Re: [Gm2] O/S issues

[gaius]

address@hidden writes:

> john o goyo <address@hidden> writes:
>
>> On 12/01/2010 09:53 AM, address@hidden wrote (in part):
>>> Question to all: are there any known major bugs outstanding which
>>> would prevent 1.0 from being released?
>>
>> For Solaris 10/sparc users, there is the issue of incorrect code being
>> generated for HIGH() in certain circumstances.  However, I appreciate
>> that most users will be running on PCs so this bug should not prevent
>> releasing 1.0.
>>
>
> Hi John,
>
> yes true it is a show stopper on Solaris - the HIGH regression tests
> you posted passes on x86_64 and x86_32 so I agree it should not
> prevent 1.0 from the x86 series.  I'll have a go at trying to
> reproduce this on a ppc system (ps3 squeeze), if this fails to
> reproduce the same bug then I think 1.0 should be flushed out for the
> x86* linux systems
>
> regards,
> Gaius

Hi John,

I've spent some time investigating this on the ppc ps3 debian and in
conclusion I do see a similar problem with your length test (albeit
the first test succeeds and the second fails).  I suspect this is due
to an assumption in the Modula-2 gcc tree generation code which
assumes that sizeof(long)>=sizeof(char *).  Removing this assumption
is not too difficult (it should never have been made of course) - but
the changes will be sprinkled over 3,4 modules.

  So it is likely to introduce some instability while the code is
being changed.  On reflection I'm inclined to note this as work to do
after the release of 1.0.

regards,
Gaius