This is the mail archive of the
libstdc++@gcc.gnu.org
mailing list for the libstdc++ project.
Re: test suite run time
- From: Jonathan Wakely <jwakely dot gcc at gmail dot com>
- To: Tim Shen <timshen91 at gmail dot com>
- Cc: "libstdc++ at gcc dot gnu dot org" <libstdc++ at gcc dot gnu dot org>, Mike Stump <mikestump at comcast dot net>
- Date: Thu, 31 Oct 2013 16:05:25 +0000
- Subject: Re: test suite run time
- Authentication-results: sourceware.org; auth=none
- References: <CD5C2D71-BF36-401A-9224-8DDFBC1DAA07 at comcast dot net> <CAH6eHdRvsWtrkiWbcZqL72vmz+6-9XA61p-_rW+umuN7F6gF1A at mail dot gmail dot com>
On 2 October 2013 19:43, Jonathan Wakely wrote:
> On 17 September 2013 22:48, Mike Stump wrote:
>> I was doing a -j32 build and test suite run and noticed:
>>
>> normal7) \
>> dirs="`cd $$srcdir; echo 2[68]_*/*`";; \
>>
>> was the long leg.
>
> Yes, that's taking *significantly* longer than the other legs, we
> should rebalance them. I'll have a look at that some time.
The slow part of the normal7 target is the testsuite/28_regex
directory, which has lots of new tests in. I don't know why most
regex tests take several seconds to compile, is it just the number of
template instantiations or are we missing something from a precompiled
header, or should we have some explicit instantiations in the library?
A small improvement would be to move the 26_numerics tests to a
different target:
diff --git a/libstdc++-v3/testsuite/Makefile.am
b/libstdc++-v3/testsuite/Makefile.am
index 4c92fef..2f4158f 100644
--- a/libstdc++-v3/testsuite/Makefile.am
+++ b/libstdc++-v3/testsuite/Makefile.am
@@ -148,9 +148,9 @@ check-DEJAGNU $(check_DEJAGNU_normal_targets):
check-DEJAGNU%: site.exp
normal5) \
dirs="`cd $$srcdir; echo 23_*/[luv]*`";; \
normal6) \
- dirs="`cd $$srcdir; echo 2[459]_*/*`";; \
+ dirs="`cd $$srcdir; echo 2[4569]_*/*`";; \
normal7) \
- dirs="`cd $$srcdir; echo 2[68]_*/*`";; \
+ dirs="`cd $$srcdir; echo 28_*/*`";; \
normal8) \
dirs="`cd $$srcdir; echo 27_*/*`";; \
normal9) \
That still leaves normal7 as the slowest step, but only a couple of
minutes longer.
Should we try to do better with more careful juggling, or should we
try to speed up the regex tests?