git-clone ignores submodule.recurse configuration

git-clone ignores submodule.recurse configuration

[Ralf Thielow]

Today I played around a bit with git submodules and noticed
that the very handy configuration "submodule.recurse" is not
working for the git-clone command.

"git help config" tells me that submodule.recurse affects
all commands that have a --recurse-submodules option.
git-clone seems to be an exception which is also mentioned in
a workflow example in Documentation/gitsubmodules.txt that
says:

  # Unlike the other commands below clone still needs
  # its own recurse flag:
  git clone --recurse <URL> <directory>
  cd <directory>

So this seems to be a known issue. Is someone already
working on this or has a plan to do it? Or is there a reason
not doing this for git-clone but for git-pull?

Re: git-clone ignores submodule.recurse configuration

[Brandon Williams]

On 12/01, Ralf Thielow wrote:
> Today I played around a bit with git submodules and noticed
> that the very handy configuration "submodule.recurse" is not
> working for the git-clone command.
> 
> "git help config" tells me that submodule.recurse affects
> all commands that have a --recurse-submodules option.
> git-clone seems to be an exception which is also mentioned in
> a workflow example in Documentation/gitsubmodules.txt that
> says:
> 
>   # Unlike the other commands below clone still needs
>   # its own recurse flag:
>   git clone --recurse <URL> <directory>
>   cd <directory>
> 
> So this seems to be a known issue. Is someone already
> working on this or has a plan to do it? Or is there a reason
> not doing this for git-clone but for git-pull?

When we introduced the "submodule.recurse" config we explicitly had it
not work with clone because a recursive clone ends up pulling data from
other sources aside from the URL the user explicitly provides.
Historically there have been a number of security related bugs with
respect to cloning submodules so we felt it was best to require a
recursive clone to be requested explicitly, at least for the time being.

-- 
Brandon Williams

Re: git-clone ignores submodule.recurse configuration

[Stefan Beller]

On Fri, Dec 1, 2017 at 1:47 PM, Brandon Williams <bmwill@google.com> wrote:
> On 12/01, Ralf Thielow wrote:
>> Today I played around a bit with git submodules and noticed
>> that the very handy configuration "submodule.recurse" is not
>> working for the git-clone command.

The rationale here is that submodule.recursive ought to make your
conglomerate of repositories (superproject + some submodules)
look and feel like one giant repository.  Thinking further this sounds
like a user wants to set it globally, hence it may be expected to
work for fresh clones, too.

However at clone time, we don't know which submodules the
user wants. The first submodule recursive switches were an
all or nothing, but I think we need a better selection there.
For fetch there is an "only-those-needed" selection as
"on-demand". Similar for push.

For clone we expected the user wanting to specify which
submodules are interesting:

  git clone --recurse-submodule=<pathspec>

(Note to self: the `[=<pathspec]` part is deep down in the
man page, not in the synopsis, we may want to put that
more prominently there)

Maybe when submodule.recursive is set, we should assume
a pathspec of "." (everything), also see to `submodule.active`,
which is set using the argument from git-clone.

>> "git help config" tells me that submodule.recurse affects
>> all commands that have a --recurse-submodules option.
>> git-clone seems to be an exception which is also mentioned in
>> a workflow example in Documentation/gitsubmodules.txt that
>> says:
>>
>>   # Unlike the other commands below clone still needs
>>   # its own recurse flag:
>>   git clone --recurse <URL> <directory>
>>   cd <directory>
>>
>> So this seems to be a known issue. Is someone already
>> working on this or has a plan to do it? Or is there a reason
>> not doing this for git-clone but for git-pull?
>
> When we introduced the "submodule.recurse" config we explicitly had it
> not work with clone because a recursive clone ends up pulling data from
> other sources aside from the URL the user explicitly provides.
> Historically there have been a number of security related bugs with
> respect to cloning submodules so we felt it was best to require a
> recursive clone to be requested explicitly, at least for the time being.
>
> --
> Brandon Williams