Don't canonicalise file keys

[lionel-]

Branched from #1245
Progress towards #1212

While I was working on breakpoints I ran into difficulties matching files from different locations: frontend URIs via DAP requests and source references from R whose paths went through normalizePath(). The latter does file canonicalisation, which means that in addition to resolving . and .., it also follows symlinks. The frontend URIs are not canonicalised, which caused CI failures when running tests in projects storedc in temporary folders, with mismatches between /tmp and /private/tmp. The fix I implemented was to canonicalise everywhere and encode that canonicalisation happened with the UrlId file, which we stored as file keys in our maps tracking state (e.g. breakpoints) by file.

The UrlId normalisation also dealt with other issues that surfaced over time, such as windows letter drive casing because to fix issues with lowercase on one end and uppercase on the other.

The same sort of issue came up with the LSP: we get URIs from the frontend, but we also encounter source() calls that may be relative and we need to resolve them. So I reached for the same UrlId approach to make the different file paths agree with each other.

It turns out that doing any sort of normalisation that requires I/O is a red flag. If a file gets deleted on disk, and we get an event about that file, there is literally no way to match it back to what we know. For the LSP that's a real issue because we get file deletion events after the file is deleted.

The other thing to watch out for: Always report back the unormalised URI/path to the frontend, because that's what they use to map files on their end. Any changes might cause the frontend to treat the file as different.

With this PR:

• We no longer canonicalise files that we store as keys. We only do lexical normalisation: resolve . and ... That matches r-a and ty.
• We still do canonicalisation at comparison time, in a secondary lookup. This way we can still match file paths coming from different sources.
• We keep track of original file URI/paths for communication with the frontend. For now only in the DAP, this will be done for the LSP in another PR.

[DavisVaughan]

It turns out that doing any sort of normalisation that requires I/O is a red flag. If a file gets deleted on disk, and we get an event about that file, there is literally no way to match it back to what we know.

LOL OUCH

[DavisVaughan]

I simultaneously:

• am sad we dont have a canonical uri everywhere, but get it, especially with the deleted file example
• am happy we dont have to do IO for uri normalization

[DavisVaughan]

dont we also normalize ..?

[DavisVaughan]

you mention it in the pr, but i dont see it anywhere here, so thats why im asking

We only do lexical normalisation: resolve . and ...

[DavisVaughan]

would be nice to see a test in this file either way!

[lionel-]

I can't find it either! Must have made a rebase mistake along the way or other... I distinctly remember component parsing code.

Claude says "Url::parse already collapses dot segments for file: URLs, but Url::from_file_path doesn't". Maybe I missed that the agent decided to remove that code at some point because of the parse behaviour 😬

[DavisVaughan]

Yea IIUC this isn't actually what is happening

In handle_set_breakpoints() we do this

let uri = match UrlId::from_file_path(path)

and we store that uri in by_url here.

So I'm not entirely sure "the frontend sees the uri it sent". It might be lexically normalized?

[DavisVaughan]

Which is either fine, and needs a doc comment update, or isn't fine, and we need a third map that gets us back to the original path: String the DAP provided us 😬

[DavisVaughan]

And either way, a test could be good

[lionel-]

I think we should be consistent and send the verbatim path back here, thanks for catching this

[lionel-]

Instead of a third map, I've just included the verbatim path in the first map. I wonder if I should store the verbatim in the path variant of FilePath, in the next PR.