On 29/04/18 01:57 PM, Xi Ruoyao wrote:
On 2018-04-29 12:32 -0400, Dennis Clarke wrote:
I know the subject line is not really accurate, could not be. Over the
long years of the gcc project there were always pages of various peoples
build results at https://gcc.gnu.org/gcc-X/buildstat.html for some value
of X. That started to die off back in 2015 or so.
>><snipage>
Dennis Clarke
I think they are manally updated. And, GCC 8.1 is not released yet. It's
a RC version.
Oh? The web site says otherwise https://gcc.gnu.org/ where to me I see
the words "Supported Releases" and that means released. Right?
I think the issue is: most people building and testing GCC are trying out
experimental version from SVN trunk.
A lot do. Some folks are only really interested in a release version
unless there is some interesting problem worth looking at :
libquadmath and quadmath.h do not exist on ppc64
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=85440
Actually that bug still says "resolved fixed" when really it isn't even
a valid bug. The code for libquadmath won't work on any architecture
that can not ( yet? ) support the binary128 floating point type in
hardware or software emulation. Whereas the POWER9 has hardware support
and I'd love to try out gcc 8 there, but it isn't released .. or is it?
LFS guys (like me) build GCC releases,
me too .. on a few odd ball systems and not always virtual machines.
but the test result in LFS chroot environment is not accurate (without GDB,
and we can't send the result because we don't have mail clients in chroot).
Even if you did .. the result ( if a valid release ) won't appear
anywhere but in the mail list. I used curl to fetch the results data
for the past few years and it is largely all experimental releases.
Maybe someone can create an automatic system to build and test GCC releases?
yikes. :-P
I think we still need a human to look over the results and then
determine to what degree the results are reasonably clean or just
another nightmare of fails in gfortran and g++ and go etc etc.
Example of a nightmare :
https://gcc.gnu.org/ml/gcc-testresults/2018-04/msg01589.html
Here is a release thing of beauty :
https://gcc.gnu.org/ml/gcc-testresults/2018-04/msg01455.html
I know that the testsuite for Perl tends to kick out a few lines at the
bottom that claim 99.8% passed or similar. The testsuite in gcc just
dumps out numbers that may leave one to always ask "why would anyone
trust this thing?"
=== g++ Summary ===
# of expected passes 105237
# of unexpected failures 105
# of expected failures 395
# of unsupported tests 4600
So what is that ? Good? Disaster? About 105 unexpected failures
means C++ can not be trusted on this platform? I am never too sure as
there here never been a perfect result and there may never be. An
automated test platform would have to report the above as :
rhel74$ echo "4k 1 105 105237 105 395 4600 +++ / - 100 * pq" | dc
99.9100
So pretty darn clean right? However a 0.09% failure in the wrong place
would be a catastrophic failure for code that needs to run inside a
Medtronic heart implant or some flight control systems or turbo fan fuel
pump sensors. Those things are most likely hand coded assembly and
tested to death. No pun intended.
Anyways, I was just wondering where GCC 8.1 is and it seems to be RSN.
Dennis
ps: see https://gcc.gnu.org/pub/gcc/snapshots/LATEST-8/
