Package: g++-4.4
Version: 4.4.4-8
(Leaving choice of severity to gcc maintainers; at the current stage ppl cannot
migrate to testing although it would fix an RC bug.)
ppl 0.10.2-7 FTBS on armel [1] (exclusively) with test suite failures; this was
extremely surprising as 0.10.2-6 was fine [2] and there were no code changes.
Still, the failure turned out to be reproducible on abel.d.o. Looking at one of
the failures (tests/Octagonal_Shape/upperboundifexact1.cc, disabled all but
test02, test03, test04, using cmdline [3]) yields the following interesting
result:
- Changing optimization to -O0 makes things work.
- Using -O1 or -O2 causes the failure.
- Using -O3 makes things work again.
- Using all the -f... options that "g++ -O1 -Q --help=optimizers" claims to be
enabled at -O1 does not induce the failure. At this point I feel unable to
debug this any further because I do not know which more fine-grained options
to add to see which (set of) optimization triggers the problem.
I might want to add that 0.10.2-6 was built using g++-4.4 4.4.3-1 (successful),
0.10.2-7 failed with g++-4.4 4.4.4-8.
Using the command line as shown in [3] the error is pretty easy to see; correct
and stated in the test suite is
=== test02 ===
*** octs1 ***
A >= -2, A <= -1, B >= 0, B <= 2
*** octs2 ***
A >= 1, A <= 2, B >= 0, B <= 2
but we get
=== test02 ===
*** octs1 ***
A >= -2, A <= 4294967295, B >= 0, B <= 2
*** octs2 ***
A >= -4294967295, A <= 2, B >= 0, B <= 2
Please let me know if there's anything else I can do; the complete sources are
still there on abel.d.o.
Best regards,
Michael
[1] https://buildd.debian.org/fetch.cgi?&pkg=ppl&ver=0.10.2-7&arch=armel&stamp=1280947057&file=log
[2] https://buildd.debian.org/fetch.cgi?&pkg=ppl&ver=0.10.2-6&arch=armel&stamp=1267963469&file=log
[3] touch upperboundifexact1.cc ; make upperboundifexact1 \
TEST_CPPFLAGS="-DOCTAGONAL_SHAPE_INSTANCE=int64_t -DNOISY -DVERY_NOISY" \
CXXFLAGS="-frounding-math -g -O2" ; PPL_NOISY_TESTS=1 ./upperboundifexact1
Attachment:
pgp68lk5jJzPS.pgp
Description: PGP signature