Multi match with non const patterns #38485
Merged
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
robot-ch-test-poll1
added
the
pr-improvement
rschu1ze
force-pushed
the
multi-match-with-non_const-patterns
branch
from
2e6c2fe
to
82442b4
Compare
|
@Mergifyio update |
✅ Branch has been successfully updated |
rschu1ze
force-pushed
the
multi-match-with-non_const-patterns
branch
2 times, most recently
from
17fc728
to
c40a9fb
Compare
#38434 (comment) #38434 (comment) #38434 (comment) (the latter is no longer relevant as the affected places were removed in the meantime)
After making function multi[Fuzzy]Match(Any|AnyIndex|AllIndices)() work
with non-const needles, 12 more functions started to fail in test
"00233_position_function_family":
multiSearchAny()
multiSearchAnyCaseInsensitive()
multiSearchAnyUTF8
multiSearchAnyCaseInsensitiveUTF8()
multiSearchFirstPosition()
multiSearchFirstPositionCaseInsensitive()
multiSearchFirstPositionUTF8()
multiSearchFirstPositionCaseInsensitiveUTF8()
multiSearchFirstIndex()
multiSearchFirstIndexCaseInsensitive()
multiSearchFirstIndexUTF8()
multiSearchFirstIndexCaseInsensitiveUTF8()
Failing queries take the form
select 0 = multiSearchAny('\0', CAST([], 'Array(String)'));
rschu1ze
force-pushed
the
multi-match-with-non_const-patterns
branch
from
c40a9fb
to
1eed72b
Compare
Difficult to test because x86 and ARM behave differently due to vectorscan not available everyehwere ... The same stuff (multiFuzzyMatch) is tested elsewhere.
|
Dockerhub died before integration/stateless tests ran. Will need to fix some tests anyways, hopefully Dockerhub is up the next time. |
|
@Mergifyio update |
✅ Branch has been successfully updated |
|
The stress test errors are related to keeper and unrelated. |
Avogar
reviewed
Jul 6, 2022
Avogar
reviewed
Jul 6, 2022
Avogar
reviewed
Jul 6, 2022
Avogar
reviewed
Jul 6, 2022
Avogar
reviewed
Jul 6, 2022
|
I decided to experiment a bit and found logical error. It wasn't introduced in this PR, the same error occurs in master. @rschu1ze As you are already familiar with this code, maybe it will be easy to fix it for you. But let's do it in a separate PR with bugfix changelog category |
|
What about functions |
Generally speaking, searching with non-const needles is more expensive than with const needles as the searcher needs to be re-initialized once per row (which is slow for fast Volnitsky search and fast for slow std lib search). So, non-const needles are nice-to-have from a functional POV but not recommended. This PR is about three functions for which a user wanted non-const needles. While doing so, I had to add support for non-const needles to more functions. multiSearchAllPositions() wasn't one of these. Looking at it's implementation, it should not be too difficult to also allow non-const needles but that would go into a separate PR then. |
rschu1ze
commented
Jul 6, 2022
| } | ||
| #else | ||
| /// fallback if vectorscan is not compiled | ||
| /// -- the code is copypasted from vectorVector() in MatchImpl.h and quite complex code ... all of it can be removed once vectorscan is |
There was a problem hiding this comment.
This branch is called in a quite obscure edge case ... for functions mutliMatchAny() / multiMatchAnyIndex() with non-const needles on platforms where vectorscan is disabled (all non-x86 at the moment). It is a bit fragile and I had to write it from scratch using some copypasta from elsewhere because the technique used in the fallback for const needles didn't work conceptually.
I did some local testing to make sure it's okay. However, if problems in CI tests pop up, I would likely remove it altogether and throw a "not implemented" exception instead. The same is done for function multiMatchAllIndices() which is sort of related. There is a good chance that vectorscan can be enabled on non-x86 so we could throw the fallback away anyways.
Interesting, thanks for finding this. Will take a look. |
rschu1ze
force-pushed
the
multi-match-with-non_const-patterns
branch
from
192debd
to
1de5e9a
Compare
rschu1ze
commented
Jul 7, 2022
|
All tests are green except the ubsan stress test ("Backward compatibility check: Server failed to start"). But the failure is is caused by network issues and keeper. Other stress tests ran fine. Guess we are good to merge. |
Avogar
approved these changes
Jul 7, 2022
This was referenced Jul 8, 2022
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.
Resolves #38046
Functions multiMatchAny(), multiMatchAnyIndex(), multiMatchAllIndices() and their fuzzy variants now accept non-const pattern array argument
While working on this PR, I noticed that functions
broke and needed the same treatment (non-const needles). So I fixed that too which eventually blew this PR a bit up.
Changelog category (leave one):
Changelog entry (a user-readable short description of the changes that goes to CHANGELOG.md):
Functions multiMatchAny(), multiMatchAnyIndex(), multiMatchAllIndices() and their fuzzy variants now accept non-const pattern array argument