ok op@

When reusing a temporary file, do not forget to reset its file position to zero. Otherwise, opening multiple raw objects in a row can fail during delta application with an unexpected file size reported by stat(2). Problem reported by Aaron Lin Problem reproduced and fix verified with a repository provided by Aaron. I have unsuccessfully tried to write a regression test for this. ok op@

When sending the configuration to the newly reloaded gotd process we must send the path to the secrets file if it is being used. Otherwise, the new process will die with "unexpected amount of data received in imsg" errors. Test coverage will be added later as part of some new gotsysd tests which I have in progress.

Reset reload secrets state on reload failure, such that the next reload attempt will not be rejected with an 'unexpected imsg' error.

These symbols are also part of configuration file syntax and should therefore be avoided.

We need to enforce additional restrictions on reference names specified in gotsys.conf in order to prevent escaping to arbitrary gotd.conf syntax. Add test coverage.

This directive will not be implemented in the final design. Instead we will be adding authentication to gotwebd with associated config syntax.

SMTP relays for email notifications will only be configured server-side.

While we must install staged blob contents during revert if there are staged changes, the base blob ID recorded in the file index must always reflect the actual base blob from the base commit. Setting the base blob ID to that of a staged blob causes problems. The staged blob's ID is tracked in a separate field. ok op@

Use a dedicated flag to keep track of HEAD ref creation state, rather than checking pe == NULL.

Avoids the need to use got clone -b to get anything at all from a server implementation which does not advertise a HEAD symref. Patch by martijn@

When a reference update sends an empty pack file because the objects already exist on a different branch, gotd was wrongly rejecting the reference update. The logic which protects references from moving kicked in, but didn't account for this particular case. Run a YCA check to ensure the client is only appending new commits to the branch and allow the ref update to proceed if so. Add test coverage. Problem reported by martijn@

Because copy_ref_delta() was forgetting to update the *outsize output parameter, when clients sent ref-deltas gotd was miscalculating the on-disk size of the pack file and would report "bad offset in pack file" if an object was stored at an offset beyond the miscalculated smaller size.

Fixes regress failures due to 'git gc' complaining, as of 2.50.0: '# pack-refs with:' does not start with '# pack-refs with: ' reported by naddy@ ok op@