Re: [Valgrind-developers] Linux/TileGX port: last call for comments

[Philippe W. <phi...@sk...>]

On Fri, 2015-04-10 at 02:30 +0000, Zhigang Liu wrote:
> [Zhigang] The regress test shows 20+ failures as post in my previous email today.
> Most of them are due to the mismatch source file line # given by stack trace.
> The mismatched line number are very close to expected within +/-3.
> Apparently, we need improve the stack back trace later.
You might compare the stacktraces produced by Valgrind and
the stacktraces produced by gdb (using V gdbserver).
This might indicate if the problem lies in the Valgrind unwinder or
just in the way the compiler generates unwind info.

> Regarding the maintaining: we can periodically(say once a week) pull the trunk and
>  run on TileGX linux machine and post the results on this mail-list.
There is a standard script that is used by all nightly builds to build,
compare and send the results to the valdev result mailing lists.
The idea is to run that nightly (so as to detect problems/regressions)
asap.
If your company can donate a machine to gcc compile farm, the nightly
build can be setup there.
Otherwise, a cronjob starting the build in a quiet period is the best
thing to do.

Philippe