bruce.dubbs at gmail.com
Sun Jun 26 14:58:55 PDT 2011
Matthew Burgess wrote:
> On 26/06/2011 20:09, Bruce Dubbs wrote:
>> I have found that if I change this to
>> 1. sh Configure -des -Dprefix=/tools
>> 2. make
>> 3. cp -v perl cpan/podlators/pod2man /tools/bin
>> 4. mkdir -pv /tools/lib/perl5/5.14.1
>> 5. cp -Rv lib/* /tools/lib/perl5/5.14.1
>> then all packages in the book build without complaint. The build time
>> does not seem to change a lot.
>> The change in line 3 is due to the new distribution, but the first 2
>> lines are the more significant. We build the whole distro, but still do
>> not do a complete `make install'.
> Ah, thanks for that. My previous attempt with 5.14.0 made the same
> changes to 1 & 3, but didn't change 2 which is why it obviously failed
> for me. I don't see any problems with the proposed instructions; I
> assume you did a full chapter 6 test suite run to verify it? I think
> it's Glibc & Coreutils that are particularly picky about the Perl
> installation, if I recall correctly.
You are right about that the reason for perl in Chapter 5 is Glibc and
Coreutils. The tests for Perl in Chapter 6 are run after both of those
and the Chapter 6 install instructions don't change. I still get "All
I do get some FAILs in coreutils:
I'm not sure why, but I don't think they are related to perl. I also
don't think they are significant.
For glibc, I get:
make: *** [/sources/glibc-build/rt/tst-cpuclock2.out] Error 1
make: *** [/sources/glibc-build/elf/tst-audit4.out] Error 132
make: *** [/sources/glibc-build/elf/tst-audit6.out] Error 132
make: *** [/sources/glibc-build/elf/tst-audit7.out] Error 132
Google says the audit tests are due to an optimization issue in gcc-4.6.
The fix is to add -mno-vzeroupper for those tests.
I don't think we need to address this.
More information about the lfs-dev