pmbootstrap doesn't figure out dependencies right when compiling KDE

Created by: ollieparanoid

@PureTryOut reported in IRC:

if I tell it to compile a package which has deps on other packages that aren't compiled yet, it'll install all the deps for all packages combined before starting to compile the first in the list, even with --strict

Created by: PureTryOut

What originally happened (installing too many deps) does not seem to happen anymore. The chroots get zapped properly before building a new package.

However, sometimes it can not find a dependency if it has not been compiled yet. I just had a case of this again.

I tried building plasma-settings which has a dependency on kirigami2, neither of them had been compiled (for this architecture) before. Instead of first compiling kirigami2, it just errored out.

>>> plasma-settings: Analyzing dependencies...
ERROR: unsatisfiable constraints:
  .makedepends-plasma-settings-0:
    masked in: cache
    satisfies:
               world[.makedepends-plasma-settings]
  kirigami2 (missing):
    required by:
>>> ERROR: plasma-settings: all failed

Note that kirigami2 in this case is a runtime dependency, not a build dependency. Maybe that makes a difference?

Eduit: Ok further testing: it only happens with runtime dependencies. All other dependencies get build with --strict properly.

By * postmarketOS Bot on 2017-11-04T14:14:01

Created by: ollieparanoid

Thanks for the detailed information, that makes a lot of sense. We actually only care about the makedepends in the usual, non-strict mode, as a performance hack. That way, you can build device- packages without building the actual linux packages. But for the strict-mode, we should really build the real depends, too.

I've quickly written down a patch, which should fix this in theory. Could you try if that works for you?

--- a/pmb/build/package.py
+++ b/pmb/build/package.py
@@ -58,12 +58,11 @@ def package(args, pkgname, carch, force=False, buildinfo=False, strict=False):

     # Initialize build environment, install/build makedepends
     pmb.build.init(args, suffix)
-    if len(apkbuild["makedepends"]):
-        if strict:
-            for makedepend in apkbuild["makedepends"]:
-                package(args, makedepend, carch_buildenv, strict=True)
-        else:
-            pmb.chroot.apk.install(args, apkbuild["makedepends"], suffix)
+    if strict:
+        for depend in apkbuild["makedepends"] + apkbuild["depends"]:
+            package(args, depend, carch_buildenv, strict=True)
+    else if len(apkbuild["makedepends"]):
+        pmb.chroot.apk.install(args, apkbuild["makedepends"], suffix)
     if cross:
         pmb.chroot.apk.install(args, ["gcc-" + carch_buildenv,
                                       "g++-" + carch_buildenv,

Also I've noted that this behavior wasn't mentioned in build internals so I'll document it there now.

By * postmarketOS Bot on 2017-11-04T15:27:44

Created by: ollieparanoid

In my opinion, this is a blocker for merging your plasma work (#440). So let's look at it again.

Summary

abuild considers all depends as build-time dependencies. pmbootstrap only considers makedepends as build-time dependencies. This is a performance hack, and allows fast building of device-* packages without building the kernel it depends on.

Plan to resolve this

I think to fix this incompatibility with Alpine's packaging, we should adopt to how Alpine does it by default. To preserve the performance hack (which is quite useful for mass-editing and building device-* packages), I would put it behind a flag:

pmbootstrap build -i / pmbootstrap build --ignore-depends

What I'm saying is, that it should also work that way even if --strict is not defined. It would still make sense to have --strict to start with a clean chroot and only install the depends, which are required (like it works now).

As always, feedback is welcome.

By * postmarketOS Bot on 2017-11-21T17:27:27