Mercurial fetcher may suffer from issues similar to git smudging

[roberth]

Describe the bug

I am unfamiliar with mercurial and I haven't confirmed whether these are truly a reproducibility problems, but we have to make sure that the inputs to the fetcher fully determine the returned file system objects.

• Keyword Extension might adversely affect the working directory contents.
• Large Files Extension may or may not be enabled, leading to different results
• possibly more

tl;dr Mercurial's more narrow use may cause important issues to go unnoticed while users work around them, but require a breaking change for the problems to be fixed properly.
Mercurial isn't as frequently used as git, and it might take the involvement of an unconfigured or "misconfigured" mercurial to figure this out. Users might solve the problem by changing configuration, but this would make any expression that depends on a mercurial tree to be second class unless we solve these issues. Solving them may cause breaking changes, which is why I'm writing this issue as a non-hg user.

Steps To Reproduce

1. Enable or disable a Mercurial extension
2. Clear ~/.cache/nix
3. Fetch from a repo that is affected by the extension
4. Observe that a different store object is returned

Expected behavior

Nix produces the same results regardless of what's in the Mercurial configuration.

nix-env --version output

Additional context

fetchTree

Priorities

Add 👍 to issues you find important.

[Ericson2314]

I am tempted to demote builtins.fetchMercurial to being an unstable feature. IMO it was sort of an oversight to grandfather it in (vs we correctly did not grandfather in unstable features like the new CLI).