Fix panics due to bad offsets when calling rowan `token_at_offset`
withered-magic opened this issue · comments
Currently only see this behavior in Zed, but pretty often can get errors like
stderr: server: initializing state and starting event loop
stderr: server: determining workspace output_base
stderr: server: running "bazel info build-language"
stderr: thread '<unnamed>' panicked at external/crates__rowan-0.15.15/src/cursor.rs:751:9:
stderr: Bad offset: range 0..43 offset 47
stderr: note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace
stderr: Rayon: detected unexpected panic; aborting
Haven't seen this at all in VSCode, so I'm thinking it's maybe something Zed specific about the order in which it delivers textDocument/didChange notifications and positional requests like goto def
Actually I can get this to happen for rust-analyzer as well
stderr: thread 'Worker' panicked at /Users/runner/.cargo/registry/src/index.crates.io-6f17d22bba15001f/rowan-0.15.15/src/cursor.rs:786:13:
stderr: Bad range: node range 0..3, range 4..4
stderr: :
stderr: Bad range: node range 0..3, range 4..4
rust-analyzer catches these types of panics, so we should probably implement similar functionality (right now we only handle salsa::Cancelled)
@zaucy FYI, addressing this here in case you've seen this with Zed
I actually haven't ran into this! Or I just haven't noticed. Could be platform dependent? I'm mainly running on Windows right now.
Hmm, potentially, not familiar enough with how Zed does LSP to be sure though.. either way it should be fixed across platforms now!