ReDoS: revert new superlinear algorithm. #13127
Conversation
|
This affect Java as well, right? |
Oh, yes, that's right. Edit: The Java tests are fine, I'll request review from the team. |
|
Was CI not run for Python? |
Because no file in the |
I see.
Cool
I do want that :-) |
That ought to be fixed for all languages. Otherwise main will be broken by accident very soon. |
Every language has some implementation in
I know how to do it for the CI workflows defined in Edit: Looks like internal DX is on it, see below. |
|
@erik-krogh is anything blocking us from merging this? The above discussion about CI checks doesn't seem like it should block this particular PR. |
No. I'll merge. |
We just traded one performance tradeoff with another.
I've gotten too many reports of polynomial-redos timing out recently.
So I've gone back to the ugly approach of limiting the search space for "complex" regular expressions.
But I'm keeping the
succ->pumpEndrename suggestion from #12628.Evaluations: Python, Ruby (<- includes the regression that motivated this PR), JavaScript all look good.
We flag slightly different results, which is expected, we go back to the resutls we had before #12628.