Force shallow clone of json 3-rd party

[Nir-Az]

CMake ignore the shallow clone, json history is huge (185 [MB] when the repo itself is ~7 [MB]) and take much time to download.
Jenkins fail on that many times on timeout.
This fix it.

Before:

After:

Also notified json on GH
nlohmann/json#4370

[maloel]

CMake ignore the shallow clone

Source? It's supposed to work as of v3.6, no?

[maloel]

I think better to keep the quotes in case the binary dir has spaces in it - though I'm not sure it matters.

[Nir-Az]

Should I try or keep it as is?

[maloel]

I think quotes are better

[Nir-Az]

Done

[Nir-Az]

CMake ignore the shallow clone

Source? It's supposed to work as of v3.6, no?

Its not working , see LibCI which use > 3.20
Tried it also locally

[maloel]

CMake ignore the shallow clone

Source? It's supposed to work as of v3.6, no?

Its not working , see LibCI which use > 3.20 Tried it also locally

That's not in doubt -- I believe you.
But I'd still like to see a source that says this is a known issue and how to get around it...

[Nir-Az]

CMake ignore the shallow clone

Source? It's supposed to work as of v3.6, no?

Its not working , see LibCI which use > 3.20 Tried it also locally

That's not in doubt -- I believe you. But I'd still like to see a source that says this is a known issue and how to get around it...

https://gitlab.kitware.com/cmake/cmake/-/issues/17770

[maloel]

https://gitlab.kitware.com/cmake/cmake/-/issues/17770

Great, thanks. It says:

For some git versions [>=1.7.10] the --no-single-branch option is passed, and this option will cause all branch HEAD's to be downloaded, instead of just the branch/tag/commit that one wanted to download

So we're trying to take out this flag.

Couple of things:

• Sounds like we need a general function git_shallow_clone that we would use here and in many other places, right? This is not a weird repo where this problem exists: someone else may as well complain to us that our repo is too big. Can we refactor to an actual function that we document and call instead of replicating this change everywhere? (possibly in another PR, up to you)
• This may be a problem with ExternalProject_Add, but another option would be to use FetchContent, no?
  • We haven't so far because it's not available before CMake 3.11, but now that we're removing U18, our minimum CMake version can be increased too?
  • Maybe FetchContent is the reason this was never fixed it in ExternalProject_Add?

Jenkins fail on that many times on timeout.

I have yet to see this... but agree that this is better if it indeed helps.

[Nir-Az]

I agree it will be helpful to deep dive on this, I will open a ticket for it.
I suggest we merge this one as it's causing many issues and try to work on a more generic approach.

[Nir-Az]

https://gitlab.kitware.com/cmake/cmake/-/issues/17770

Great, thanks. It says:

For some git versions [>=1.7.10] the --no-single-branch option is passed, and this option will cause all branch HEAD's to be downloaded, instead of just the branch/tag/commit that one wanted to download

So we're trying to take out this flag.

Couple of things:

• Sounds like we need a general function git_shallow_clone that we would use here and in many other places, right? This is not a weird repo where this problem exists: someone else may as well complain to us that our repo is too big. Can we refactor to an actual function that we document and call instead of replicating this change everywhere? (possibly in another PR, up to you)

• This may be a problem with ExternalProject_Add, but another option would be to use FetchContent, no?

  • We haven't so far because it's not available before CMake 3.11, but now that we're removing U18, our minimum CMake version can be increased too?
  • Maybe FetchContent is the reason this was never fixed it in ExternalProject_Add?

Jenkins fail on that many times on timeout.

I have yet to see this... but agree that this is better if it indeed helps.

BTW, here you did a lot of work to make it run on cmake configure step (call it from a different file), it is also relevant for a full external project step which also compile (like in libcurl)
So a generic function is complicated to do for all cases.
We need to think about it

[maloel]

Please point to the article you pointed to in the PR

[maloel]

And also explain advice.detachedHead=false

[Nir-Az]

Done