make check fails

[John Ralls]

On Jun 13, 2013, at 3:59 AM, Christian Stimming <christian at cstimming.de> wrote:

> Am Montag, 10. Juni 2013, 18:35:11 schrieb Alex Aycinena:
>> Hi,
>> 
>> Christian - I believe you did something recently in this area and make
>> check doesn't complete; perhaps related?
> 
> Unfortunately yes (due to the PWARN -> PINFO change). Thanks for pointing this 
> out. I've explained it in the commit message.


Christian,

You wrote in the commit message:
"Sigh. It turns out the utest-Split.c relies on the "warning" log level
in order to check for specific code paths. This sucks. The log level
"warning" should please be reserved for things that are actual warnings,
not for code path checks that are used in the unittests."

That's not correct. utest-Split may rely on the presence of the message to
check the code path, and in order to do so needs to match the loglevel,
logdomain, and message string. If you change one of those in the tested
function, you just change it to match in the test and it should work. In this
case you want to change the "foo differ" warnings to infos. No big deal,
just change the loglevel in the tests to match.

Now, ideally unit tests would only be checking the message to make sure 
that it's emitted: That's a side-effect of the function, and a unit test should
test all of the function's effects. Unfortunately most of the engine consists
of over-long functions with multiple code paths whose behavior is difficult
to test correctly coupled with excessive class interdependencies . In order to 
be able to refactor them for better testability I have to first get them tested 
somehow so that I'll know when I break something in the refactor. 

Regards,
John Ralls