Re: CVS commit: pkgsrc/lang/perl5

[Richard PALO <richard%netbsd.org@localhost>]

Le 26/06/14 13:31, Jonathan Perkin a écrit :
> * On 2014-05-10 at 14:39 BST, Richard PALO wrote:
>
> > Module Name:    pkgsrc
> > Committed By:   richard
> > Date:           Sat May 10 13:39:57 UTC 2014
> >
> > Modified Files:
> >         pkgsrc/lang/perl5: distinfo
> >         pkgsrc/lang/perl5/patches: patch-ca
> >
> > Log Message:
> > Put back -fstack-protector on solaris in perl after libtool fix.
>
> I'm not sure which libtool fix this is referring to, but the reason I
> disabled this in the first place is because it causes problems with
> packages which do not correctly link against -lssp.
>
> Re-introducing it has again caused packages to fail, you can see this
> most clearly in graphics/graphviz which is currently the 3rd top
> broken package:
>
>    
> http://us-east.manta.joyent.com/pkgsrc/public/reports/upstream-trunk32/20140625.2120/graphviz-2.38.0nb1/build.log
>
> It's too late to fix this now during the freeze as any change to perl
> would result in massive rebuilds, but it's something we need to do
> after the branch.  My personal preference is to just disable it, but
> again I'm not sure what the libtool fix is or whether it's supposed to
> be handling this.

The libtool fix was to:
1. allow -fstack-protector* passthrough for the link options
        (reverted by joerg@, BTW I asked him again recently about this)
2. remove -nostdlib for g++ on SunOS to allow (1) work for g++
   in addition to gcc (reverted by joerg@, then later recommitted!).
3. disable link time library position optimization for SunOS in order
   to support -Bdirect issue on x86_64 (see revert in 1)

There has been plenty of time to resolve this collaboratively.

I believe simply the patch-ac fragment that was reverted will resolve 
this crippling of gcc/g++ on platforms where the stack protector canary 
isn't built directly in to (g)libc (which is why gcc provides it in the 
first place!)