On Sat, Dec 14, 2013 at 10:06:13PM +0100, Ludovic Court??s wrote: John Darrington skribis: > On Sat, Dec 14, 2013 at 07:19:03PM +0100, Ludovic Court??s wrote: > John Darrington skribis: > > > *gnu/packages/gnupg.scm (libgpg-error): Added gcc as a native input, > > and set the CC_FOR_BUILD variable. > > The implicit inputs when cross-compiling include both the cross-gcc and > the native gcc (see build-system/gnu.scm.) > > AFAICS setting CC_FOR_BUILD=gcc should be enough (it uses ???cc??? by > default.) Can you confirm. > > You are right. setting gcc as a native input seems to be unnecessary. Somewhat on > a tangent: > Would it not be a good idea to have CC_FOR_BUILD=gcc the default setting at least > fot the gnu build-system. Yeah, perhaps we could add that when cross building (we have only two occurrences currently, but it may be needed quite often.) Right. PSPP should have it, but does not currently. > > *gnu/packages/gnupg.scm (libgcrypt): Added libgpg-error as a native input and > > set the --with-gpg-error-prefix configure option. > > This is wrong: libgcrypt would end up linking against the native libgpg-error. > > I think you are mistaken. The purpose of --with-gpg-error-prefix= is to set the path > of the library which it should be linked against. The issue is, that libgpg-error provides > both a binary (which must be run at configure time) AND a library which has to be linked. Oh indeed, I was unaware of that. So that explains why libgpg-error must be passed both as a native and target input. However, is --with-gpg-error-prefix really needed? By default the cross gcc will find the target libgpg-error in its search path, and not the native one, so I would guess this is OK. That was not my experience. I will try it again, in case I missed something. J' -- PGP Public key ID: 1024D/2DE827B3 fingerprint = 8797 A26D 0854 2EAB 0285 A290 8A67 719C 2DE8 27B3 See http://sks-keyservers.net or any PGP keyserver for public key.