Re: [Boost-testing] acc date-time failures (was Missing toolsets from 1.33.1 to 1.34)

From: Boris Gubenko (Boris.Gubenko_at_[hidden])
Date: 2006-06-12 15:59:39

Jeff Garland wrote:
> I've made the markup for the testfacet test cases which accounts for
> 2 of the 7 failures.

Thanks!

> The rest of the failures look like acc compiler issues to me.

The failing tests are:

testdate_input_facet[_dll]
testlocal_time
testlocal_time_input_facet
testtime_input_facet

They all compile cleanly with aCC6 when using STLport instead of the
Rogue Wave library, so, it looks like the issue here is not the
compiler per se, but STL.

What's interesting is that on Tru64, which has a slightly older RW
library than HP-UX -- rw libstd v2.0 and v2.2, respectively -- these
tests succeed. I'm wondering whether there is some missing
conditionalization for HP-UX somewhere and whether Tru64 is using the
old code which you are considering ditching in 1.35.

> [...] but honestly I'm ready to ditch that code in 1.35 and mark any
> compiler that can't deal with the new i/o unusable.

Sounds a little too harsh to me, especially given the fact, that it
*may* render cxx on Tru64 unusable, but it is your call.

> the best course is probably to just mark up the  rest of the
> failures

I'm not sure what the implications are: will it make these tests
permanent failures on HP-UX?

> and hope someone at hp looks at the remaining compiler issues.

I'll look at these issues in a couple of days when I'm done with HP-UX-
specific patches for STLport so they can make STLport 5.1 (the current
version of STLport 5.x is busted on HP-UX, both ia64 and PA-RISC).

Boris

----- Original Message ----- 
From: "Jeff Garland" <jeff_at_[hidden]>
To: "Running Boost regression tests" <boost-testing_at_[hidden]>
Sent: Monday, June 12, 2006 10:15 AM
Subject: Re: [Boost-testing] acc date-time failures (was Missing toolsets 
from 1.33.1 to 1.34)

> Thomas Witt wrote:
>> Boris,
>>
>> Boris Gubenko wrote:
>>> From: Thomas Witt <witt_at_[hidden]>
>>>> Let me clarify this. I am perfectly willing to accept acc patches
>>>> for 1.34. That is as long they don't break anything else.
>>>>
>>
>>> There are low-hanging fruits though: for example, date_time library
>>> tests testfacet[_dll] fail on HP-UX for exactly same reason they fail
>>> on Tru64: because of the bug in the Rogue Wave library. They are marked
>>> as expected failures on Tru64 and should be marked as such on HP-UX as
>>> well. I'd like to modify explicit-failures-markup.xml accordingly in
>>> both CVS HEAD and RC, but I don't have write access to CVS. So, how do
>>> I do it? Submit a patch for .xml file?
>
> I've made the markup for the testfacet test cases which accounts for 2
> of the 7 failures.  The rest of the failures look like acc compiler
> issues to me.  The remaining failures basically make using any of the
> more recent i/o features unavailable.  I could enable the old i/o code,
> which is more forgiving to compilers, but honestly I'm ready to ditch
> that code in 1.35 and mark any compiler that can't deal with the new i/o
> unusable.  Seems like the best course is probably to just mark up the
> rest of the failures and hope someone at hp looks at the remaining
> compiler issues.  Thoughts?
>
> Jeff
>
>
>
>
>
>
> _______________________________________________
> Boost-Testing mailing list
> Boost-Testing_at_[hidden]
> http://lists.boost.org/mailman/listinfo.cgi/boost-testing
>
>