Arch Linux User Repository

I think I was confused by the kwinft package for Manjaro actually being the equivalent of kwinft-git. I had thought limiting to a branch would still apply minor subversion fixes, but that's not the case.

Saw your recent update. Tarballs are the preferred source for stable packages. I had assumed you wanted or needed to use the git repo, in which case tag or commit would have been acceptable. Selecting by branch changes the package type.

The difference is you are specifying a branch, while wrapland-git uses the default master branch. However, even doing it this way, you should still have a -git suffix and a pkgver() function. If you don't want to do that, you should pin to a specific tag or commit. As long as upstream is consistent with tagging, tags are easier to maintain than commits.

I appreciate you providing script snippets; I'm not good with this (the package started out as just a simple installer for the release tarballs and only recently morphed into what it is). Before I make changes, could you take a look at commit history and confirm that what I am doing now is truly different from what wrapland-git does? I think the recent changes to the package might have been fundamentally misguided.

To make this not a git package, you should use the tag or commit.

_tag="wrapland@0.527.0"
_commit="de7685a95cabeac60b65b17cf5389aed254321c0"
source=(
  "git+...#tag=$_tag"
  "git+...#commit=$_commit"  
)

To keep version numbers similar to what you've been using:

pkgver="5.27.0"
_tag="wrapland@0.${pkgver/.}"

I'm referring to a pkgver() function to produce version strings like 5.27.0.r0.gde7685a9, as described in VCS package guidelines.

pkgver() {
  cd "wrapland"
  git describe --long --tags \
    | sed -E 's/^[^0-9]+0\.([0-9])/\1\./;s/([^-]*-g)/r\1/;s/-/./g'}
}

Since you're checking out branches and not tagged releases, this really is a git package and should be renamed.

It looks like they use 5.27 for the branches and 0.527.0 for who knows what. Doesnt really matter to me which you use. It's the .r#.g# part that I was referring to: 5.27.0.r0.gde7685a9

Thanks for correcting my mistake, qt5-tools is a dependency again. I'll also remove the prepare() portion and replace with the branch/tag in URL. As for the pkgver(), are you asking why wrapland and disman are not pkgver(0.527.0) instead of pkgver(5.27)? I actually don't remember why I changed it in the past, but I'll revert it if it's confusing. I'll push the update as soon as you respond.

1. The qt5- depends are not optional. See error message in earlier comment. You should be testing your packages by building in a clean chroot to ensure your package will build on machines other than yours.

2. You shouldn't be putting packages in dependency lists just for your "convenience".

3. Some git operations are acceptable, but switching branches / tags / commits is standard to put in source. Following the convention makes your intent clear. Not doing so makes your package more difficult to review. You still need a proper pkgver() function when using branches to differentiate revisions.

AFAIK qt5/6-tools is not even a required dependency, and it's been moved to optional dependency status. The switch to qt6-* packages was just for my own convenience, since I have replaced every qt5-* packages that aren't dependencies.

As for the git checkout, since kwinft has not received a release update for some time, I thought doing a git checkout would allow the minor commits to be built so that kwinft coincides better with kwin minor version updates. KInfoCenter thinks I'm running KWin 5.27.9. I would think that changing the git URL to include #branch=Plasma/5.27 would do the same thing, but is using git checkout some faux pas that I'm not aware of?

kwayland5 is still based on qt5. So uses qt5-tool not qt6-tools. This (and other kwinft-related packages) builds fine when the appropriate qt5-* package is used.

I didn't notice that you're doing a git checkout when I opened the deletion requests. Please explain why you aren't using #branch= in the source array, or update to do so. Then I'll close the deletion requests. I think a proper pkgver() function may still be needed though.