On Wed, Feb 24, 2016 at 05:59:00PM -0500, David Turner wrote:
> @@ -1207,6 +1208,29 @@ int git_config_early(config_fn_t fn, void *data, const
> char *repo_config)
> }
>
> if (repo_config && !access_or_die(repo_config, R_OK, 0)) {
> + char *storage = NULL;
> +
> + /*
> + * make sure we always read the ref storage config
> + * from the extensions section on startup
> + */
> + ret += git_config_from_file(ref_storage_backend_config,
> + repo_config, &storage);
> +
> + register_ref_storage_backends();
> + if (!storage)
> + storage = xstrdup("");
> +
> + if (!*storage ||
> + !strcmp(storage, "files")) {
> + /* default backend, nothing to do */
> + free(storage);
> + } else {
> + if (set_ref_storage_backend(ref_storage_backend))
> + die(_("Unknown ref storage backend %s"),
> + ref_storage_backend);
> + }
> +
Coverity complains that "storage" leaks here, and I think it does in the
case that we non-default storage, and we successfully set up the
backend. That's a pretty minor point.
However, after looking at this code, I'm rather confused about a few
things.
One is that we read the config and use ref_storage_backend_config[1] to
store the string value in "storage", and then we check whether that is
non-default, and if it is, I'd expect us to feed it to
set_ref_storage_backend. But we don't; we feed ref_storage_backend
instead!
What is that value? It looks like it is the string we set in
check_repo_format when we load the extensions list there. So why
are we re-reading the config here at all? Couldn't we just use
ref_storage_backend in the first place?
My second confusion is why this is happening in git_config_early(). That
function is called during the setup of check_repository_format_gently(),
which is why I think you wanted to put the code here. But it's _also_
called as part of a regular git_config(). Which means we're parsing the
repo config and setting the ref backend all over again, every time we
look at config for other reasons.
So I think this setup probably should be in
check_repository_format_gently(), and should be able to trigger off of
the existing ref_storage_backend string we've already saved (and we
should bail immediately there if we don't know about the backend, as it
means we _don't_ match the repo's extensions and cannot proceed).
-Peff
[1] The ref_storage_backend_config function uses xstrdup(), which I
think will segfault for a value-less boolean config, like:
[extensions]
# notice no "=" sign!
refstorage
The same bug is in check_repo_format, where we assigned
ref_storage_backend. The normal way to do this is to use:
return git_config_string(var, value, &ref_backend_storage);
which checks for the boolean case and complains.
--
To unsubscribe from this list: send the line "unsubscribe git" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
