On Thu, May 18, 2017 at 12:00 PM, Brandon Williams <[email protected]> wrote:
> On 05/18, Stefan Beller wrote:
>> >> +static enum {
>> >> + SUBMODULE_CONFIG_NOT_READ = 0,
>> >> + SUBMODULE_CONFIG_NO_CONFIG,
>> >> + SUBMODULE_CONFIG_EXISTS,
>> >> +} submodule_config_reading;
>> >
>> > Any way we can have this not be a global, but rather a parameter? You
>> > could pass in a pointer to this value via the callback data parameter in
>> > the submodule_config function.
>>
>> As said in the reply to Junio, this patch has been sitting on my hard drive
>> for a while and was written before you started the attempt to de-globalize
>> the state of git.
>>
>> Ideally this setting would be part of the repository object. For example
>> the repository object would have a "submodule_config" pointer, initialized
>> to NULL, which can then be set to the read config or a static empty_config
>> if no such config exists.
>
> I'm not quite sure I agree, or rather we may be talking about two
> different things or I'm misinterpreting the patches. From these patches
> it seems like 'submodule_config' that you are refering to is not the
> actual submodule configuration but rather some options that are stored
> in .git/config or other various config locations (home, system, etc).
You are reading the patch correctly.
> What would need to be part of the repository object (and is in my WIP
> that I'll hopefully send out so i can get some feedback) would be the
> submodule_cache which is the internal representation of a repository's
> .gitmodules files.
and in the light of this patch, we'd want to have a cache the flag if the
regular config contains any submodule related things, so for now a static
seems to be the best option and once we have the repo object we'd
have a bit from the flags section (assuming we'll have a flags
section down the road):
struct repo {
....
struct regular_config *config_ptr;
unsigned config_has_submodule_stuff : 1;
unsigned config_loaded_submodule_config : 1;
}
Then the caching decision would be:
static int submodule_config(const char *var, const char *value, void *cb)
{
struct repo *r = cb;
if (!strcmp(var, "submodule.fetchjobs")) {
r->config_has_submodule_stuff = 1;
parallel_jobs = git_config_int(var, value);
if (parallel_jobs < 0)
die(_("negative values not allowed for
submodule.fetchJobs"));
return 0;
} else if (starts_with(var, "submodule.")) {
r->config_has_submodule_stuff = 1;
return parse_submodule_config_option(var, value);
} else if (!strcmp(var, "fetch.recursesubmodules")) {
r->config_has_submodule_stuff = 1;
config_fetch_recurse_submodules =
parse_fetch_recurse_submodules_arg(var, value);
return 0;
}
return 0;
}
void load_submodule_config(struct repo *r)
{
# assume r->config_has_submodule_stuff and config_loaded_submodule_config
# was set to 0 on repo init
if (r->config_loaded_submodule_config)
return;
git_config(submodule_config, r);
r->config_loaded_submodule_config = 1;
}
That said, I agree that the
>> >> +static enum {
>> >> + SUBMODULE_CONFIG_NOT_READ = 0,
>> >> + SUBMODULE_CONFIG_NO_CONFIG,
>> >> + SUBMODULE_CONFIG_EXISTS,
>> >> +} submodule_config_reading;
with its global state is counterproductive for your series, but I see
an easy integration path. As we do not have the repo struct,
I proposed it this way to make submodule progress.
Thanks,
Stefan
