| Message ID | 20161009143249.GQ2052@waldemar-brodkorb.de |
|---|---|
| State | Not Applicable |
| Headers | show |
On Sun, Oct 9, 2016 at 4:32 PM, Waldemar Brodkorb <wbx@openadk.org> wrote: > Hi Thomas, > Thomas De Schampheleire wrote, > >> On Oct 9, 2016 15:05, "Thomas Petazzoni" <thomas.petazzoni@free-electrons.com> >> wrote: >> > >> > Hello, >> > >> > On Sun, 9 Oct 2016 13:55:52 +0200, Thomas Petazzoni wrote: >> > >> > > All three patches applied to master. Thanks! >> > >> > And I reverted the texinfo patch and the binutils patch. The binutils >> > patch badly breaks the build of the toolchain. See: >> > >> > http://autobuild.buildroot.net/results/90d/ >> 90d0537bc12cbb2a7c6d344b3c6c9e1a5046d791/build-end.log >> > http://autobuild.buildroot.net/results/9a2/ >> 9a2b30bfd08c3a4af43f0523276eb5f799699aaa/build-end.log >> > http://autobuild.buildroot.net/results/d8a/ >> d8a4afeeec8d6b070b17a067d0ded5aefe4e3f6a/build-end.log >> > >> > Please do a toolchain build before sending the next iterations of the >> > patches. A basic ARM toolchain would not even build with your >> > patches :-/ >> >> Will check, sorry... > > Binutils isn't installed into output/host. > May be try something like this: > diff --git a/package/binutils/binutils.mk > b/package/binutils/binutils.mk > index 4e87fcb..8aa3fc7 100644 > --- a/package/binutils/binutils.mk > +++ b/package/binutils/binutils.mk > @@ -59,10 +59,8 @@ endif > # and sometimes needs specific makeinfo versions to work > BINUTILS_CONF_ENV += MAKEINFO=true > BINUTILS_MAKE_OPTS += MAKEINFO=true > -BINUTILS_INSTALL_TARGET_OPTS += MAKEINFO=true > HOST_BINUTILS_CONF_ENV += MAKEINFO=true > HOST_BINUTILS_MAKE_OPTS += MAKEINFO=true > -HOST_BINUTILS_INSTALL_OPTS += MAKEINFO=true No, I'm sure that at least in some combinations, the install command needs to get the makeinfo override. I have compared the logs and results now, and found the issue. It seems that for FOO_INSTALL_{TARGET_,STAGING_,}OPTS there is a default applied that already includes 'install'. The += does not actually append because it is evaluated before the defaults in package/pkg-autotools.mk. So, without my patch the install command essentially was 'make install', but with my patch it became 'make MAKEINFO=true' without target. The silly thing is, I actually did see this failure during my testing, but dismissed it too quickly, thinking it surely could not have been caused by my changes, and would 'probably' be caused by some funny state I left my repo in during my testing. [bumps head into wall] I'll do some further testing first ;-) and then resend the patches. Thanks, Thomas
diff --git a/package/binutils/binutils.mk b/package/binutils/binutils.mk index 4e87fcb..8aa3fc7 100644 --- a/package/binutils/binutils.mk +++ b/package/binutils/binutils.mk @@ -59,10 +59,8 @@ endif # and sometimes needs specific makeinfo versions to work BINUTILS_CONF_ENV += MAKEINFO=true BINUTILS_MAKE_OPTS += MAKEINFO=true -BINUTILS_INSTALL_TARGET_OPTS += MAKEINFO=true HOST_BINUTILS_CONF_ENV += MAKEINFO=true HOST_BINUTILS_MAKE_OPTS += MAKEINFO=true -HOST_BINUTILS_INSTALL_OPTS += MAKEINFO=true # gcc bug with Os/O2/O3, PR77311 # error: unable to find a register to spill in class 'CCREGS'