there can silently exist package order dependencies between package functions

I find it hilarious that Dave Reisner points this out, because it is his packages that are full of this problem: one package function putting files in place for another, such that all the functions have to be done, and be done in the order he expects them to be done or packging will fail. Anyway it looks like the issue is settled... On Tue, Sep 22, 2015 at 11:57 PM, Que Quotion wrote:

Case in point, I had some trouble pakaging several packages today.

One was libsystemd, which I've already submitted a patch for. https://bugs.archlinux.org/task/46382

Another was libutil-lnux, which will need a little more help. https://bugs.archlinux.org/task/46383

In both cases $pkgname was used where $pkgbase should be (fixed in an earlier report for systemd) and then both packages use one package_something() function to put files in place for another package_somethingelse() function--both by moving files from $pkgdir into $srcdir. This seems like bad practice to me, isn't necessary, and could be done better: My proposal for systemd shortens the PKGBUILD by a few lines and doesn't touch $srcdir.

Removing the "--pkg" feature of makepkg will only allow more bad practice that will be harder to undo should we change our minds down the line.

On Tue, Sep 22, 2015 at 11:40 PM, Que Quotion wrote:

...

In other words, rather than fix the packages to be safer and more logical, go ahead and let people write sloppy PKGBUILDs?

For what it's worth, I think this is a bad idea.

The changes needed to fix the packages are small, and this would reduce the ability of users to automate upgrading from ABS.

On Tue, Sep 22, 2015 at 11:33 PM, Allan McRae wrote:

...

On 23/09/15 00:20, Pierre Neidhardt wrote:

...

I might have missed something in the previous messages, but is there any rationale on this, beside debatable worthiness?

Many packages break when using this.

Some are because assumptions made on build() about the value of $pkgname which changes when --pkg was used. These should use $pkgbase anyway...

Also, there are quite a few split packages that rely on the first first split-package being packaged for the rest of the package functions to work. This will become more prevalent when a way to do a "make install" followed by moving files into different packages becomes available (i.e. rpm packaging style - I am working on this).

So... essentially, it is (debatably) broken and will become more broken in the future