Re: Bug#786895: lintian: incompatible-java-bytecode-format warning needs update for Java 1.7

[Niels Thykier <niels@thykier.net>]

On 2015-05-26 16:35, Axel Beckert wrote:
> Hi,
> 
> Markus Koschany wrote:
>> I think it is time to adjust the lintian warning about
>> "incompatible-java-bytecode-format". The current version of Lintian
>> still warns about Java 1.7 bytecode, however Java 1.7 is the default
>> in Jessie and it is reasonable to update the check now.
>>
>> It seems to be sufficient to change the value of
>> "max-bytecode-version" in data/java/constants from 50 to 51.
> 
> There is a related bug report against lintian at
> https://bugs.debian.org/783815 (Using oracle java 8 in a package
> generate unknow-java-class-version).
> 
> I'm actually unsure about the semantics of max-bytecode-version.
> Citing from my (still unanswered) mail to the above bug-report:
> 
> The code introduced in 2.5.23 may have an off-by-one error, but it
> also may be on purpose (the changelog entry sys "Catch uses of Java8
> byte code"):
> 
>             if (   $mver <= $MAX_BYTECODE->value('min-bytecode-version') - 1
>                 or $mver
>                 >= $MAX_BYTECODE->value('max-bytecode-existing-version')) {
>                 # First public major version was 45 (Java1), latest
>                 # version is 52 (Java8).
>                 tag 'unknown-java-class-version', $jar_file,
>                   "($class -> $mver)";
> 
> The ">=" in there likely should be just a ">" if Java 8 byte code
> (i.e. for use with openjdk-8 from experimental) is fine. Or am I
> misunderstanding the semantics of max-bytecode-existing-version?
> 
> 		Regards, Axel
> 

Hi,

The off-by-one seems to be a regression/mistake in be04767.  It suggests
it intends to make Java8 known.  Unfortunately, the off-by-one causes to
exactly not do this.

Beyond fixing the off-by-one a.k.a. s/>=/>/, we should probably also fix:

"""
        # If the lowest version used is:
        $bad = 1
          if $jmajlow == $MAX_BYTECODE->value('max-bytecode-version')
          + 1; # Java7 - consider bad per request.
"""

Which I suspect should be a >= under the assumption that there could
eventually be two (or more) versions between max-bytecode-version and
max-bytecode-existing-version.

Beyond that, everything else should be updating the data file accordingly.

Thanks,
~Niels