Re: [BUG]: backfill min batch size test case failure on s390x

[Patrick Steinhardt <ps@pks.im>]

Hi,

On Tue, Apr 22, 2025 at 12:31:18PM +0000, Pranav P wrote:
> Hi,
>
> When running `make test` on an s390x machine in Debian it is failing on 'do partial clone 2, backfill min batch size'
> Reference: https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1102106
>
> After processing the command line arguments structure member min_batch_size should have had the value 20
>
> Instead of having the value 20 (--min-batch-size=20) it was having a very large value
>
> min_batch_size in `struct backfill_context` is of type `size_t` and since in the function cmd_backfill, in the
> options struct it is passed on to OPT_INTEGER, which eventually causes
>
> ```
> *(int *)opt->value = strtol(arg, (char **)&s, 10);
> ```
> in parse-options.c line 188. This is writing the data in the first 4 bytes of min_batch_size and on big endian
> systems this will lead min_batch_size to be a big number. This issue is immediately visible in little endian systems.
>
> Changing OPT_INTEGER to OPT_MAGNITUDE seems to be working on x86 and s390x

Yup, indeed.

> ```
> diff --git a/builtin/backfill.c b/builtin/backfill.c
> index 18f9701487..33e1ea2f84 100644
> --- a/builtin/backfill.c
> +++ b/builtin/backfill.c
> @@ -123,7 +123,7 @@ int cmd_backfill(int argc, const char **argv, const char *prefix, struct reposit
>                 .sparse = 0,
>         };
>         struct option options[] = {
> -               OPT_MAGNITUDE(0, "min-batch-size", &ctx.min_batch_size,
> +               OPT_INTEGER(0, "min-batch-size", &ctx.min_batch_size,
>                             N_("Minimum number of objects to request at a time")),
>                 OPT_BOOL(0, "sparse", &ctx.sparse,
>                          N_("Restrict the missing objects to the current sparse-checkout")),
> ```
>
> But on systems where size_t which not be unsigned long, this might lead to an issue.
> So, one other suggestion I have is to change the data type of min_batch_size from size_t to int. But I am not able to
> determine whether a practical upper bound for min_batch_size would exceed what an int variable can store.
> With that clarification, I can a raise patch for the issue.

True, as well. Overall the current state is a bit unfortunate because
it's so easy to get this wrong, and the compiler won't even spit out a
warning.

> I am fairly new to opensource and was following the `git bugreport`. So I am extremely sorry for any lack of clarity in the report.

No need to be sorry, this bug report contains all the details we'd need
and is of quite high quality :) The only thing is that we've already got
this bug reported via [1].

The underlying issue is addressed via [2], which has already been merged
to `next`. It also fixes the underlying issue you observed with
different integer widths as well as with signedness.

Thanks!

Patrick

[1]: https://lore.kernel.org/git/89257ab82cd60d135cce02d51eacee7ec35c1c37.camel@physik.fu-berlin.de/
[2]: https://lore.kernel.org/git/20250401-b4-pks-parse-options-integers-v1-0-a628ad40c3b4@pks.im/