Hi Frank,
On 23 July 2012 22:47, Frank Ch. Eigler <fche@xxxxxxxxxx> wrote:
> Hi -
>
> nathans mentioned on IRC a configure/build failure on RHEL4 of the
> fche/prefix branch, from which he backported the gcc
> -fstack-protector-all / -fpie / etc. security-related patch (commit
> ba49d527).
>
> I reproduced the build failure on a rhel4 vm. It appears to be due to
> a too-old autoconf version there. However, a ../configure file
> subsituted from a more recent fedora version, the rhel4 configure &
> build completed successfully.
>
> I suspect this could be a cause for reevaluating the policy that
> .../configure is excluded from the git sources, and is required to be
> regenerated during build time. Some FOSS projects dogmatically
> exclude generated content from their source code repositories. Others
> don't mind the bit of git clutter, if it makes developers' lives
> easier. PCP may wish to try the latter.
At this stage for 3.5.5 I'm looking for something simple, and would prefer
to pull in those changes (gcc/ld options) if we can. Reworking configure
this way may well be a good thing (it'd simplify windows builds too, as I
recall fighting autoconf there as well) ... but a bit risky for a small bugfix
release.
Since AIUI this change is checking whether gcc supports some new flags
we might be able to achieve the same more simply by either avoiding the
use of cflags (not sure if that's an option), and/or doing a check more like
the pthread check (with the do_threads=check approach). Since it looks
like the problem is the CFLAGS tweak persists, when it shouldn't, this one
must be a fairly common problem.
If its all too problematic, we can pull that change from 3.6.5 and go with a
configure-included approach for the release that pulls in the rest of the
reworked configure scripts, perhaps.
Don't get too hung up on other RHEL4 fixes though - check with Mark, but
I think he stopped building there awhile ago for other reasons (?) - I might
be misremembering that.
Thanks.
--
Nathan
|