-
Notifications
You must be signed in to change notification settings - Fork 1.7k
Proper handling $crate and local_inner_macros #7133
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Conversation
6a0c387
to
fe5340d
Compare
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Great!
bors r+ |
Build succeeded! And happy new year! 🎉 |
This has substantially slowed down the analysis: Confirmed locally:
|
Maybe we could store |
Unfortunately I think we should consider reverting this until we find a solution. |
The original rustc implementation used TLS to store the hygiene information and I did try it and it improves speed a lot. But I don't feel it is the right solution for us. I opened https://rust-lang.zulipchat.com/#narrow/stream/185405-t-compiler.2Fwg-rls-2.2E0/topic/Hygiene.20Frames.20Information for the discussion. |
7137: Revert "Proper handling $crate and local_inner_macros" r=jonas-schievink a=jonas-schievink Reverts #7133 It caused a fairly significant performance regression. bors r+ Co-authored-by: Jonas Schievink <[email protected]>
7145: Proper handling $crate Take 2 [DO NOT MERGE] r=edwin0cheng a=edwin0cheng Similar to previous PR (#7133) , but improved the following things : 1. Instead of storing the whole `ExpansionInfo`, we store a similar but stripped version `HygieneInfo`. 2. Instread of storing the `SyntaxNode` (because every token we are interested are IDENT), we store the `TextRange` only. 3. Because of 2, we now can put it in Salsa. 4. And most important improvement: Instead of computing the whole frames every single time, we compute it recursively through salsa: (Such that in the best scenario, we only need to compute the first layer of frame) ```rust let def_site = db.hygiene_frame(info.def.file_id); let call_site = db.hygiene_frame(info.arg.file_id); HygieneFrame { expansion: Some(info), local_inner, krate, call_site, def_site } ``` The overall speed compared to previous PR is much faster (65s vs 45s) : ``` [WITH old PR] Database loaded 644.86ms, 284mi Crates in this dir: 36 Total modules found: 576 Total declarations: 11153 Total functions: 8715 Item Collection: 15.78s, 91562mi Total expressions: 240721 Expressions of unknown type: 2635 (1%) Expressions of partially unknown type: 2064 (0%) Type mismatches: 865 Inference: 49.84s, 250747mi Total: 65.62s, 342310mi rust-analyzer -q analysis-stats . 66.72s user 0.57s system 99% cpu 1:07.40 total [WITH this PR] Database loaded 665.83ms, 284mi Crates in this dir: 36 Total modules found: 577 Total declarations: 11188 Total functions: 8743 Item Collection: 15.28s, 84919mi Total expressions: 241229 Expressions of unknown type: 2637 (1%) Expressions of partially unknown type: 2064 (0%) Type mismatches: 868 Inference: 30.15s, 135293mi Total: 45.43s, 220213mi rust-analyzer -q analysis-stats . 46.26s user 0.74s system 99% cpu 47.294 total ``` *HOWEVER*, it is still a perf regression (35s vs 45s): ``` [WITHOUT this PR] Database loaded 657.42ms, 284mi Crates in this dir: 36 Total modules found: 577 Total declarations: 11177 Total functions: 8735 Item Collection: 12.87s, 72407mi Total expressions: 239380 Expressions of unknown type: 2643 (1%) Expressions of partially unknown type: 2064 (0%) Type mismatches: 868 Inference: 22.88s, 97889mi Total: 35.74s, 170297mi rust-analyzer -q analysis-stats . 36.71s user 0.63s system 99% cpu 37.498 total ``` Co-authored-by: Edwin Cheng <[email protected]>
This PR introduces
HygineFrames
to store the macro definition/call site hierarchy in hyginee and when resolvinglocal_inner_macros
and$crate
, we use the token to look up the corresponding frame and return the correct value.See also: https://rustc-dev-guide.rust-lang.org/macro-expansion.html#hygiene-and-hierarchies
fixe #6890 and #6788
r? @jonas-schievink