Specifically, #18744 was the PR that was supposed to fix the old bug, but it fixed it incorrectly (and didn't add a test!) The underlying reason was that we marked metavariables in expansions as joint if they were joint in the macro call, which is incorrect.
This wrong fix causes other bug, #19497, which this PR fixes by removing the old (incorrect) fix.
I expected this to be faster (due to less allocations and better cache locality), but benchmarked it is not (neither it is slower). Memory usage, however, drops by ~50mb (of `analysis-stats .`). I guess tt construction is just not hot.
This also simplifies using even less memory for token trees by compressing equal span, which I plan to do right after.
Some workflows are more easily expressed with a flat tt, while some are better expressed with a tree. With the right helpers, though (which was mostly a matter of trial and error), even the worst workflows become very easy indeed.
I didn't follow rustc precisely, because I think it does some things wrongly (or they are FIXME), but I only allowed more code, not less. So we're all fine.
Swap Subtree::token_trees from Vec to boxed slice
Performs one of the optimizations suggested in #16325, but a little bit more. Boxed slices guarantee `shrink_to_fit` aswell as saving a pointer width as no capacity has to be stored.
Most of the diff is:
- Changing `vec![]` to `Box::new([])`
- Changing initialize -> fill into fill -> into_boxed_slice
- Working around the lack of an owned iterator or automatic iteration over a `Box<[T]>`
I would like to use my own crate, [small-fixed-array](https://lib.rs/small-fixed-array), although I understand if it isn't mature enough for this. If I'm given the go ahead, I can rework this PR to use it instead.
Don't provide `add_missing_match_arms` assist when upmapping match arm list failed
Fixes#15310
We shouldn't provide the assist when we fail to find the original match arm list.
Note that this PR will temporarily make the assist not applicable when attribute macro operates on the match expression in question, just like the case in #15310, for most of the current stable toolchain users. This is because the sysroot-abi proc-macro-srv on the current stable [discards] spans for `Group` delimiters in some code paths, which the popular `proc-macro2` crate almost always calls, and it makes the identity of match arm list's brackets lost, leading to the upmapping failure. This has been fixed by #14960, which will land in the next stable, 1.71.
[discards]: 8ede3aae28/src/tools/rust-analyzer/crates/proc-macro-srv/src/abis/abi_sysroot/ra_server.rs (L231)
