automata: fix incorrect offsets reported by reverse inner optimization #1063
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sadly it seems that my days of squashing optimization bugs are still
before me. In this particular case, the reverse inner literal
optimization (which is a new optimization introduced in regex 1.9)
resulted in reporting incorrect match offsets in some cases. The
offending case here is:
$ regex-cli find match meta --no-table -p '(?:(\d+)[:.])?(\d{1,2})[:.](\d{2})' -y '888:77:66'
0:1:9:888:77:66
The above reports a match at 1..9, but the correct match is 0..9. The
problem here is that the reverse inner literal optimization is being
applied, which splits the regex into three (conceptual) pieces:
1. `(?:(\d+)[:.])?(\d{1,2})`
2. `[:.]`
3. `(\d{2})`
The reverse inner optimization works by looking for occurrences of (2)
first, then matching (1) in reverse to find the start position of the
match and then searching for (3) in the forward direction to find the
end of the match.
The problem in this particular case is that (2) matches at position `3`
in the `888:77:66` haystack. Since the first section of numbers is
optional, the reverse inner optimization believes a match exists at
offset `1` by virtue of matching (1) in reverse. That is, the
`(\d{1,2})` matches at 1..3 while the `(?:(\d+)[:.])?` doesn't match at
all. The reverse search here is correct in isolation, but it leads to an
overall incorrect result by stopping the search early. The issue is that
the true leftmost match requires (2) to match at 6..7, but since it
matched at 3..4 first, it is considered first and leads to an incorrect
overall match.
To fix this, we add another "trip wire" to the reverse inner
optimization (of which there are already several) that tries to detect
cases where it cannot prove that the match it found is actually the
leftmost match. Namely, if it reports a match offset greater than the
start of the search and otherwise *could* have kept searching, then we
don't know whether we have the true leftmost match. In that case, we
bail on the optimization and let a slower path take over.
This is yet another example of how the nature of regex searching, and in
particular leftmost searching, inhibits the composition of different
regex strategies. Or at least, makes them incredibly subtle.
Fixes #1060
|
I ran this under rebar's curated benchmark suite, and there were no obvious regressions: |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Sadly it seems that my days of squashing optimization bugs are still before me. In this particular case, the reverse inner literal optimization (which is a new optimization introduced in regex 1.9) resulted in reporting incorrect match offsets in some cases. The offending case here is:
The above reports a match at 1..9, but the correct match is 0..9. The problem here is that the reverse inner literal optimization is being applied, which splits the regex into three (conceptual) pieces:
(?:(\d+)[:.])?(\d{1,2})[:.](\d{2})The reverse inner optimization works by looking for occurrences of (2) first, then matching (1) in reverse to find the start position of the match and then searching for (3) in the forward direction to find the end of the match.
The problem in this particular case is that (2) matches at position
3in the888:77:66haystack. Since the first section of numbers is optional, the reverse inner optimization believes a match exists at offset1by virtue of matching (1) in reverse. That is, the(\d{1,2})matches at 1..3 while the(?:(\d+)[:.])?doesn't match at all. The reverse search here is correct in isolation, but it leads to an overall incorrect result by stopping the search early. The issue is that the true leftmost match requires (2) to match at 6..7, but since it matched at 3..4 first, it is considered first and leads to an incorrect overall match.To fix this, we add another "trip wire" to the reverse inner optimization (of which there are already several) that tries to detect cases where it cannot prove that the match it found is actually the leftmost match. Namely, if it reports a match offset greater than the start of the search and otherwise could have kept searching, then we don't know whether we have the true leftmost match. In that case, we bail on the optimization and let a slower path take over.
This is yet another example of how the nature of regex searching, and in particular leftmost searching, inhibits the composition of different regex strategies. Or at least, makes them incredibly subtle.
Fixes #1060