thread 'LspServer' panicked at 'called Option::unwrap() on a None value', crates/vfs/src/lib.rs:143:38

[bjorn3]

I got the following error several times in a row while working on a project which has a rust-lang/rust checkout in a subdirectory. The final message that vscode won't restart rust-analyzer happened while I had a file in src/test/ui open.

thread 'LspServer' panicked at 'called `Option::unwrap()` on a `None` value', crates/vfs/src/lib.rs:143:38
stack backtrace:
   0: rust_begin_unwind
             at /rustc/4b91a6ea7258a947e59c6522cd5898e7c0a6a88f/library/std/src/panicking.rs:584:5
   1: core::panicking::panic_fmt
             at /rustc/4b91a6ea7258a947e59c6522cd5898e7c0a6a88f/library/core/src/panicking.rs:142:14
   2: core::panicking::panic
             at /rustc/4b91a6ea7258a947e59c6522cd5898e7c0a6a88f/library/core/src/panicking.rs:48:5
   3: rust_analyzer::global_state::GlobalState::process_changes
   4: rust_analyzer::main_loop::<impl rust_analyzer::global_state::GlobalState>::handle_event
   5: rust_analyzer::main_loop::<impl rust_analyzer::global_state::GlobalState>::run
   6: rust_analyzer::main_loop::main_loop
   7: rust_analyzer::run_server
note: Some details are omitted, run with `RUST_BACKTRACE=full` for a verbose backtrace.
[Info  - 1:26:23 PM] Connection to server got closed. Server will restart.

rust-analyzer version: rust-analyzer version: 0.3.1203-standalone

rustc version: rustc 1.65.0-nightly (750bd1a7f 2022-09-14)

relevant settings: None that I know of.

[Veykril]

The panic comes from fetching file contents of a file that is not in the vfs ... I assume you git checkout at least once before this started?

[bjorn3]

Yes, although even after I restarted rust-analyzer it has been giving similar issues several times. Maybe the fact that the rust checkout is in the .gitignore of the parent repo is related?

[Veykril]

I doubt that, so this panic is odd, I wonder if there is a race condition here because we apparently created a ChangeFile that is not a deletion but where the file id was already deleted. The panic happens here https://github.com/rust-lang/rust-analyzer/blob/master/crates/rust-analyzer/src/global_state.rs#L210

[Veykril]

Okay I can see one way how this can crash once (not repeatedly unless the same scenario is replicated). If in one process_changes call, the VFS received a modify and a delete for a given file, then when we process the modify we will crash on the linked line, because the changes happen immediately to the VFS and so process_changes observes the state only after the VFS has changed completely.

[bjorn3]

To give a bit more info, the scripts/test_rustc_tests.sh script checks out a specific commit, reverts any changes to the working tree, applies some patches and deletes certain files and directories and after that runs the rustc test suite. Maybe it is a race condition between handling the reverting of all changes and subsequently applying patches and removing files? Or between separate inotify instances for different directories? By the way this is on Linux in case that is relevant.

[Veykril]

Ye that pretty much describes a series of action that can cause the problem described in my previous comment, but I am still unsure how that would get the server stuck in an inconsistent state, unless panicking in that code can poison something there 🤔