A payload other than "cpio" makes rpm print an error much too late in
the installation process. This patch adds an earlier check. It also
prints a different message for delta-rpms (see SuSE's deltarpm
Created attachment 107058 [details]
I have no problem with an earlier payload check in principle,
and lib/rpminstall.c is about as early as it gets for the
There are several other pathways that will need the
check plugged however, the most important of which
is rpmtsAddInstallElement() in lib/depends.c, as that
performs similar "early" detection for bindings that
your patch does for the rpm CLI.
The other pathways are those used to install *.src.rpm.
Lots of this mess needs scrap and rewrite, lest the introduction
of a 4th (i.e. other than binary/src and now "drpm") flavor
of *.rpm cause this bug to be dicovered again and again and again.
Now that rpm-4.4.4 supports "ustar" payloads, this is not the right check to do.
The better implementation is to add a "rpmlib(PayloadIsDeltaRpm)" (or equivalent) tracking
dependency when building delta rpm headers, and then failing in rpmtsCheck() if the
version of rpmlib does not supprt the necessary functionality.
That's perfect for binary rpm's, but source rpm installs (are there delta rpm payloads
in source rpm's?) would still need additional handling because rpmtsCheck() (and
the install paths) are different.
I'm perfectly okay living without the check for srpm's because non-cpio payloads for
srpm's is plain and simply a bad idea.
WONTFIX in the sense that the included patch is the wrong check, but I'm perfectly happy
with a rpmlib(...) tracking dependency for delat rpm's underneath a #ifdef in rpm sources some day.