[lfs-dev] Test failures in current svn

Ken Moffat zarniwhoop at ntlworld.com
Thu May 22 15:49:49 PDT 2014


On Sun, May 18, 2014 at 09:15:02PM +0100, Ken Moffat wrote:
> On Sun, May 18, 2014 at 06:30:59PM +0200, Armin K. wrote:
> > On 05/18/2014 01:53 AM, Ken Moffat wrote:
> 
> > > 
> > > binutils :
> > > 
> > >  ld-plugin failures (do we really want to enable this ?)
> > > FAIL: PR ld/12758
> > > FAIL: PR ld/12760
> > > FAIL: LTO 3 symbol
> > > FAIL: PR ld/13183
> > > FAIL: LTO 3a
> > > FAIL: LTO 11
> > > 
> > 
> > gcc-4.9 issue I think. I believe archlinux has fix for this one.
> > 
> 
>  Seems likely. But we now enable the lto plugin for gcc.  I cannot
> tell if that is or is not wise until I can finish the system, but at
> the moment I feel (slightly) concerned.
> 

 Well, I've now completed my normal desktop (on x86_64) after
thinking that the build was dead because I couldn't build a kernel [
phenom issue, probably - noted in the other thread ] and everything
has built without issues.  Most of it isn't yet tested, and I've no
idea if anything I build actually _uses_ LTO, but at the moment it
all looks good.

> > > gcc :
> > > 
> > > FAIL: c-c++-common/tsan/thread_leak1.c
> > > (several times, for different optimizations)
> > > 
> > > FAIL: gcc.c-torture/compile/limits-exprparen.c
> > > (several times, for different optimizations, all of them ICEd)
> > > 
> > > FAIL: g++.dg/ipa/devirt-11.C
> > > (several times, for different values of -std=)
> > > 
> > 
> > I think latest patch modified something regarding ipa but I'm not sure.
> > Maybe that caused it?
> 
>  Could be.  Again, what will really matter is the results for the
> complete system.
> 
 Since everything built fine, I was probably worrying unnecessarily.

>  The two that most concern me are the new perl failures,
> particularly the one that now gets undef.
> 
 I still think that anything using real numbers in perl (if there is
such a thing!) will benefit from testing to see if it still works.
So, for a server I would be reluctant to try gcc-4.9 at the moment.
On my desktop, the only heavy perl user is get_iplayer (download
phone-size video from the BBC - for people in GB) and that is
working without any obvious errors.

 So for x86_64 desktop applications I think the build is ok, even if
the test results are not.  My "Norwegian Blue" build is back from
the dead.

ĸen
-- 
das eine Mal als Tragödie, dieses Mal als Farce


More information about the lfs-dev mailing list