fix(trim-paths): workspace remap under -Zroot-dir

[weihanglo]

What does this PR try to resolve?

With -Zroot-dir,
rustc runs from the specified root dir,
so its working directory is not covered by workspace remap rule and absolute paths were leaking there.

The proposed fix here:

• Cargo unconditionally adds the workspace remap as an {"from":".","to":...} record in the unremap file.
• When the rusc cwd -Zroot-dir changes, the workspace remap record would reflect the change.

Part of #12137

How to test and review this PR?

See commit message for why this is fine.

Note

LLM disclosure: This was found during the experiment of #17309 with LLM.
bootstrap invokes cargo with -Zroot-dir={src},
and std started leaking checkout path from there.

[rustbot]

r? @epage

rustbot has assigned @epage.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

Why was this reviewer chosen?

The reviewer was selected based on:

• Owners of files modified in this PR: @epage, @weihanglo
• @epage, @weihanglo expanded to epage, weihanglo

[epage]

Should we test non-root items as well or do you think the root is enough?

[weihanglo]

Should we test non-root items as well or do you think the root is enough?

Ah. Not enough. I only look at how rustc uses that. However, path dependencies outside the workpace may leak. For example, current if we have

workspace root -> /root
-Zroot-dir     -> /external
bar pkg root  ->  /external/bar` 
bar rustc cwd ->  /external

then bar remap will be /external/bar -> /cargo/path/bar-0.0.1, which cannot really remap out DW_AT_comp_dir as it is /external rustc cwd. A bit unsure how to proceed. We could for this case additionally have a /external -> /some-random-path but I don't know if it is worth addressing or adding to the unremap file.

[rustbot]

This PR was rebased onto a different master commit. Here's a range-diff highlighting what actually changed.

Rebasing is a normal part of keeping PRs up to date, so no action is needed—this note is just to help reviewers.

[weihanglo]

Should we test non-root items as well or do you think the root is enough?

Ah. Not enough. I only look at how rustc uses that. However, path dependencies outside the workpace may leak. For example, current if we have

workspace root -> /root
-Zroot-dir     -> /external
bar pkg root  ->  /external/bar` 
bar rustc cwd ->  /external

then bar remap will be /external/bar -> /cargo/path/bar-0.0.1, which cannot really remap out DW_AT_comp_dir as it is /external rustc cwd. A bit unsure how to proceed. We could for this case additionally have a /external -> /some-random-path but I don't know if it is worth addressing or adding to the unremap file.

Tracked in the tracking issue.