Cached crate metadata falls out of sync

rust-analyzer version: rust-analyzer version: 0.3.1995-standalone (c86d6230f 2024-06-10) [/Users/djc/.vscode/extensions/rust-lang.rust-analyzer-0.3.1995-darwin-arm64/server/rust-analyzer]

rustc version: 1.79.0 (129f3b996 2024-06-10)

editor or extension: VS Code v0.3.1995

relevant settings: nothing special

repository link (if public, optional): N/A

In the largish Cargo workspace (~30 crates) at work, I'm seeing behavior where doing something like a git pull and then triggering RA to reanalyze will often fail to refresh upstream crates within the workspace. We have a bunch of library small crates and then a few larger binary crates downstream of that, and if API changes happen to my utils crate (which my application crate depends on via a path = "../utils" dependency) -- like changing the number of function arguments or adding new fields to a type, RA will signal breakage in the downstream users until I explicitly trigger RA to refresh the upstream crate (generally by hitting Cmd + S to save the file).

I forget if RA ever got this right, but if it did it's been a long time since it has.

My expectation would be that, once a refresh is triggered, all files in the workspace could be checked (maybe filtering by mtime) such that the rebuild can happen for all crates that need it.

Reproduce this in a roughly 30-crate Cargo workspace after a git pull, changing an upstream path dependency and triggering rust-analyzer reanalysis. Compare the downstream diagnostics before and after saving the upstream file with Cmd+S. Done means a refresh checks changed workspace crates and updates downstream errors without requiring an explicit save.