internal: Bump salsa

[lnicola]

CC #14693 (comment)

[lnicola]

@Veykril can you give this a try (your old commit probably works too, though)?

[bors]

☔ The latest upstream changes (presumably #14664) made this pull request unmergeable. Please resolve the merge conflicts.

[Veykril]

Checked it out on my machine, the perf decrease is also in the 10% for me so I guess its fine (maybe analysis overall gains from this elsewhere if we are lucky)

[lnicola]

Oof, okay, I hoped it would be slightly less :-(. Do you think we should merge this?

[Veykril]

dot completion went from 56 to 63, path from 145 to 161 (though my machine is also a bit noisy)

I think its fine to merge (as part of bumping everything to syn 2), if it worsens more things like analysis stats we might want to look into not having the slot changes

[lnicola]

It does show up in analysis-stats:

# baseline, 7d48bbadd449db111289d12db15d3a0dde1a9b14, best of 3
Database loaded:     506.62ms, 284minstr (metadata 300.22ms, 25minstr; build 102.00ms, 9815kinstr)
  crates: 44, mods: 918, decls: 19627, fns: 14681
Item Collection:     11.24s, 100ginstr
  exprs: 462280, ??ty: 33 (0%), ?ty: 129 (0%), !ty: 1                                                                                                                
  pats: 104232, ??ty: 4 (0%), ?ty: 11 (0%), !ty: 193
Inference:           38.46s, 280ginstr
Total:               49.70s, 381ginstr

# pr, best of 3
Database loaded:     509.21ms, 284minstr (metadata 301.47ms, 25minstr; build 105.61ms, 9815kinstr)
  crates: 44, mods: 918, decls: 19627, fns: 14681
Item Collection:     11.25s, 99ginstr
  exprs: 462280, ??ty: 33 (0%), ?ty: 129 (0%), !ty: 1                                                                                                                
  pats: 104232, ??ty: 4 (0%), ?ty: 11 (0%), !ty: 193
Inference:           41.70s, 296ginstr
Total:               52.96s, 395ginstr

[Veykril]

hmm, not sure what to think about 3 seconds here to be honest 😅

[lnicola]

I think it might be a bit less than that because of noise. Another couple of runs:

# baseline
Total:               48.82s, 383ginstr
Total:               50.38s, 385ginstr
Total:               50.79s, 400ginstr

# pr
Total:               49.01s, 377ginstr
Total:               53.74s, 406ginstr
Total:               52.95s, 410ginstr
Total:               50.36s, 385ginstr

The instruction counts are all over the place, too.

[Veykril]

The instruction counts are all over the place, too.

That sounds terrible 😬. But okay, then let's just try it with the commits.

[bors]

☔ The latest upstream changes (presumably #14802) made this pull request unmergeable. Please resolve the merge conflicts.

[lnicola]

This would still be nice to have, but we'll have to fork or get someone to publish a new version before.