I narrowed the problem down to qt5-declarative 5.15.8+kde+r24-2.1. Upgrading to 5.15.9 (in testing) solves the issue.

Some qt5/qt6 messup probably.

I start to suspect qt6-translations which has not been rebuild, it is a 'any' package, but
if it ends up on a i486 build machine, it will fail (as Qt6 fails completely currently
on i486). We must find a way to force any packages to the pentium4 subarchitecture, as
this is the one expected to have most prerequisites available to build the package.

It's ok on testing.

The message on stable is:

Cannot mix incompatible Qt library (5.15.8) with this library (5.15.9)

I'll push qt5 to stable..

fixed on i686, for some reason pentium4 needs to have some qt5 rebuilt..

cool, qt5-webengine ends in a one definition rule problem:

release/../../../../qtwebengine/src/3rdparty/chromium/base/memory/ref_counted.h:145:19: warning: type ‘struct RefCountedThreadSafeBase’ violates the C++ One Definition Rule [-Wodr]
  145 | class BASE_EXPORT RefCountedThreadSafeBase {
      |                   ^

...

/usr/src/debug/qt5-webengine/qtwebengine/src/core/api/qwebenginecallback_p.h:136:12: note: the conflicting type defined in another translation unit has virtual table with more entries
during RTL pass: pro_and_epilogue
release/gen/services/network/public/mojom/content_security_policy.mojom.h: In member function ‘Clone’:
release/gen/services/network/public/mojom/content_security_policy.mojom.h:1666:1: internal compiler error: in find_oldest_value_reg, at regcprop.cc:460
 1666 | }
      | ^
Please submit a full bug report, with preprocessed source (by using -freport-bug).
See <https://bugs.archlinux.org/>

aha, this one I hit before: FS#282

Broken on pentium4 for me still. Several months later.