Regression tests, UNICODE vs ASCII

[Geoffrey Hausheer]

So I've finally started writing tests for functions that have both Unicode 
and Ascii equivalents, and I'm wondering how to write the tests.  I re-read 
the thread about this that occurred back in January, and there didn't seem to 
be a clear decision.

With the current framework, it doesn't appear to be possible to just write 
one test, and have it build both unicode and ascii versions, and it seems 
like at least Alexandre was strongly against this anyway.

But in my case, I know nothing at all about writing Unicode tests, or how to 
explicitly test unicode functionality, so I see that I have three options:
1) Don't write any 'W' tests, and use only 'A' functions in my tests
2) Duplicate my 'A' tests as 'W' tests, and acknowledge that there is likely 
Unicode specific stuff that is not tested
3) Learn something about Unicode, and figure out how to test the functions 
correctly.

Now obvioulsy (3) is the best choice, but honestly, I am not motivated 
towards this at the moment.  Which leaves (1) and (2), with (2) being the way 
I'm leaning.  But the tests are getting really big, and pretty messy, and 
duplicating them and managing both versions looks like a nightmare (there are 
lots of TODOs, and I expect the tests to behave differently on different OS).

So what is the recomendation that I do?
In the simplest case, I can just copy the Ascii test to a new file and 
replace all 'A' with 'W', and just keep the tests seperated that way.

Anyhow, any comments would be appreciated.

Thanks,
.Geoff