worktree: Include file size in DiskState to fix stale buffer reload

[lex00]

Related to #38109
Supersedes #48691
See also #49436 (independent implementation of the same fix)

Problem

When an external tool writes a file using std::fs::write() (truncate + write), the scanner can read the 0-byte file during the truncation window. The buffer reloads to empty. If the subsequent write produces the same mtime, file_updated() sees no DiskState change and never triggers a second reload. The buffer is permanently stuck empty.

Root cause analysis: #38109 (comment)

Fix

Add size: u64 to DiskState::Present { mtime, size }. The derived PartialEq now detects size changes even when mtime matches, so a 0→N byte transition always triggers ReloadNeeded. The scanner already tracks file size in Entry.size — this threads it through File::for_entry() and local_worktree_entry_changed() into the reload comparison.

Remote/collab callers pass size: 0 since remote buffers don't make local reload decisions (the reload path checks !self.is_via_collab()).

Changes

6 files:

• worktree.rs: Add size to DiskState::Present, thread through File::for_entry() and entry change handler
• buffer.rs: Update DiskState enum and file_updated() comparison
• buffer_store.rs: Pass size from entry to local_worktree_entry_changed()
• image_store.rs: Same for images
• copilot.rs: Pass size: 0 for virtual copilot buffers
• project_tests.rs: Two new integration tests

Related issues:

• May improve Zed fails to update internal representation of large files updated outside of Zed #20872, Zed editor doesn't know if the files are changed elsewhere. Especially with GRPC and ORM code generation tools.  #15791, Agent stops due "Unsaved changes" when he's the only one making changes #43990

Note: the two integration tests use allow_parking() + flush_fs_events() and must be run individually (--exact), not in parallel with other RealFs tests.

• Tests
• Code Reviewed
• Manual QA

Release Notes:

• Fixed an issue where open buffers could become permanently empty when external tools (AI agents, formatters) wrote files while they were open in the editor.

[probably-neb]

Thanks for taking a crack at this. I'm going to close as duplicate of #49436.
I realize you've added tests, however they aren't actually reproducing the issue - just testing that the derived equals works with the size added so I don't think they're worth keeping. Especially given that real FS tests tend to be quite flaky in CI.
Thanks again!

[lex00]

Thanks for taking a crack at this. I'm going to close as duplicate of #49436. I realize you've added tests, however they aren't actually reproducing the issue - just testing that the derived equals works with the size added so I don't think they're worth keeping. Especially given that real FS tests tend to be quite flaky in CI. Thanks again!

It wasn't really about the tests :) I was the one who first called it out on 38109 in the comments and made a PR fixing it, which I accidentally closed, my mistake. The PR that re-established the commit references me. Anyway no biggie, take care.

[lex00]

original for reference:
https://github.com/lex00/zed/pull/1#event-22617648782