[pacman-dev] [PATCH] makepkg: Use a recursive git clone.

[Allan McRae]

On 12/03/13 14:04, Allan McRae wrote:
> On 12/03/13 00:08, Jan Steffens wrote:
>> On Tue, Mar 5, 2013 at 10:58 AM, Allan McRae <allan at archlinux.org> wrote:
>>> On 04/03/13 23:06, Allan McRae wrote:
>>>> On 04/03/13 22:58, Jan Steffens wrote:
>>>>> On Mon, Mar 4, 2013 at 1:20 PM, Allan McRae <allan at archlinux.org> wrote:
>>>>>> This would mean --holdver would not work when a project has submodules
>>>>>> given these would be updated.  Also, that would require an internet
>>>>>> connection during "extracting" the sources, but --holdver is suppose to
>>>>>> allow offline VCS builds.   Is there a way we can avoid this?
>>>>>>
>>>>>> Anyone else got anything to add here?
>>>>>>
>>>>>> Allan
>>>>>
>>>>> The state of the submodules is fixed in the commit of the parent repo
>>>>> - the submodules are locked to commits as well.
>>>>>
>>>>> With --recursive, building with an internet connection would work, but
>>>>> without would not, as I think it will attempt to get the submodules
>>>>> from their sources, not the parent repo.
>>>>
>>>> Is it possible to improve on that?  Can we checkout the submodules earlier?
>>>>
>>>>> Without --recursive, building would fail in any case, as the
>>>>> submodules aren't initialized at all.
>>>>>
>>>>> So I think adding --recursive is the right thing to do.
>>>>>
>>>
>>> Pulling patch - if someone cn figure out how to have submodules and have
>>> all the source downloaded earlier, that would be good...
>>
>> I now think this should be reverted. Initializing and updating the
>> submodules every build causes it to download all the submodules every
>> build. That's hardly optimal, and sometimes the submodules aren't even
>> needed for building.
>>
>> Rather, I think we need to have the PKGBUILD handle submodules
>> manually. So, revert this patch and write PKGBUILDs like this:
>>
>> source=(git://foobar.com/git/foo.git
>>   git://foobar.com/git/foo-module-bar.git
>>   git://foobar.com/git/foo-module-baz.git)
>> md5sums=(SKIP SKIP SKIP)
>>
>> build() {
>>   cd foo
>>
>>   git submodule init
>>   git config submodule.qux/bar.url "$srcdir/foo-module-bar"
>>   git config submodule.qux/baz.url "$srcdir/foo-module-baz"
>>   git submodule update
>>
>>   make
>> }
>>
>> Here, foo is our top-level repo, and foo-module-ba[rz] are the
>> submodules used. The names of the submodules in the foo repo are
>> "qux/bar" and "qux/baz". The PKGBUILD initializes the submodules
>> (copying the config from .gitmodules to the local .git/config), then
>> sets the clone URL for the submodules to the clones we already have.
>> The update then clones from there.
>>
>> Does somepony have an idea how to automate this? In the meantime,
>> please revert the patch so submodules can be handled manually, as
>> shown.
>>
> 
> I am also in favour of reverting this.  I have struck the repeated
> downloading of submodules for each build and it annoyed me.
> 
> The "workaround" you provided works fine and completely separates the
> download and build stages again.  Using submodules is not all that
> common, so I'd say just document this in the wiki and be done.
> 
> Any other comments?
> 

Just a further point - do this in prepare() not build().