On 19/01/2017 11:20, James Cowgill wrote: > Hi, > > On 18/01/17 22:39, Nico Schlömer wrote: >> I'm co-maintaining the Trilinos package [1] in Debian and recently found >> a bunch of new lintian warnings of the kind >> binary-or-shlib-defines-rpath [2]. It say in the description of the warning: >> ``` >> To fix this problem, look for link lines like: >> >> gcc test.o -o test -Wl,--rpath,/usr/local/lib >> >> or >> >> gcc test.o -o test -R/usr/local/lib >> >> and remove the -Wl,--rpath or -R argument. >> ``` >> Indeed, the Trilinos installation contains many of those lines, but they >> are necessary too. When executing the test binaries (which are compiled >> in the build tree alongside the libraries), they have to find the linked >> shared libraries. Messing with the rpath is necessary. >> >> That's not true later on when the libraries are _installed_, of course. >> For this, CMake has the switch CMAKE_SKIP_INSTALL_RPATH [3], which >> serves exactly that purpose. For some reason, lintian doesn't seem to be >> happy with that though. > The CMAKE_SKIP_INSTALL_RPATH option only affects the rpaths which CMake > itself inserts. It does not affect any rpaths manually added with other > -Wl,--rpath options. The culprit here is probably openmpi which adds all > of these rpath entries: > > $ mpicc -show > [...] -Wl,-rpath -Wl,/usr//lib -Wl,-rpath > -Wl,/usr/lib/x86_64-linux-gnu/openmpi/lib [...] > > Maybe it shouldn't do that. The /usr//lib one is clearly useless and the > it seems that most of the libraries from > /usr/lib/x86_64-linux-gnu/openmpi/lib are symlinked into > /usr/lib/x86_64-linux-gnu anyway. > Agreed. Will remove these on the next upload.
Best regards Alastair -- Alastair McKinstry, <alast...@sceal.ie>, <mckins...@debian.org>, https://diaspora.sceal.ie/u/amckinstry Misentropy: doubting that the Universe is becoming more disordered.
signature.asc
Description: OpenPGP digital signature
