Re: gnustep-base on OpenBSD mips54

[Richard Frith-Macdonald]

On 10 Jan 2011, at 18:41, Sebastian Reitenbach wrote:

> Sebastian Reitenbach wrote:
>> Hi,
>> 
>> 
>> Fred Kiefer wrote:
>> 
>>> I just submitted the second and third part of your patch. The first part
>>> may or may not be useful, somebody using FreeBSD should comment on that.
>>> 
>>> Before you go ahead with GNUstep applications you should try to get the
>>> base test suite running and see whether this fully works on your system.
>>> Only then should we try to analyse the gui issues. GNUstep gui uses
>>> proxies for most images and this will only work, when the whole
>>> NSInvocation stuff in base is working correctly.
>>> 
>>> When you are sure to rule out a base issue the next step would be to run
>>> the application with gdb and set a break point on the assert line in
>>> [NSCell -setImage:] and if the code gets there analyse the content of
>>> anImage. It should either be an NSImage or more likely a GSThemeProxy.
>>> 
>>> 
>> It took a while to run those tests, but I'm done now. And indeed, there
>> are some tests failing with regard to NSInvocations.
>> See attached logfile.
>> 
>> Compiling NSInvocation.m I get a lot of warning messages on the mips64 box:
>> NSInvocation.m:918: warning: cast increases required alignment of target
>> type
>> NSInvocation.m:918: warning: cast increases required alignment of target
>> type
>> NSInvocation.m:918: warning: cast increases required alignment of target
>> type
>> NSInvocation.m:918: warning: cast increases required alignment of target
>> type
>> NSInvocation.m:918: warning: cast increases required alignment of target
>> type
>> NSInvocation.m:918: warning: cast increases required alignment of target
>> type
>> NSInvocation.m:918: warning: cast increases required alignment of target
>> type
>> 
>> I don't see those messages on my i386 box, maybe that could be one of
>> the reasons?

Those could be a compiler bug ... why is it reporting a lot of casts on the 
last line of the file ... they certainly aren't really there, so something is 
wrong with the compiler, even if it's just that it's giving the wrong line 
number for real errors.

As far as test failures are concerned, I think you first need to test that 
libffi is working in your platform.  Last time I looked there was no support 
for MIPS64 on any flavour of BSD, but maybe your FreeBSD is supported now?  I 
know that MIPS64 gnu/linux is supposed to be supported, so if it doesn't work 
on your FreeBSD system you might be able to fix it or find some BSD/MIPS expert 
to fix it for you.

Anyway, your log of the testsuite showed a *LOT* of failures including problems 
with NSBundle, you should really get all the tests for the base library (apart 
from the latest NSLocale additions) passing before attempting to go any 
further, and getting libffi working is a key stage before getting the base 
library working.