Bug: bad errno report from lock_ref_sha1_basic() when contents bad

Hi Junio,

User "mbrochh" on IRC reported a result along the lines of

  error: unable to resolve reference refs/remotes/origin/master: Success
  ! [new branch] master -> origin/master (unable to update local ref)

from 'git fetch' (or pull actually, but clearly it's the fetch part failing). This error message is from lock_ref_sha1_basic(), and suggests a confused error exit taken with errno=0.

I can reproduce the problem by writing a bit of garbage to a remote ref and then fetching, though I usually get some other error in the message (usually ENOENT, which is equally wrong).

As far as I can tell it's caused by resolve_ref_unsafe() returning NULL when the content checks fail:

	if (get_sha1_hex(buffer, sha1) || (buffer[40] != '\0' && !isspace(buffer[40]))) {
		if (flag)
			*flag |= REF_ISBROKEN;
		return NULL;
	}

This used to warn before your 5595635 (resolve_ref(): report breakage to the caller without warning, 2011-10-19), but otherwise suffered from the same problem.

As far as I can see lock_ref_sha1_basic() needs a similar check for REF_ISBROKEN (and presumably the only fix is to nuke it so maybe it should even suggest that to the user). But in general, the "errno may be crucial or worthless" interface of lock_ref_sha1_basic() seems a bit broken...

Thomas Rast <trast@inf.ethz.ch> writes:

Probably.

Sorry, but I am not sure what you mean by this. The function is fairly careful using errno to base its decision and even saves errno when it calls helpers that may overwrite it, so the "interface" does not want "may be crucial or worthless". The implementation may have been broken over time by many patches, but that is a different issue. Which codepaths clobber errno, and is it something you can easily fix with a patch or two?

Junio C Hamano <gitster@pobox.com> writes:

I meant something else, sorry for being so unclear and confused. In particular my comment was supposed to be about resolve_ref_unsafe().

It has two different failure modes:

* Some FS syscall went wrong, and errno is crucial for the user to debug
  it (e.g., ENOENT).

* Reading went fine, but we failed to parse the resulting buffer.  errno
  has some unknown value.

The user of the function needs to be aware of this distinction, and handle it accordingly. Otherwise it risks emitting outright wrong error messages, e.g., in my case I got

  thomas@thomas:~/g(next u+107)$ echo garbage >.git/refs/remotes/origin/master
  thomas@thomas:~/g(next u+107)$ g fetch
  error: unable to resolve reference refs/remotes/origin/master: No such file or directory
  From git://git.kernel.org/pub/scm/git/git
   ! [new branch]      master     -> origin/master  (unable to update local ref)

(The IRC user who asked about this was lucky because "Success" immediately told us that it was an error path not related to syscalls.)