Front page | perl.perl5.porters |
Postings from June 2016
Re: The new Unescaped-left-brace-warning can be noisy
Thread Previous
|
Thread Next
From:
Karl Williamson
Date:
June 20, 2016 22:14
Subject:
Re: The new Unescaped-left-brace-warning can be noisy
Message ID:
57686A9F.30700@khwilliamson.com
On 06/20/2016 03:57 PM, Karl Williamson wrote:
> On 06/20/2016 12:39 PM, Andreas Koenig wrote:
>> Karl,
>>
>> FWIW, on GWILLIAMS/IRI-0.005.tar.gz 'make test' used to produce ~433
>> bytes, now ~3348343. This causes the normal reporting mechanisms to
>> refuse to work, so this does not show up on cpantesters.
>>
>> If you are looking for tests that trigger the new warnings, here are
>> others:
>>
>> http://www.cpantesters.org/cpan/report/f5acc38e-363d-11e6-82f7-cd4e388fbd9d
>>
>> (KJETILK/RDF-Trine-Node-Literal-XML-0.16.tar.gz)
>
> Every one of the above warnings is almost certainly a bug in the module,
> as all of them are of the form /\{,\d\}/. It appears that they believe
> that one can omit the lower bound of a quantifier. Indeed that is part
> of the purpose of this deprecation, to change the language so that this
> lower bound can be omitted. But it isn't the case today, and the code
> compiles into matching the sequence "{" "," \d "}", silently until now.
>>
>> http://www.cpantesters.org/cpan/report/ee393738-362c-11e6-83b5-bcf7378fbd9d
>>
>> (GWILLIAMS/Attean-0.017.tar.gz)
>
> This is the exact same thing, omitting the lower bound of a quantifier.
> The new warning has caught an existing bug in the code.
>>
>> http://www.cpantesters.org/cpan/report/c9f77224-35f8-11e6-974b-b539388fbd9d
>>
>> (IVANWILLS/File-CodeSearch-v0.7.4.tar.gz)
>
> This is probably just the warning legitimately deprecating an existing
> legitimate usage. There are two constructs failing ".{-1}" and .{-2}".
> It's possible those are very weird attempts at a quantifier, but it
> seems to me more likely that the author meant to literally match the
> sequence "{" "-" "1" "}". The report shows three instances of this
> warning.
>
>>
>> Thanks,
>>
>
> So I'm not sure what to do about this. The tests show all three of the
> examples are either bugs in the modules, or show correctly a
> now-deprecated usage.
>
> I looked at IRI, and it has 5 occurrences of the {,\d} paradigm. I will
> look further to see if those are the things that could be causing all
> the noise, and issue a PR on it.
I have now issued the PR. On my system, I downloaded this module and
got some noise, but not the huge amount you reported. After adding in a
lower bound of 0, to the five instances, no warnings were raised, and
all tests passed.
>
> Maybe the regex compiler should not show the entire pattern when it
> finds problems, but instead some amount (perhaps a large amount) of
> context around the error.
>
>
>
Thread Previous
|
Thread Next