New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
NetBSD: multiple test failures with '-Duselongdouble' #17134
Comments
From @jkeenanIf you never test for something, you'll never know where some failures That's what we're learning as we test more varied configurations on more When we build perl with '-Duselongdouble' on NetBSD, perl builds but we netbsd-7.2 g++-4.8.5 http://perl5.test-smoke.org/report/93327 Test failures: AFAICT these are among the first smoke test runs done on NetBSD with Note that these failures do not include cpan/Scalar-List-Utils/t/uniq.t, Note that these failures also do not reflect the clang++ problems with If there's anyone with NetBSD experience (or even mere access) who can Thank you very much. perl perl perl |
From @jkeenanOn Mon, 19 Aug 2019 00:17:32 GMT, jkeenan@pobox.com wrote:
I have installed a NetBSD-8.0 VM and have built perl blead with -Duselongdouble. I am getting the same test failures Carlos is seeing in his smoke-test reports. See attachments. -- |
From @jkeenan../ext/POSIX/t/math.t .. Test Summary Report ../ext/POSIX/t/math.t (Wstat: 256 Tests: 59 Failed: 0) |
From @jkeenan# Failed test 1459 - '%a' '2.71828182845904509' -> '0xa.df85458a2bb48p-2' cf '0xa.df85458a2bb4a9bp-2' at op/sprintf2.t line 687 Test Summary Report op/sprintf2.t (Wstat: 0 Tests: 1694 Failed: 1) |
From @jkeenanSummary of my perl5 (revision 5 version 31 subversion 4) configuration: Characteristics of this binary (from libperl): |
The RT System itself - Status changed from 'new' to 'open' |
From @jkeenanOn Sat, 24 Aug 2019 04:19:03 GMT, jkeenan wrote:
The failures in the cpan/ExtUtils-MakeMaker tests all implicate TAP::Formatter::File and POSIX.PM: ##### Along the same lines, I am unable to install cpanm on NetBSD with '-Duselongdouble'. I get: ##### Now, when I test ext/POSIX/t/math.t with a longdouble build, I find that I get *multiple* crashes which are only revealed when I comment out crashing tests. Here are 3 groups of crashing tests and the crash message associated with each. ##### .../blead/lib/perl5/5.31.4/amd64-netbsd-ld/auto/POSIX/POSIX.so: Undefined PLT symbol "log2l" (symnum = 151) .../blead/lib/perl5/5.31.4/amd64-netbsd-ld/auto/POSIX/POSIX.so: Undefined PLT symbol "tgammal" (symnum = 47) Thank you very much. |
From @craigberryOn Sat, Aug 24, 2019 at 10:14 AM James E Keenan via RT
The reason for this is fairly obvious from a quick look at this #elif defined(USE_LONG_DOUBLE) && \ So the POSIX extension is assuming that if frexpl, ilogbl, and sqrtl <http://gnats.netbsd.org/47122> <https://lists.gnu.org/archive/html/bug-gnulib/2019-01/msg00114.html> Changing either Configure or POSIX itself to do a detailed probe for |
From @jkeenanOn Sun, 25 Aug 2019 15:48:07 GMT, craig.a.berry@gmail.com wrote:
I have no idea. We rarely get bug reports from "NetBSD people". We have been getting reports about problems with longdoubles lately, so one of our smoke-testers started to add configurations that built with longdoubles, and did so on multiple OSes. So this problem has only been reported by "Perl people". In NYC I know FreeBSD people and OpenBSD people but not NetBSD people. I did a git checkout of v5.24.4 on NetBSD-8.0, built with -Duselongdouble and got the same kind of failures in ext/POSIX/t/math.t (and, consequently, in MakeMaker test) and even *more* test failures in t/op/sprintf2.t. My belief is that Configure does not generally test for C99 functions, but from your analysis it appears that POSIX.pm does assume that if we have certain C99 functions we've got a lot more. Is that a fair statement? Thank you very much. -- |
From @craigberryOn Sun, Aug 25, 2019 at 6:28 PM James E Keenan via RT
Yes, that is a fair statement. The fallback is that it would use the |
From @khwilliamsonOn 8/25/19 9:47 AM, Craig A. Berry wrote:
What if we just another canary or two based on what's missing in netbsd |
From @jkeenanOn Mon, 26 Aug 2019 18:08:56 GMT, craig.a.berry@gmail.com wrote:
Note that the failures in ext/POSIX/t/math.t on NetBSD with -Duselongdouble are crashes, not mere test failures. There are at least 3 separate groups of crashes; only the first is being reflected in the smoke test results. Thank you very much. -- |
[snip] [# Craig wrote:]
@khwilliamson Can you explain what you meant by 'canary' here? I'm unclear. Thanks. |
I'm sorry; I apparently didn't respond to this, unless I did it on irc. Anyway, by canary, I meant some function for which we think its presence or absence would mean other similar functions will also be present or absent. Now, I'm wondering if there's any connection between this ticked and #17668 |
I don't think so, if it did have an effect it would be bus errors or segmentation faults |
Just to add my observations from trying to build all perls from v5.8.9 up to v5.30.3 on a NetBSD 9.0 CPAN smoker with -Duselongdouble. I got viable perls (i.e. no POSIX problems) from v5.8.9 up to v5.20.3. v5.22.0 up to v5.30.3 all exhibit the POSIX problems. |
Based on @bingos's research, I did a lot of bisecting on NetBSD-8.0 between v5.20.0 and v5.22.0. Based on that, I believe that the problems described in this ticket entered at at least two different commits (though the commits were part of a larger project). So I have created two new tickets, #17853 and #17854 to handle these. I believe that they will supersede this ticket, so I will close this ticket in about a week unless there is objection. Thank you very much. |
Closing ticket per plan above. |
Migrated from rt.perl.org#134376 (status was 'open')
Searchable as RT134376$
The text was updated successfully, but these errors were encountered: