harmony-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Gregory Shimansky <gshiman...@apache.org>
Subject Re: [testing] first M7 candidate (r681495) testing status
Date Fri, 22 Aug 2008 12:48:19 GMT
Nathan Beyer said the following on 21.08.2008 17:51:
> Can we tweak the Locale during the test case, so that the Locale
> matches the fixture data?

The locale during test case is not enough. It is locale during build 
time is what is important. Test produce lots of classes with 
international characters in their names. These classes have to be 
somehow written to the file system when they are compiled.

If system locale is set to some value that doesn't support some 
international characters, then compiled class files contain "?" symbols 
in their names. Obviously class files with such names cannot be found 
later when they are searched by class loader.

I don't know in which international characters are used in the tests, 
but it looks like they weren't picked up from a particular language but 
were selected randomly from the whole possible set of the unicode symbols.

So only with locales that are truly unicode, not some subset of it (e.g 
"C", "ASCII", "POSIX", "ru_RU.KOI8-R, zh_CN.GB2312 are all subsets for 
some particular language and don't support any other languages), it is 
possibly to successfully build and execute these tests.

> On Thu, Aug 21, 2008 at 8:36 AM, Gregory Shimansky
> <gshimansky@apache.org> wrote:
>> chunrong lai said the following on 21.08.2008 6:51:
>>> hi,:
>>>  I got an update about the failure in java.lang.ClassGenericsTest.test_2
>>> in
>>> DRLVM_test.
>>>  I see the test case is using some harded-UTF-encoded class-name during
>>> the
>>> test. So the test only passes if the locale is set to "en_US.UTF-8" or
>>> "zh_CN.utf8". If the locale is set to "C" or "zh_CN.GB2312" it just fails.
>>> RI also can not pass the test case in such locales. So it looks like a
>>> test
>>> case (or environment setting) problem.
>> Yes, I investigated this test's failure two years ago :) See [1] and replies
>> from Salikh, Alexey and Rana.
>> [1] http://markmail.org/message/ku3juqhnyhysy5gb
>> --
>> Gregory


View raw message