Re: [PATCH v7 03/17] hook: add list command

[Emily Shaffer <emilyshaffer@google.com>]

On Sat, Jan 30, 2021 at 07:10:11PM -0800, Jonathan Tan wrote:
> 
> > Teach 'git hook list <hookname>', which checks the known configs in
> > order to create an ordered list of hooks to run on a given hook event.
> > 
> > Multiple commands can be specified for a given hook by providing
> > multiple "hook.<hookname>.command = <path-to-hook>" lines. Hooks will be
> > run in config order. If more properties need to be set on a given hook
> > in the future, commands can also be specified by providing
> > "hook.<hookname>.command = <hookcmd-name>", as well as a "[hookcmd
> > <hookcmd-name>]" subsection; at minimum, this subsection must contain a
> > "hookcmd.<hookcmd-name>.command = <path-to-hook>" line.
> 
> I learned later that this isn't true - in patch 6, the commit message
> and one of the tests therein describe being able to skip a previously
> inline command by making a hookcmd section of the same name and just
> specifying "skip = true" there (without any command).
> 
> Maybe just delete the "at minimum" part.

Thanks, nice catch. I dropped "at minimum" and s/must/should/. Since I
received feedback during the dogfood internally that there's no
documentation on "hookcmd.<thing>.skip" whatsoever, I'll try to make
a more detailed overview of hookcmd acting as an alias in the
user-facing docs in that commit.

> 
> > +static int list(int argc, const char **argv, const char *prefix)
> >  {
> > -	struct option builtin_hook_options[] = {
> > +	struct list_head *head, *pos;
> > +	struct hook *item;
> 
> You asked for review on nits too so here's one: "item" should be
> declared in the list_for_each block. (That also makes it easier to see
> that we don't need to free it.)

Ok, done.

> 
> > diff --git a/hook.c b/hook.c
> > new file mode 100644
> > index 0000000000..937dc768c8
> > --- /dev/null
> > +++ b/hook.c
> > @@ -0,0 +1,115 @@
> > +#include "cache.h"
> > +
> > +#include "hook.h"
> > +#include "config.h"
> 
> Usually we put all the includes together without any intervening blank
> lines.

Done.

> 
> > +static void append_or_move_hook(struct list_head *head, const char *command)
> > +{
> > +	struct list_head *pos = NULL, *tmp = NULL;
> > +	struct hook *to_add = NULL;
> > +
> > +	/*
> > +	 * remove the prior entry with this command; we'll replace it at the
> > +	 * end.
> > +	 */
> > +	list_for_each_safe(pos, tmp, head) {
> > +		struct hook *it = list_entry(pos, struct hook, list);
> > +		if (!strcmp(it->command.buf, command)) {
> > +		    list_del(pos);
> > +		    /* we'll simply move the hook to the end */
> > +		    to_add = it;
> 
> "break" here?

I think it's safe to do so, but I think I left it out in case duplicates
do make it into the list somehow. But if they're always being inserted
via "append_or_move_hook()" that should not be an issue, so I'll add the
break.

In fact, fixing this exposed a bug. Later, I add using 'pos' instead of
'head':
  list_add_tail(&to_add->list, pos);
When I'm guaranteed to iterate to the end of the list, that works fine,
because in this implementation the last element of the list has "->next"
set back to "head". But when 'pos' isn't sure to be at the end of the
list, it breaks the list. Whoops.. :)


> 
> > +		}
> > +	}
> > +
> > +	if (!to_add) {
> > +		/* adding a new hook, not moving an old one */
> > +		to_add = xmalloc(sizeof(struct hook));
> 
> Style is to write sizeof(*to_add), I think.

Sure.

> 
> [snip]
> 
> > +struct hook_config_cb
> > +{
> > +	struct strbuf *hookname;
> 
> struct declarations have "{" not on a line on its own.

Sure, I can change it. Although, 'git grep -E "struct \w+$"' tells me
there are a few other offenders :) But the count (60ish minus
doc/relnotes references) is much lower than 'git grep -E "struct \w+
\{$"' (800+ matches) so I'm definitely doing it wrong :D

> 
> Also, "hookname" could just be a char *?

