Re: [boost] [thread][1.34][tss] Borland and gcc tss_lib failures

From: Anthony Williams (anthony_w.geo_at_[hidden])
Date: 2006-12-11 08:38:57

Roland Schwarz <roland.schwarz_at_[hidden]> writes:

> Anthony Williams wrote:
>> Both Borland and gcc-mingw are failing the tss_lib tests, due to lack of
>> tss_cleanup_implemented, implying we don't have a means of doing automatic TSS
>> cleanup when static linking with these compilers.
>> 
>> Unless anyone has any good reason not to (such as an idea on how to implement
>> such automatic cleanup), I will mark these as expected failures tomorrow.
>
> I am not sure this is the right way to do it.
>
> cleanup implemented can be done by user code, by explicitly calling the 
> cleanup on process end, i.e. before end of main.
>
> Only automatic cleanup is missing, which I think is simply an 
> optimization, but not strictly required.
>
> Please give me some days to look into it (perhaps weekend?), I'll try to 
> find out the proper way to handle the issue. (No I am not suggesting 
> that I will implement automatic cleanup.) Possibly the cleanup 
> implemented should be "fake implemented" by the test code for these 
> compilers.
>
> Ok?

If you're willing to look into this, great, I'll leave it until you've had a
chance.

Yes, it's the (lack of) automatic cleanup that's causing the problem, and I
agree that user code can call the cleanup functions when exiting a thread in
order to ensure everything is cleaned up. However, this is not the same, and
is clearly missing functionality, so some test should be marked as an expected
failure.

Maybe we should split out an automatic cleanup test, which can then be marked
"expected failure", and have the other tss_lib tests do manual cleanup. That
way the rest of the TSS tests have a chance of running.

Anthony
-- 
Anthony Williams
Software Developer
Just Software Solutions Ltd
http://www.justsoftwaresolutions.co.uk