threads / discuss / 47722

git add --all does not respect submodule.<name>.ignore

Subject: git add --all does not respect submodule.<name>.ignore

## tl;dr

3 messages between Jan 30, 2018 and Jan 30, 2018.

replies: 2people: 3as markdown or json

Michael Scott-Nelson· Jan 30, 2018, 20:25 UTC · lore

Giving a submodule "ignore=all" or "ignore=dirty" in .gitmodule successfully removes that module from `git status` queries. However, these same diffs are automatically added by git-add, eg `git add .` or `git add --all` adds the submodules that we want ignored. This seems inconsistent and confusing.

Workflow reason: We want to be able to have supers and subs checked-out, make changes to both, but only update the SHA-1 pointer from super to sub when explicitly forced to do so, eg `git add -f subName`. This workflow prevents continuous merge conflicts from clashing SHA-1 pointers while still allowing `git add --all`, allowing a sort of middle ground between submodules and an untracked library.

Teaching git-add about submodule.ignore and/or teaching .gitignore about submodules would be awesome.

Also experimented with `git update-index --assume-unchanged subName`,
but I believe that it does not get committed and besides also does not
seem to have a way to `git add -f`.
---
Note: currently on git version 2.14.1, but looking through the
changelogs, did not see any changes since then that would enable this
workflow.

-Michael Scott-Nelson

Stefan Beller· Jan 30, 2018, 21:56 UTC · re: Michael Scott-Nelson · lore

Re: git add --all does not respect submodule.<name>.ignore

On Tue, Jan 30, 2018 at 12:25 PM, Michael Scott-Nelson <mscottnelson@gmail.com> wrote:

Show 5 quoted lines

> Giving a submodule "ignore=all" or "ignore=dirty" in .gitmodule
> successfully removes that module from `git status` queries. However,
> these same diffs are automatically added by git-add, eg `git add .` or
> `git add --all` adds the submodules that we want ignored. This seems
> inconsistent and confusing.

My prime suspect for this change would be https://github.com/git/git/commit/5556808690ea245708fb80383be5c1afee2fb3eb

Show 7 quoted lines

> Workflow reason:
> We want to be able to have supers and subs checked-out, make changes
> to both, but only update the SHA-1 pointer from super to sub when
> explicitly forced to do so, eg `git add -f subName`. This workflow
> prevents continuous merge conflicts from clashing SHA-1 pointers while
> still allowing `git add --all`, allowing a sort of middle ground
> between submodules and an untracked library.

For that you want to set the ignore flag locally in .git/config instead of .gitmodules. The .gitmodules seems like a convenient place to "share submodule config", but that is a slippery slope and I think that was a mistake by the project.

(If you control the environment, you could also put the ignore flags into the system wide config)

> Teaching git-add about submodule.ignore and/or teaching .gitignore
> about submodules would be awesome.

I wonder if we can address this issue with even more config options. (sounds bad, but is the first thought I have)

> Also experimented with `git update-index --assume-unchanged subName`,
> but I believe that it does not get committed and besides also does not
> seem to have a way to `git add -f`.

The assume-unchanged bit is a performance optimisation for powerusers, but its documentation words it in a less dangerous way, such that it sounds as if it is a UX feature instead of a performance thing. I'd stay away from that know.

Stefan

Jacob Keller· Jan 30, 2018, 23:38 UTC · re: Stefan Beller · lore

Re: git add --all does not respect submodule.<name>.ignore

On Tue, Jan 30, 2018 at 1:56 PM, Stefan Beller <sbeller@google.com> wrote:

Show 6 quoted lines

> The assume-unchanged bit is a performance optimisation for powerusers,
> but its documentation words it in a less dangerous way, such that it sounds
> as if it is a UX feature instead of a performance thing. I'd stay away from
> that know.
>
> Stefan

In almost all cases where someone incorrectly recommends assume-unchanged, I think they could just use skip-worktree to get the actual behavior they want.

Thanks, Jake