Re: [PATCH 8/8] submodule--helper: use OPT_SUBCOMMAND() API

[Glen Choo <chooglen@google.com>]

Ævar Arnfjörð Bjarmason <avarab@gmail.com> writes:

> On Thu, Nov 03 2022, Glen Choo wrote:
>
>> Ævar Arnfjörð Bjarmason <avarab@gmail.com> writes:
>> [...]
>>> +	struct option options[] = {
>>> +		OPT_SUBCOMMAND("clone", &fn, module_clone),
>>> +		OPT_SUBCOMMAND("add", &fn, module_add),
>>> +		OPT_SUBCOMMAND("update", &fn, module_update),
>>> +		OPT_SUBCOMMAND("foreach", &fn, module_foreach),
>>> +		OPT_SUBCOMMAND("init", &fn, module_init),
>>> +		OPT_SUBCOMMAND("status", &fn, module_status),
>>> +		OPT_SUBCOMMAND("sync", &fn, module_sync),
>>> +		OPT_SUBCOMMAND("deinit", &fn, module_deinit),
>>> +		OPT_SUBCOMMAND("summary", &fn, module_summary),
>>> +		OPT_SUBCOMMAND("push-check", &fn, push_check),
>>> +		OPT_SUBCOMMAND("absorbgitdirs", &fn, absorb_git_dirs),
>>> +		OPT_SUBCOMMAND("set-url", &fn, module_set_url),
>>> +		OPT_SUBCOMMAND("set-branch", &fn, module_set_branch),
>>> +		OPT_SUBCOMMAND("create-branch", &fn, module_create_branch),
>>> +		OPT_END()
>>> +	};
>>> +	argc = parse_options(argc, argv, prefix, options, usage, 0);
>>> +	subcmd = argv[0];
>>> +
>>> +	if (strcmp(subcmd, "clone") && strcmp(subcmd, "update") &&
>>> +	    strcmp(subcmd, "foreach") && strcmp(subcmd, "status") &&
>>> +	    strcmp(subcmd, "sync") && strcmp(subcmd, "absorbgitdirs") &&
>>> +	    get_super_prefix())
>>> +		/*
>>> +		 * xstrfmt() rather than "%s %s" to keep the translated
>>> +		 * string identical to git.c's.
>>> +		 */
>>> +		die(_("%s doesn't support --super-prefix"),
>>> +		    xstrfmt("'%s %s'", cmd, subcmd));
>>
>> FYI I'm preparing a series to get rid of the SUPPORT_SUPER_PREFIX checks
>> in both the top level and in submodule--helper (i.e. revisiting my
>> complaints from [1]), but I haven't sent it out yet because I haven't
>> found the right way to motivate that change. Feel free to chime in on
>> that series when it comes out.
>>
>> [1] https://lore.kernel.org/git/20220630221147.45689-5-chooglen@google.com/
>
> Maybe you have different plans, but keep at the WIP re-roll of what I
> have after this, I've also removed all of that.
>
> Basically, ending up with:
> 	
> 	--- a/builtin.h
> 	+++ b/builtin.h
> 	@@ -224,7 +224,14 @@ int cmd_sparse_checkout(int argc, const char **argv, const char *prefix);
> 	 int cmd_status(int argc, const char **argv, const char *prefix);
> 	 int cmd_stash(int argc, const char **argv, const char *prefix);
> 	 int cmd_stripspace(int argc, const char **argv, const char *prefix);
> 	+int cmd_submodule(int argc, const char **argv, const char *prefix);
> 	 int cmd_submodule__helper(int argc, const char **argv, const char *prefix);
> 	+int cmd_submodule__helper_clone(int argc, const char **argv, const char *prefix);
> 	+int cmd_submodule_absorbgitdirs(int argc, const char **argv, const char *prefix);
> 	+int cmd_submodule_foreach(int argc, const char **argv, const char *prefix);
> 	+int cmd_submodule_status(int argc, const char **argv, const char *prefix);
> 	+int cmd_submodule_sync(int argc, const char **argv, const char *prefix);
> 	+int cmd_submodule_update(int argc, const char **argv, const char *prefix);
> 	 int cmd_switch(int argc, const char **argv, const char *prefix);
> 	 int cmd_symbolic_ref(int argc, const char **argv, const char *prefix);
> 	 int cmd_tag(int argc, const char **argv, const char *prefix);
>
> And changes like:
> 	
> 	@@ -366,7 +365,7 @@ static void runcommand_in_submodule_cb(const struct cache_entry *list_item,
> 	 
> 	                strvec_pushl(&cpr.args, "--super-prefix", NULL);
> 	                strvec_pushf(&cpr.args, "%s/", displaypath);
> 	-               strvec_pushl(&cpr.args, "submodule--helper", "foreach", "--recursive",
> 	+               strvec_pushl(&cpr.args, "submodule--helper-foreach", "--recursive",
> 	                             NULL);
> 	 
> 	                if (info->quiet)
>
> I.e. when we call "foreach" we dispatch to cmd_submodule_foreach(), but
> when "foreach" needs to recurse it doesn't invoke a "git submodule
> foreach", instead it invokes "git submodule--helper-foreach".
>
> The reason for doing that is that we can't promote the helper to a
> built-in without also leaking implementation detail that we support the
> now internal-only --super-prefix in the command itself.
>
> So this code becomes:
> 	
> 	@@ -3352,43 +3304,17 @@ static int module_add(int argc, const char **argv, const char *prefix)
> 	 
> 	 int cmd_submodule__helper(int argc, const char **argv, const char *prefix)
> 	 {
> 	-       const char *cmd = argv[0];
> 	-       const char *subcmd;
> 	        parse_opt_subcommand_fn *fn = NULL;
> 	        const char *const usage[] = {
> 	                N_("git submodule--helper <command>"),
> 	                NULL
> 	        };
> 	        struct option options[] = {
> 	-               OPT_SUBCOMMAND("clone", &fn, module_clone),
> 	-               OPT_SUBCOMMAND("add", &fn, module_add),
> 	-               OPT_SUBCOMMAND("update", &fn, module_update),
> 	-               OPT_SUBCOMMAND("foreach", &fn, module_foreach),
> 	-               OPT_SUBCOMMAND("init", &fn, module_init),
> 	-               OPT_SUBCOMMAND("status", &fn, module_status),
> 	-               OPT_SUBCOMMAND("sync", &fn, module_sync),
> 	-               OPT_SUBCOMMAND("deinit", &fn, module_deinit),
> 	-               OPT_SUBCOMMAND("summary", &fn, module_summary),
> 	-               OPT_SUBCOMMAND("push-check", &fn, push_check),
> 	-               OPT_SUBCOMMAND("absorbgitdirs", &fn, absorb_git_dirs),
> 	-               OPT_SUBCOMMAND("set-url", &fn, module_set_url),
> 	-               OPT_SUBCOMMAND("set-branch", &fn, module_set_branch),
> 	-               OPT_SUBCOMMAND("create-branch", &fn, module_create_branch),
> 	+               OPT_SUBCOMMAND("push-check", &fn, cmd_submodule_push_check),
> 	+               OPT_SUBCOMMAND("create-branch", &fn, cmd_submodule_create_branch),
> 	                OPT_END()
> 	        };
> 	        argc = parse_options(argc, argv, prefix, options, usage, 0);
> 	-       subcmd = argv[0];
> 	-
> 	-       if (strcmp(subcmd, "clone") && strcmp(subcmd, "update") &&
> 	-           strcmp(subcmd, "foreach") && strcmp(subcmd, "status") &&
> 	-           strcmp(subcmd, "sync") && strcmp(subcmd, "absorbgitdirs") &&
> 	-           get_super_prefix())
> 	-               /*
> 	-                * xstrfmt() rather than "%s %s" to keep the translated
> 	-                * string identical to git.c's.
> 	-                */
> 	-               die(_("%s doesn't support --super-prefix"),
> 	-                   xstrfmt("'%s %s'", cmd, subcmd));
> 	 
> 	        return fn(argc, argv, prefix);
> 	 }
>
> I.e. for all the "super-prefix" ones we dispatch directly (and maybe I
> should just do that too for those two stragglers).
>
> It's ugly, but it's only ugly on the inside, but if you can think of a
> better way to do it...

