Skip memmove in del_at_beg when not needed.

[yuyichao]

This was likely previously done on linux in the glibc implementation.

Fix #24494

[dhoegh]

Is it possible to add some kind of test, to protect against a possible regression in the future, as the regression is massive and apparently snug under the radar in 0.5 and 0.6?

[yuyichao]

No there's no behavioral difference. You can add something to benchmark though.

[yuyichao]

(Though nonosoldier won't be able to catch this either......)

[dhoegh]

Too bad:( Could the test case in the issue be added as a separate test set, and hence if it regresses the test set run time goes from 0.5s to 300s?

[yuyichao]

Historically that kind of test always cause more trouble and never catch any regression. There's nothing special about this pattern anyway.

[StefanKarpinski]

Testing for this kind of thing is exactly why we have BaseBenchmarks – and we do regularly run those tests to make sure things haven't regressed. Please do open a PR to add this test there.

[ararslan]

This doesn't backport cleanly to release-0.6 (or at least to my backports branch). Would you be willing to submit a PR to release-0.6 with the correct change, or just push a commit to aa/backports-0.6.2?

[yuyichao]

It's exactly the same code on 0.6, just with different context. Anyway, it's here https://github.com/JuliaLang/julia/compare/aa/backports-0.6.2...yyc/test/release-0.6?expand=1

[ararslan]

Perfect, thanks so much!