Re: [PATCH] upload-pack.c: use of parse-options API
- From
Jeff King <peff@peff.net>
- Date
- May 18, 2016, 18:08 UTC
- Message-ID
- <20160518180800.GC5796@sigill.intra.peff.net>
- In-Reply-To
- <20160518164019.26443-1-Antoine.Queru@ensimag.grenoble-inp.fr>
On Wed, May 18, 2016 at 06:40:19PM +0200, Antoine Queru wrote:
> Option parsing now uses the parser API instead of a local parser. > Code is now more compact.
Sounds like a good goal.
Show 5 quoted lines
> -static const char upload_pack_usage[] = "git upload-pack [--strict] [--timeout=<n>] <dir>";
> +static const char * const upload_pack_usage[] = {
> + N_("git upload-pack [--strict] [--timeout=<n>] <dir>"),
> + NULL
> +};Do we need to enumerate the options here now? The usage message should list the options from "struct options", which make these redundant.
Something like:
git -upload-pack [options] <dir>
probably makes more sense.
Of course, it's hard to read the usage message because...
Show 7 quoted lines
> + struct option options[] = {
> + OPT_HIDDEN_BOOL(0, "stateless-rpc", &stateless_rpc, NULL),
> + OPT_HIDDEN_BOOL(0, "advertise-refs", &advertise_refs, NULL),
> + OPT_BOOL(0, "strict", &strict, NULL),
> + OPT_INTEGER(0, "timeout", &timeout, NULL),
> + OPT_END()
> + };You've left the description text for each of these options as NULL, so running "git-upload-pack -h" segfaults.
I'm not sure whether it is worth hiding the first two options. We typically hide "internal" options like this for user-facing programs, so as not to clutter the "-h" output. But upload-pack isn't a user-facing program. Anybody who is calling it directly with "-h" may be interested in even its more esoteric options.
Show 27 quoted lines
> - for (i = 1; i < argc; i++) {
> - char *arg = argv[i];
> -
> - if (arg[0] != '-')
> - break;
> - if (!strcmp(arg, "--advertise-refs")) {
> - advertise_refs = 1;
> - continue;
> - }
> - if (!strcmp(arg, "--stateless-rpc")) {
> - stateless_rpc = 1;
> - continue;
> - }
> - if (!strcmp(arg, "--strict")) {
> - strict = 1;
> - continue;
> - }
> - if (starts_with(arg, "--timeout=")) {
> - timeout = atoi(arg+10);
> - daemon_mode = 1;
> - continue;
> - }
> - if (!strcmp(arg, "--")) {
> - i++;
> - break;
> - }
> - }A common pitfall in converting a for-loop to parse-options is when there is anything stateful in the parsing loop. But I don't see anything here. The trickiest thing is that --timeout implies daemon_mode. You've handled that below as:
> + if (timeout != 0) > + daemon_mode = 1;
I think that is OK. It would not be correct if, for example, some other code set "timeout" to a non-zero value (e.g., a config option), but I don't see that here.
As a style nit, we usually spell comparison-with-zero as just:
if (timeout) daemon_mode = 1;
Looking at at daemon_mode, though, it appears that it cannot be set in any other way except here. And it does nothing except to disable progress in some cases. Weird. There may be some opportunities for refactoring or simplification there, but I that is outside the scope of your patch.
> + argc = parse_options(argc, (const char **)argv, NULL, options, upload_pack_usage, 0);
Perhaps this is a good opportunity to use "const" in the declaration of main(), as most other git programs do. Then you can drop this cast.
Show 7 quoted lines
> setup_path();
>
> - dir = argv[i];
> + dir = argv[0];
>
> if (!enter_repo(dir, strict))
> die("'%s' does not appear to be a git repository", dir);Prior to your patch, we used to check:
- if (i != argc-1) - usage(upload_pack_usage);
which ensured that "dir" was non-NULL. But with your patch, we may pass NULL to enter_repo. It fortunately catches this, but then we pass the NULL to die, which can segfault (though on glibc systems, stdio is kind enough to replace it with the "(null)").
Related, we silently accept extra arguments after your patch (whereas before we showed the usage message). You probably want to check "argc == 1", and otherwise show the usage message.
-Peff