I just analyzed the FTBFS of libpfm4 in unstable/i386:
-make: Entering directory '/build/libpfm4-4.7.0'
-Compiling for 'i386' target
+make: Verzeichnis „/build/libpfm4-4.7.0“ wird betreten
+Compiling for 'x86_64' target
Compiling for 'Linux' system
The build system uses 'uname -m' to determine which CPUs are the target,
and obviously in the i386 chroot for the second build this returns
'x86_64'. I had the same problem when building in i386 pbuilder on an
amd64 host, but explicitly running it under 'linux32' fixed this for me.
Using 'uname -m' might not have been the best choice for upstream to
detect the architecture ...
This was only a problem in the second build, so that pbuilder seems to
configured differently than the first one.
The buildds obviously don't show this problem, so seem to run under
linux32 on an amd64 host.
Reproducible-builds mailing list