Front page | perl.perl5.porters |
Postings from October 2016
Re: newly crashing test
Thread Previous
From:
Dan Collins
Date:
October 22, 2016 20:23
Subject:
Re: newly crashing test
Message ID:
CA+tt54JQEq1c_6XGAhxvxmW9iu+pvSbj_8WUQWdqwUyHtn8vmw@mail.gmail.com
Resending because my response was blocked from p5p:
Interesting. Yeah, I'd like to be able to capture that better. For example,
on Linux, all you see is this:
ok 215
not ok 216 - RT \#3054: Recursive operator overloading should not crash the
interpreter # TODO RT #3054: Recursive operator overloading overflows the C
stack
# Failed test 216 - RT \#3054: Recursive operator overloading should not
crash the interpreter at t/test{DOT}pl line 1058
# got "Segmentation fault"
# expected "ok"
# PROG:
# use overload '""' => sub { "$_[0]" };
# print bless {}, __PACKAGE__;
# print "ok\n";
# STATUS: 35584
The relevant code is in t/test{DOT}pl but I'm not familiar with OS X, so I
wouldn't know how to work around that.
On Sat, Oct 22, 2016 at 3:57 PM, Jarkko Hietaniemi <jhi@iki.fi> wrote:
> On Saturday-201610-22 15:51, Dan Collins wrote:
>
>> Jarkko - I can't be sure from the line you posted. Does t/uni/overload.t
>> segfault (and therefore `make test` fails), or does the child process
>> created by that test segfault (it's supposed to, that's the bug).
>>
>
> The latter. "make test" passes.
>
> So I guess it's not deadly, just ugly as sin.
>
> Maybe the crashing could be captured somehow better?
>
> If it's the former, we can skip that test on everything but linux, where
>> it works as intended, or if necessary, we can skit it entirely.
>>
>> --
>> Dan
>>
>> On Sat, Oct 22, 2016 at 3:12 PM, Jarkko Hietaniemi <jhi@iki.fi
>> <mailto:jhi@iki.fi>> wrote:
>>
>> The new test added in
>>
>> http://perl5.git.perl.org/perl.git/commitdiff/111da7867221d4
>> c747d002c775d64f4cc5c00f80
>> <http://perl5.git.perl.org/perl.git/commitdiff/111da7867221d
>> 4c747d002c775d64f4cc5c00f80>
>>
>> seems to consistently crash in my OS X boxes:
>>
>> sh: line 1: 81278 Segmentation fault: 11 /Users/jhi/perl5/perl/perl
>> "-I../lib" "tmp81276A" 2>&1
>>
>> Looks like it has already been found to be problematic in FreeBSD:
>>
>> http://perl5.git.perl.org/perl.git/commitdiff/e7619ccdfa0f85
>> 62897ad922b44837e5f65194e9
>> <http://perl5.git.perl.org/perl.git/commitdiff/e7619ccdfa0f8
>> 562897ad922b44837e5f65194e9>
>>
>> Maybe the problem should be fixed before adding the test?
>>
>> (Adding per-platform exceptions may not be a maintainable solution.)
>>
>> While in general TODO tests are a good idea, if the test is flakey
>> and/or potentially resource-grabbing, up to the point of
>> machine-killing, maybe less brilliant of an idea.
>>
>
>
Thread Previous