On 2013-Nov-17 16:32:46 -0500, Eitan Adler <li...@eitanadler.com> wrote: >On Wed, Nov 13, 2013 at 7:40 AM, Bernhard Fröhlich <de...@bluelife.at> wrote: >> I think the idea and approach is correct but I went for a slighly more >> conservative >> approach if _GCC_RUNTIME was not set which is true in case base gcc is used= > >This can not possibly be correct as the variable begins with a "_". >The initial proposed patch is also incorrect for a similar reason. >_GCC_RUNTIME is not valid for use in a ports Makefile.
I completely agree. As I initially stated, this is solely a work- around for a bug in lang/gcc*. In fact the existing use of _GCC_RUNTIME in bsd.gcc.mk is equally wrong and is also just a work- around for lang/gcc* bugs. >Perhaps it would help if USE_GCC or USES=compiler exposed a variable >indicating the runtime? This is not correct either. A port using lang/gcc should not need to know about the compiler internals. If code compiled with gcc46 needs to be linked against non-standard libraries then using gcc46 for linking should be sufficient - it shouldn't be necessary to add magic "-Wl,-rpath=..." options. (IMO, the whole -L/-rpath approach is broken: The runtime searchpath should default to the linktime searchpath. Unfortunately, fixing that isn't possible). -- Peter Jeremy
pgpG0tXPpCO8h.pgp
Description: PGP signature
