I have successfully reproduced the problem.  Reproducer:

download oe-core/bitbake
download meta-openembedded

configure a new build
edit conf/bblayers.conf -- add meta-openembedded/meta-oe

bitbake ncurses
bitbake uhd

(uhd has a minor bug in it's dependencies, it requires ncurses but does not yet state it.)

Result of the above matches Philip's comments.

Contents of packages-split/uhd-dev/usr/lib/pkgconfig/uhd.pc:


prefix=/usr
exec_prefix=${prefix}
libdir=${exec_prefix}/lib
includedir=${prefix}/include

Name: UHD
Description: Ettus Research - USRP Hardware Driver
URL: http://code.ettus.com/redmine/ettus/projects/uhd/wiki
Version: 003.005.000-0-g5cb9779d
Requires:
Requires.private: 
Conflicts:
Cflags: -I${includedir} 
Libs: -L${libdir} -luhd
Libs.private: 


---

This causes the pkgconfig dependency code to pull out a dep on "Version" 003.005.000-0-g5cb9779d of the "pkgconfig(uhd)".  RPM is assuming the versioning should be handled by standard rules, which are "version-release", where "-" is not supported in either version or release.  Thus three '-' means it can't figure out which is version and which is release, thus the error is generated.

Still don't have a solution, but I can explain the fault at least.

Probably the most general/simplest fix would
be to add
    ... | tr -- - _
(or equivalent) to the script that extracts the version
from the *.pc file.

The file that needs changing is pkgconfigdeps.sh,
the version is being extracted by these 2 lines

$pkgconfig --print-provides "$filename" 2> /dev/null | while read n r v ; do
...
$pkgconfig --print-requires "$filename" 2> /dev/null | while read n r v ; do

and you would need to split out the EVR field and change
all '-' dash characters to '_' underscore characters in the EVR field.

I received this in the email also:


--- host/cmake/Modules/UHDVersion.cmake	2013-01-24 16:11:20.956506736 +0100
+++ host.b/cmake/Modules/UHDVersion.cmake	2013-01-24 17:11:22.966407775 +0100
@@ -69,7 +69,7 @@ ENDIF()
 
 ########################################################################
 IF(UHD_RELEASE_MODE STREQUAL "release")
-    SET(UHD_VERSION "${UHD_VERSION_MAJOR}.${UHD_VERSION_MINOR}.${UHD_VERSION_PATCH}-${UHD_GIT_HASH}")
+    SET(UHD_VERSION "${UHD_VERSION_MAJOR}.${UHD_VERSION_MINOR}.${UHD_VERSION_PATCH}+${UHD_GIT_HASH}")
 ELSE()
-    SET(UHD_VERSION "${UHD_VERSION_MAJOR}.${UHD_VERSION_MINOR}.${UHD_VERSION_PATCH}-${UHD_GIT_COUNT}-${UHD_GIT_HASH}")
+    SET(UHD_VERSION "${UHD_VERSION_MAJOR}.${UHD_VERSION_MINOR}.${UHD_VERSION_PATCH}+${UHD_GIT_COUNT}+${UHD_GIT_HASH}")
 ENDIF()


This patches uhd to avoid the problem. I'll discuss with the UHD maintainer.

Actually all the splitting is done: just wrap
a back ticked sed/tr onto the $v field in the
two places that look like
    echo "pkgconfig($n) $r $v"

Yes changing the Version: string also Just Works
(but doesn't solve the problem generally and is
out-of-scope as an RPM "fix")

The original problem was resolved by a change to UHD.  The underlying issue still needs to be resolved via a patch to the pkgconfig scanning code in rpmdeps.

We will work on this for the 1.5 release.  The workaround should be good enough for 1.4, and the issue is fairly rare.

Not deployment, bug in RPM integration.

We've had no more reports of this.. but it should be fixed.  Moving it forward to 1.7.

Alex: Could you check if we have this issue with rpmdeps in rpm v4 please?

I have baked a version with two dashes into sqlite3.pc.in file, and rpm4 had no problem with that (maybe it doesn't look into .pc files at all anymore?). I'd say RESOLVED/OBSOLETE.