[Freeipa-devel] [PATCH 68] text unit test should validate using installed mo file

[Petr Viktorin]

On 03/30/2012 02:41 AM, John Dennis wrote:
> On 03/28/2012 04:40 AM, Petr Viktorin wrote:
>> Can install/po/Makefile just call test_i18n.py from the tests/ tree? It
>> doesn't import any IPA code so there's no need to set sys.path in this
>> case (though there'd have to be a comment saying we depend on this).
>> In the other case, unit tests, the path is already set by Nose.
>> Also the file would have to be renamed so nose doesn't pick it up as a
>> test module.
>
> Good idea. I moved test_i18n.py to tests/i18n.py. I was reluctant about
> moving the file, but that was without merit, it works better this way.

The downside is that the file now looks like a test utility module. It 
could use a comment near the imports saying that it's also called as a 
script, and that it shouldn't import IPA code (this could silently use 
the system-installed version of IPA, or crash if it's not there).
Alternatively, set PYTHONPATH in the Makefile.

> I also removed the superfluous comment in Makefile.in you pointed out.
>
> When I was exercising the code I noticed the validation code was not
> treating msgid's from C code correctly (we do have some C code in the
> client area). That required a much more nuanced parsing the format
> conversion specifiers to correctly identify what was a positional format
> specifier vs. an indexed format specifier. The new version of the
> i18n.py includes the function parse_printf_fmt() and get_prog_langs() to
> identify the source programming language.

More non-trivial code without tests. This makes me worry. But, tests for 
this can be added later I guess.

> Two more patches will follow shortly, one which adds validation when
> "make lint" is run and a patch to correct the problems it found in the C
> code strings which did not used indexed format specifiers.
>


-- 
Petr³