fix download-llvm logic for subtree sync branches

[RalfJung]

Fixes #101907

Cc @onur-ozkan
r? @Mark-Simulacrum

[Mark-Simulacrum]

@bors r+

[bors]

📌 Commit 88f14a8 has been approved by Mark-Simulacrum

It is now in the queue for this repository.

[klensy]

This change partially merged in #137594, weird.

[onur-ozkan]

@bors r-

It's already merged, and regressed LLVM commit finding logic (see #137661 (comment)).

[RalfJung]

I have updated the PR based on what I found here. I still think we shouldn't even try to look at the history when we know that basically none of the history has even been fetched, but that's a much larger change I will not undertake. I hope (and confirmed to the extent that I can, by playing with git log in a sparse checkout) this change now both fixes the issue I am running into and keeps CI working.

@rustbot ready

[Mark-Simulacrum]

r=me if we didn't need to revert this somewhere (I seem to vaguely recall that happening somewhere)

[RalfJung]

We had to revert an earlier version of this. So I went for a different approach now, which should fix that issue.

@bors r=Mark-Simulacrum

[bors]

📌 Commit 0ee9456 has been approved by Mark-Simulacrum

It is now in the queue for this repository.

[jieyouxu]

Hm, this might've somehow broken perf, but I'm not 100% (see #138873 (comment))

[jieyouxu]

It's like the rev-list args is somehow wrong because the rustc-perf log output has

usage: git rev-list [OPTION] <commit-id>... [ -- paths... ]
         limiting output:
           --max-count=<n>
           --max-age=<epoch>
           --min-age=<epoch>
           --sparse
           --no-merges
           --min-parents=<n>
           --no-min-parents
           --max-parents=<n>
           --no-max-parents
           --remove-empty
           --all
           --branches
           --tags
           --remotes
           --stdin
           --quiet
         ordering output:
           --topo-order
           --date-order
           --reverse
         formatting output:
           --parents
           --children
           --objects | --objects-edge
           --unpacked
           --header | --pretty
           --abbrev=<n> | --no-abbrev
           --abbrev-commit
           --left-right
           --count
         special purpose:
           --bisect
           --bisect-vars
           --bisect-all

[jieyouxu]

I think --diff-merges is a git show-only flag, not git rev-list.

[RalfJung]

I tried it locally and it works for git rev-list as well.

$ git --version
git version 2.47.2

[jieyouxu]

I posted a revert #138878 to fix perf, because I know nothing about the distinction between these two flags, so a fix-forward seems also risky.