On 11/01, Martin Ågren wrote:
> With --recurse-submodules, we add each submodule that we encounter to
> the list of alternate object databases. With threading, our changes to
> the list are not protected against races. Indeed, ThreadSanitizer
> reports a race when we call `add_to_alternates_memory()` around the same
> time that another thread is reading in the list through
> `read_sha1_file()`.
>
> Take the grep read-lock while adding the submodule. The lock is used to
> serialize uses of non-thread-safe parts of Git's API, including
> `read_sha1_file()`.
>
> Helped-by: Brandon Williams <bmwill@xxxxxxxxxx>
> Signed-off-by: Martin Ågren <martin.agren@xxxxxxxxx>
> ---
> Many thanks to Brandon for showing how this should have been done.
Of course! Happy to help :)
And this looks good, thanks for fixing my mistake!
>
> builtin/grep.c | 2 ++
> 1 file changed, 2 insertions(+)
>
> diff --git a/builtin/grep.c b/builtin/grep.c
> index 2d65f27d0..5a6cfe6b4 100644
> --- a/builtin/grep.c
> +++ b/builtin/grep.c
> @@ -431,7 +431,9 @@ static int grep_submodule(struct grep_opt *opt, struct repository *superproject,
> * store is no longer global and instead is a member of the repository
> * object.
> */
> + grep_read_lock();
> add_to_alternates_memory(submodule.objectdir);
> + grep_read_unlock();
>
> if (oid) {
> struct object *object;
> --
> 2.15.0.415.gac1375d7e
>
--
Brandon Williams
