A little off-topic, but does anyone have any recommendations for methods for managing 10s of git repos as various dependencies for wildly varying projects as opposed to this sort of mono-repo approach? I've seen some organizations try to use ivy, but never in a robust way.
I think they have a similar goal. The difference sounds like gta uses git for change detection while bazel uses hashes to detect changes (hashes include the hashes of the transitive closure of the deps; it’s a dag). The latter is more robust in the face of noop edits that don’t change the output as well as distributed caching of test/build results.
Once you realise that the reason you can't find the critical production code that's fritzing is because it's in another goddamn repo you didn't know about, you become a fan.
Having to reverse engineer CI/CD (or find someone who used to work on a project) to find where something came from is nuts. It should be as easy as search.
IMO the solution is git submodules.
The problem with submodules mostly is the horrible workflow and the missing tool integration.
Varying dependencies could result in separate parent repos that contain different subsets of submodules.
A simple script can fix tooling: first commit the submodule changes then create the parent repo commit.
You don't want to create a commit in the parent repo all the time. If you fix a typo in a comment in a submodule, why commit anything to parent? Anyways, which parent, since there can be many. There should be more than one. Otherwise why use a submodule?
Another problem is that one forgets to update submodules when switching branches. Afaik git does not do this because of possible data loss in submodules?
I like submodules in general. They are great for dependencies where you want to follow more closely than release version. Still, it would be even easier in a monorepo. The main reason is probably that every anti-dependency is visible.
