 31 | sub-expressions of that concatenation. If some good ones are found, we return |
| 32 | those and a concatenation of the Hir expressions seen up to that point. |
| 33 | */ |
| 34 | |
| 35 | use alloc::vec::Vec; |
| 36 | |
| 37 | use regex_syntax::hir::{self, literal, Hir, HirKind}; |
| 38 | |
| 39 | use crate::{util::prefilter::Prefilter, MatchKind}; |
| 40 | |
| 41 | /// Attempts to extract an "inner" prefilter from the given HIR expressions. If |
| 42 | /// one was found, then a concatenation of the HIR expressions that precede it |
| 43 | /// is returned. |
| 44 | /// |
| 45 | /// The idea here is that the prefilter returned can be used to find candidate |
| 46 | /// matches. And then the HIR returned can be used to build a reverse regex |
| 47 | /// matcher, which will find the start of the candidate match. Finally, the |
| 48 | /// match still has to be confirmed with a normal anchored forward scan to find |
| 49 | /// the end position of the match. |
| 50 | /// |
| 51 | /// Note that this assumes leftmost-first match semantics, so callers must |
| 52 | /// not call this otherwise. |
| 53 | pub(crate) fn extract(hirs: &[&Hir]) -> Option<(Hir, Prefilter)> { |
| 54 | if hirs.len() != 1 { |
| 55 | debug!( |
| 56 | "skipping reverse inner optimization since it only \ |
| 57 | supports 1 pattern, {} were given" , |
| 58 | hirs.len(), |
| 59 | ); |
| 60 | return None; |
| 61 | } |
| 62 | let mut concat = match top_concat(hirs[0]) { |
| 63 | Some(concat) => concat, |
| 64 | None => { |
| 65 | debug!( |
| 66 | "skipping reverse inner optimization because a top-level \ |
| 67 | concatenation could not found" , |
| 68 | ); |
| 69 | return None; |
| 70 | } |
| 71 | }; |
| 72 | // We skip the first HIR because if it did have a prefix prefilter in it, |
| 73 | // we probably wouldn't be here looking for an inner prefilter. |
| 74 | for i in 1..concat.len() { |
| 75 | let hir = &concat[i]; |
| 76 | let pre = match prefilter(hir) { |
| 77 | None => continue, |
| 78 | Some(pre) => pre, |
| 79 | }; |
| 80 | // Even if we got a prefilter, if it isn't consider "fast," then we |
| 81 | // probably don't want to bother with it. Namely, since the reverse |
| 82 | // inner optimization requires some overhead, it likely only makes |
| 83 | // sense if the prefilter scan itself is (believed) to be much faster |
| 84 | // than the regex engine. |
| 85 | if !pre.is_fast() { |
| 86 | debug!( |
| 87 | "skipping extracted inner prefilter because \ |
| 88 | it probably isn't fast" |
| 89 | ); |
| 90 | continue; |
| 91 | } |
| 92 | let concat_suffix = Hir::concat(concat.split_off(i)); |
| 93 | let concat_prefix = Hir::concat(concat); |
| 94 | // Look for a prefilter again. Why? Because above we only looked for |
| 95 | // a prefilter on the individual 'hir', but we might be able to find |
| 96 | // something better and more discriminatory by looking at the entire |
| 97 | // suffix. We don't do this above to avoid making this loop worst case |
| 98 | // quadratic in the length of 'concat'. |
| 99 | let pre2 = match prefilter(&concat_suffix) { |
| 100 | None => pre, |
| 101 | Some(pre2) => { |
| 102 | if pre2.is_fast() { |
| 103 | pre2 |
| 104 | } else { |
| 105 | pre |
| 106 | } |
| 107 | } |
| 108 | }; |
| 109 | return Some((concat_prefix, pre2)); |
| 110 | } |
| 111 | debug!( |
| 112 | "skipping reverse inner optimization because a top-level \ |
| 113 | sub-expression with a fast prefilter could not be found" |
| 114 | ); |
| 115 | None |
| 116 | } |
| 117 | |
| 118 | /// Attempt to extract a prefilter from an HIR expression. |
| 119 | /// |
| 120 | /// We do a little massaging here to do our best that the prefilter we get out |
| 121 | /// of this is *probably* fast. Basically, the false positive rate has a much |
| 122 | /// higher impact for things like the reverse inner optimization because more |
| 123 | /// work needs to potentially be done for each candidate match. |
| 124 | /// |
| 125 | /// Note that this assumes leftmost-first match semantics, so callers must |
| 126 | /// not call this otherwise. |
| 127 | fn prefilter(hir: &Hir) -> Option<Prefilter> { |
| 128 | let mut extractor = literal::Extractor::new(); |
| 129 | extractor.kind(literal::ExtractKind::Prefix); |
| 130 | let mut prefixes = extractor.extract(hir); |
| 131 | debug!( |
| 132 | "inner prefixes (len={:?}) extracted before optimization: {:?}" , |
| 133 | prefixes.len(), |
