cannot run `sl` anymore in a dotgit repo

[bolinfest]

Here is the error I am getting:

$ sl
reposetup failed in extension dirsync: When constructing alloc::boxed::Box<dyn commits_trait::DagCommits + core::marker::Send> from dyn storemodel::StoreInfo, "10-git-commits" reported error

Caused by:
    0: Resolve Git reference: ""
    1: [] is not a 40-byte hex string
abort: When constructing alloc::boxed::Box<dyn commits_trait::DagCommits + core::marker::Send> from dyn storemodel::StoreInfo, "10-git-commits" reported error

Caused by:
    0: Resolve Git reference: ""
    1: [] is not a 40-byte hex string

also:

$ sl doctor
checking internal storage
abort: repository /Users/mbolin/code/openai/.git not found!

but that path exists and is the .git folder in the root of my repo.

I also tried deleting /Users/mbolin/code/openai/.git/sl, but I still get the same error.

It is quite expensive to re-clone, so I was hoping to avoid resorting to that.

[bolinfest]

Possibly related, is that git gc failed with:

fatal: unable to read 2fafd4ab95b29fceea9e4a9ef0bd4add3e83f7ae
fatal: failed to run repack

[bolinfest]

In the process of trying to fix this, Codex added this to .git/config:

[core]
	promisor = true
        partialCloneFilter = blob:none

though Sapling doesn't seem to like that, so I've had to disable it.

[bolinfest]

OK, so I removed those options and got things back to where sl runs again.

But I still have this problem, which I have had for awhile where running sl log . in a subfolder errors with:

error.HttpError: Command exited with code 128
  git -c fetch.negotiationAlgorithm=noop --git-dir=/Users/mbolin/code/openai/.git fetch https://github.com/openai/openai.git --no-tags --no-write-fetch-head --recurse-submodules=no --filter=blob:none --stdin --progress
    fatal: remote error: upload-pack: not our ref 0000000000000000000000000000000000000000

[quark-zju]

It seems there are multiple issues. I'll try to comment on each one.

Issue A:

1: [] is not a 40-byte hex string

This looks similar with #1007 and was fixed with a 20250616 commit. What's your current sl version?

---

Issue B:

sl doctor

It's known that sl doctor does not work in git repos. It needs to be updated to skip parts that are irrelevant to git.

---

Issue C:

fatal: unable to read 2fafd4ab95b29fceea9e4a9ef0bd4add3e83f7ae
fatal: failed to run repack

I've seen this from internal reports using the dotsl git mode. It seems to be a cgit upstream issue related to git gc and the promisor enabled. We haven't fully understood this issue yet (requires understanding cgit or help from upstream). The workaround could be disabling git gc (or only enable the multi-pack-index gc).

---

Issue D:

fatal: remote error: upload-pack: not our ref 0000000000000000000000000000000000000000

This seems to be a new issue. It seems like the NULL_ID (Mercurial concept) is leaked somewhere. To troubleshoot, run:

git for-each-ref

And check if there are all-zero hashes associated with a reference. If so, try deleting it using git update-ref -d BAD_REF_NAME. Otherwise, might need to debug deeper into the git exchange protocol using GIT_TRACE2. If it's indeed a bad reference, before deleting, run git reflog show BAD_REF_NAME might give us some ideas about how it got created in the first place.

[bolinfest]

Sorry, I should have included the version in the original report:

$ sl version
Sapling 0.2.20250521-115337-25ed6ac4

Which is the latest stable version, which is nearly 6 months old at this point...