Hrmph. I think my C++ background is showing - to me, calling ".buf"
(which is cheap) a few times is a small price to pay to get length
info and so on for free. "hookname" can come from the user - "git
hook (list|run) repo-special-magic-hook" - so I worry about leaving it as a raw
char* with no size info associated. Am I being too paranoid?

> 	
> > +	struct list_head *list;
> > +};
> > +
> > +static int hook_config_lookup(const char *key, const char *value, void *cb_data)
> > +{
> > +	struct hook_config_cb *data = cb_data;
> > +	const char *hook_key = data->hookname->buf;
> > +	struct list_head *head = data->list;
> > +
> > +	if (!strcmp(key, hook_key)) {
> > +		const char *command = value;
> > +		struct strbuf hookcmd_name = STRBUF_INIT;
> > +
> > +		/* Check if a hookcmd with that name exists. */
> > +		strbuf_addf(&hookcmd_name, "hookcmd.%s.command", command);
> > +		git_config_get_value(hookcmd_name.buf, &command);
> 
> So we don't really care whether git_config_get_value returns 0 or 1 as
> long as it doesn't touch "command" if there is no such hookcmd. That
> fits with how git_config_get_value() is documented, so that's great.
> Perhaps better to document as:
> 
>   If a hookcmd.%s.command config exists, replace the command with the
>   value of that config. (If not, do nothing - git_config_get_value() is
>   documented to not overwrite the value argument in this case.)

OK. Thanks, done.

> 
> > +
> > +		if (!command) {
> > +			strbuf_release(&hookcmd_name);
> > +			BUG("git_config_get_value overwrote a string it shouldn't have");
> > +		}
> > +
> > +		/*
> > +		 * TODO: implement an option-getting callback, e.g.
> > +		 *   get configs by pattern hookcmd.$value.*
> > +		 *   for each key+value, do_callback(key, value, cb_data)
> > +		 */
> > +
> > +		append_or_move_hook(head, command);
> > +
> > +		strbuf_release(&hookcmd_name);
> > +	}
> > +
> > +	return 0;
> > +}
> > +
> > +struct list_head* hook_list(const struct strbuf* hookname)
> 
> "const char *hookname" should suffice?
> 
> Also, search for "* " and replace with " *" where applicable (also in
> the .h file).

See above.

> 
> > +{
> > +	struct strbuf hook_key = STRBUF_INIT;
> > +	struct list_head *hook_head = xmalloc(sizeof(struct list_head));
> > +	struct hook_config_cb cb_data = { &hook_key, hook_head };
> > +
> > +	INIT_LIST_HEAD(hook_head);
> > +
> > +	if (!hookname)
> > +		return NULL;
> > +
> > +	strbuf_addf(&hook_key, "hook.%s.command", hookname->buf);
> > +
> > +	git_config(hook_config_lookup, (void*)&cb_data);
> 
> Do we need this void* cast?

Nope. Dropped, thanks.

> 
> > +
> > +	strbuf_release(&hook_key);
> > +	return hook_head;
> > +}
> > diff --git a/hook.h b/hook.h
> > new file mode 100644
> > index 0000000000..8ffc4f14b6
> > --- /dev/null
> > +++ b/hook.h
> > @@ -0,0 +1,26 @@
> > +#include "config.h"
> > +#include "list.h"
> > +#include "strbuf.h"
> > +
> > +struct hook
> > +{
> > +	struct list_head list;
> > +	/*
> > +	 * Config file which holds the hook.*.command definition.
> > +	 * (This has nothing to do with the hookcmd.<name>.* configs.)
> > +	 */
> > +	enum config_scope origin;
> > +	/* The literal command to run. */
> > +	struct strbuf command;
> 
> "char *" would suffice?

Since 'command' comes directly from user input and is executed, I'm
nervous about using a char* instead of a strbuf even more so than for
the hookname.

> 
> The tests look fine.

Thanks for the nitpicky review, it was great! :)

 - Emily