| 134 | prefixes |
| 135 | ); |
| 136 | // Since these are inner literals, we know they cannot be exact. But the |
| 137 | // extractor doesn't know this. We mark them as inexact because this might |
| 138 | // impact literal optimization. Namely, optimization weights "all literals |
| 139 | // are exact" as very high, because it presumes that any match results in |
| 140 | // an overall match. But of course, that is not the case here. |
| 141 | // |
| 142 | // In practice, this avoids plucking out a ASCII-only \s as an alternation |
| 143 | // of single-byte whitespace characters. |
| 144 | prefixes.make_inexact(); |
| 145 | prefixes.optimize_for_prefix_by_preference(); |
| 146 | debug!( |
| 147 | "inner prefixes (len={:?}) extracted after optimization: {:?}" , |
| 148 | prefixes.len(), |
| 149 | prefixes |
| 150 | ); |
| 151 | prefixes |
| 152 | .literals() |
| 153 | .and_then(|lits| Prefilter::new(MatchKind::LeftmostFirst, lits)) |
| 154 | } |
| 155 | |
| 156 | /// Looks for a "top level" HirKind::Concat item in the given HIR. This will |
| 157 | /// try to return one even if it's embedded in a capturing group, but is |
| 158 | /// otherwise pretty conservative in what is returned. |
| 159 | /// |
| 160 | /// The HIR returned is a complete copy of the concat with all capturing |
| 161 | /// groups removed. In effect, the concat returned is "flattened" with respect |
| 162 | /// to capturing groups. This makes the detection logic above for prefixes |
| 163 | /// a bit simpler, and it works because 1) capturing groups never influence |
| 164 | /// whether a match occurs or not and 2) capturing groups are not used when |
| 165 | /// doing the reverse inner search to find the start of the match. |
| 166 | fn top_concat(mut hir: &Hir) -> Option<Vec<Hir>> { |
| 167 | loop { |
| 168 | hir = match hir.kind() { |
| 169 | HirKind::Empty |
| 170 | | HirKind::Literal(_) |
| 171 | | HirKind::Class(_) |
| 172 | | HirKind::Look(_) |
| 173 | | HirKind::Repetition(_) |
| 174 | | HirKind::Alternation(_) => return None, |
| 175 | HirKind::Capture(hir::Capture { ref sub, .. }) => sub, |
| 176 | HirKind::Concat(ref subs) => { |
| 177 | // We are careful to only do the flattening/copy when we know |
| 178 | // we have a "top level" concat we can inspect. This avoids |
| 179 | // doing extra work in cases where we definitely won't use it. |
| 180 | // (This might still be wasted work if we can't go on to find |
| 181 | // some literals to extract.) |
| 182 | let concat = |
| 183 | Hir::concat(subs.iter().map(|h| flatten(h)).collect()); |
| 184 | return match concat.into_kind() { |
| 185 | HirKind::Concat(xs) => Some(xs), |
| 186 | // It is actually possible for this case to occur, because |
| 187 | // 'Hir::concat' might simplify the expression to the point |
| 188 | // that concatenations are actually removed. One wonders |
| 189 | // whether this leads to other cases where we should be |
| 190 | // extracting literals, but in theory, I believe if we do |
| 191 | // get here, then it means that a "real" prefilter failed |
| 192 | // to be extracted and we should probably leave well enough |
| 193 | // alone. (A "real" prefilter is unbothered by "top-level |
| 194 | // concats" and "capturing groups.") |
| 195 | _ => return None, |
| 196 | }; |
| 197 | } |
| 198 | }; |
| 199 | } |
| 200 | } |
| 201 | |
| 202 | /// Returns a copy of the given HIR but with all capturing groups removed. |
| 203 | fn flatten(hir: &Hir) -> Hir { |
| 204 | match hir.kind() { |
| 205 | HirKind::Empty => Hir::empty(), |
| 206 | HirKind::Literal(hir::Literal(ref x: &Box<[u8]>)) => Hir::literal(lit:x.clone()), |
| 207 | HirKind::Class(ref x: &Class) => Hir::class(x.clone()), |
| 208 | HirKind::Look(ref x: &Look) => Hir::look(x.clone()), |
| 209 | HirKind::Repetition(ref x: &Repetition) => Hir::repetition(rep:x.with(sub:flatten(&x.sub))), |
| 210 | // This is the interesting case. We just drop the group information |
| 211 | // entirely and use the child HIR itself. |
| 212 | HirKind::Capture(hir::Capture { ref sub: &Box, .. }) => flatten(hir:sub), |
| 213 | HirKind::Alternation(ref xs: &Vec) => { |
| 214 | Hir::alternation(subs:xs.iter().map(|x: &Hir| flatten(hir:x)).collect()) |
| 215 | } |
| 216 | HirKind::Concat(ref xs: &Vec) => { |
| 217 | Hir::concat(subs:xs.iter().map(|x: &Hir| flatten(hir:x)).collect()) |
| 218 | } |
| 219 | } |
| 220 | } |
| 221 | |