Exclude files from ninja dist?

[infirit]

Looking into meson bu I could not find anything in the docs how to exclude files from ninja dist. I don't want to have my transiflex or travis configuration files included in the tarball.

[jpakkane]

There is a plan to be able to run arbitrary scripts before the tarfile is created. This would allow arbitrary post-processing (such as adding extra sample files etc). It's not yet implemented, though...

[QuLogic]

If dist used git archive (like hg), then it could get export-ignore done for "free".

[infirit]

Hmm, not looking forward to use post-processing for something like this. Personally I would expect a modern build system to have this integrated into it's core. I'll experiment with my own dist script or just look for something else.

Anyway, either close this or leave it open as a feature request.

[infirit]

@QuLogic yes git archive with export-ignore works pretty good. Thx

[Kekun]

This is a duplicate of #2287.

[infirit]

@Kekun That is a vcs specific workaround for this problem. Mercurial can be configured to exclude files in its .hgrc file. But they are workarounds and this request is to integrate this into meson independent of the vcs used.

Currently git and mercurial are supported but there are hundreds of vcs, some have features like gitattributes some don't but is meson going to support all them? It becomes quite unworkable quickly and I doubt they are going to do that. Which means they will be excluding many users that have to do their own release script.

[thmo]

There is a plan to be able to run arbitrary scripts before the tarfile is created. This would allow arbitrary post-processing (such as adding extra sample files etc). It's not yet implemented, though...

This has been implemented in the meantime. However unfortunately run_dist_scripts() is called after del_gitfiles() (which itself is called in git_clone()), hence a custom dist script cannot rely on the export-ignore attribute, as .gitattributes is not available at this time anymore.

[jpakkane]

#6491 allows you to access the source dir from your dist script. Is that sufficient or do you need something more?

[thmo]

The source dir itself might be modified, so the .gitattributes from there might be different. Hence one would have to again use git to get the version that has been deleted just before. To me, this looks very hackish and complicated.

Why not provide a hook that is called just before running del_gitfiles()?

[nirbheek]

The source dir itself might be modified

dist clones HEAD, so modifications are explicitly ignored. The idea is that dist is for testing the git state, not the dirty tree. We should definitely warn when that happens, though.

[thmo]

dist clones HEAD, so modifications are explicitly ignored. The idea is that dist is for testing the git state, not the dirty tree.

I know, and that's why I basically said that the possibility to access the source dir is not enough.