If we _really_ wanted to keep the check, an alternative might be to
teach the subcommand parser about SUPPORT_SUPER_PREFIX. But frankly, I
think this SUPPORT_SUPER_PREFIX check is far more trouble than it's
worth, and I wouldn't want you to spend your time trying to find ways to
keep it only for it to be dropped altogether.

Here's a snippet from the cover letter I'm working on, which will
hopefully convince you that we don't need to worry about this any more.

----- >8 --------- >8 --------- >8 --------- >8 --------- >8 ----

  When we introduced the internal-only "--super-prefix" in 74866d7579 (git: make
  super-prefix option, 2016-10-07), we specified that commands must prefix paths
  by it, and pathspecs must be parsed relative to it. That commit also includes
  safeguards to ensure that "--super-prefix" is used correctly, namely:

  - Only commands marked SUPPORT_SUPER_PREFIX can be invoked with "--super-prefix".
  - The super prefix is propagated via the GIT_INTERNAL_SUPER_PREFIX env var, so a
    non-SUPPORT_SUPER_PREFIX command cannot be invoked by a SUPPORT_SUPER_PREFIX
    one.

  However, its use is inconsistent, which is a strong reason to consider using
  better-scoped flags instead. For example..

    - Of the 4 commands that are SUPPORT_SUPER_PREFIX, only "read-tree" and
      "submodule--helper" do anything useful with it. "fsmonitor--daemon" has it
      to avoid the SUPPORT_SUPER_PREFIX warning [1]. "checkout--worker" doesn't have
      a documented reason, but since it can be invoked by "read-tree", it's
      presumably also a workaround.
    - "read-tree" and "submodule--helper" use different values for the super prefix;
      "read-tree" passes the path from the root of the superproject's tree to the
      submodule's gitlink, while "submodule--helper" passes the relative path of the
      original CWD to the submodule.
    - "submodule--helper" doesn't use "--super-prefix" to parse pathspecs, only to
      display paths.

  This series fixes replaces "--super-prefix" with such better-scoped flags, and
  fixes some bugs resulting from the SUPPORT_SUPER_PREFIX check.

[1] 53fcfbc84f (fsmonitor--daemon: allow --super-prefix argument, 2022-05-26)

----- >8 --------- >8 --------- >8 --------- >8 --------- >8 ----

I figured out all the details, but it would look something like:

- Add an internal-only "--tree-super-prefix" flag to "git read-tree",
  which sets a global variable that is read from unpack-trees.c.
- Add an internal-only "--wt-super-prefix" flag to "git
  submodule--helper", which sets a global variable that is read from
  submodule.c. Unlike "--super-prefix", this won't be gated behind a
  SUPPORT_SUPER_PREFIX for each subcommand, since AFAICT, every
  subcommand is using this "--wt-super-prefix" correctly, so we can just
  make sure that new subcommands do too.
- Remove the global "--super-prefix", the corresponding env var and
  SUPPORT_SUPER_PREFIX.