A note on the current state of build-many-glibcs

A note on the current state of build-many-glibcs

[Zack Weinberg]

I think this is all known problems, but a reminder seems worthwhile:
Four of the targets built by default by Joseph's build-many-glibcs
script currently fail while bootstrapping the cross-compiler, which
means, if you're using b-m-g to test an invasive patch, you can't get
coverage for them:

hppa-linux-gnu
    busted pthreads (#21016)

m68k-linux-gnu-coldfire
    ICE in emit_library_call_value
    (https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68467)

microblaze-linux-gnu
microblazeel-linux-gnu
    link failure while building librt:
    libpthread.so: invalid string offset 3902075345 >= 4918 for
section `.dynstr'
    (something wrong with binutils?)
    (doesn't seem to be a reported bug?)

zw

Re: A note on the current state of build-many-glibcs

[Joseph Myers]

On Sun, 19 Feb 2017, Zack Weinberg wrote:

> microblaze-linux-gnu
> microblazeel-linux-gnu
>     link failure while building librt:
>     libpthread.so: invalid string offset 3902075345 >= 4918 for
> section `.dynstr'
>     (something wrong with binutils?)
>     (doesn't seem to be a reported bug?)

I don't see this.  With GCC 6 I see the compilers build fine including the 
first glibc 
<https://sourceware.org/ml/libc-testresults/2017-q1/msg00162.html> (and 
then the subsequent glibc build fail because of missing DWARF EH support 
before GCC 7).  With mainline of everything I see both compilers and glibc 
build fine (and then elf/check-execstack fails - recall that if the 
baseline involves any tests failing then a simple examination of PASS / 
FAIL results from build-many-glibcs.py won't tell you if there have been 
regressions within the testsuite that mean more tests now fail or fail to 
build).

Also in the testsuite, ia64 fails elf/check-execstack and all SH 
configurations have compilation failures in the testsuite with GCC 
mainline only (GCC bugs 78459 and 78460).

-- 
Joseph S. Myers
joseph@codesourcery.com

Re: A note on the current state of build-many-glibcs

[Zack Weinberg]

On Mon, Feb 20, 2017 at 11:09 AM, Joseph Myers <joseph@codesourcery.com> wrote:
> On Sun, 19 Feb 2017, Zack Weinberg wrote:
>
>> microblaze-linux-gnu
>> microblazeel-linux-gnu
>>     link failure while building librt:
>>     libpthread.so: invalid string offset 3902075345 >= 4918 for
>> section `.dynstr'
>>     (something wrong with binutils?)
>>     (doesn't seem to be a reported bug?)
>
> I don't see this.  With GCC 6 I see the compilers build fine including the
> first glibc
> <https://sourceware.org/ml/libc-testresults/2017-q1/msg00162.html> (and
> then the subsequent glibc build fail because of missing DWARF EH support
> before GCC 7).

That's bizarre.  I'm using all the script's defaults, so GCC 6 and
whatever the binutils default is.  The host platform is x86_64-linux,
Debian unstable - that _could_ make a difference but it seems like it
shouldn't.  The error message smells like a binutils bug...

> Also in the testsuite, ia64 fails elf/check-execstack and all SH
> configurations have compilation failures in the testsuite with GCC
> mainline only (GCC bugs 78459 and 78460).

I do also see the ia64 execstack failure, but not the SH failures
(since I'm using gcc6).

zw