[RFA:] work around dejagnu bug for simulator targets
Benjamin Kosnik
bkoz@redhat.com
Tue Feb 12 19:46:00 GMT 2008
> There was a weird discrepancy between cris-elf libstdc++
> testsuite runs on Debian systems and Fedora systems, with 59
> extra failures on Fedora systems. After some digging, I found
> it was due to a bugfix imported into the debian dejagnu package
> but missing in the Fedora package and pristine dejagnu-1.4.4:
> <http://lists.gnu.org/archive/html/dejagnu/2004-06/msg00008.html>.
> The bug caused input files to be missing, which trigged aborts
> in the test-cases, looking like "real" failures. (Meh, not
> really graceful exit behavior there, but once you know the
> reason, I guess you just fix it and/or don't care.)
Thanks for looking into this, and more importantly, fixing it.
> There are signs that there'll be a dejagnu-1.5 release perhaps
> this year (just trying to curb your enthusiasm), but I guess
> we'll have to support dejagnu-1.4.4 for some time after that.
>
> This patch, the file dejapatches.exp, also tries to establish a
> general habitat for current and future gcc-local bugfixes and
> overrides (grep for 'delete.*after next [Dd]eja[Gg]nu release'
> in gcc/testsuite/lib and you might see a need). As this is
> currently only needed for libstdc++, I didn't touch the other
> config/default.exp's.
I think you fixed this correctly. Nice job on that. I like this way of
handling divergence from upstream.
If other libraries need this, they can just use the dejapatches.exp
file.
> I tested this against Debian etch, making sure the installed
> sim_download was used and on Fedora 8, making sure the fallback
> was used. Both setups now exhibit the same results for
> libstdc++-v3.
Great.
> Incidentally, v3-copy-files should use remote_upload (to
> target), not remote_download (from target). You almost can't
> tell the difference, though. :)
Please fix this too, as long as you noticed it.
> gcc/testsuite:
> * lib/dejapatches.exp: New file.
>
> libstdc++:
> * testsuite/config/default.exp: Load gcc dejapatches.exp.
This is fine by me (Paolo already ok'd, but I don't see it applied.)
Janis?
-benjamin
More information about the Libstdc++
mailing list