[dmd-beta] dmd 1.063 and 2.048 beta

[Don Clugston]

On 9 August 2010 14:07, Fawzi Mohamed <fawzi at gmx.ch> wrote:
>
> On 9-ago-10, at 13:46, Don Clugston wrote:
>
>> On 9 August 2010 13:23, Fawzi Mohamed <fawzi at gmx.ch> wrote:
>>>
>>> On 9-ago-10, at 12:37, Don Clugston wrote:
>>>
>>>> [...]
>>>> On a somewhat related topic -- Andrei commented out the unit tests in
>>>> biguintcore and biguintx86 (search for @@ to find them) stating that
>>>> they fail on OSX.
>>>> They certainly pass on Windows, and they are unchanged since the last
>>>> release. Can someone with access to OSX please uncomment them, and
>>>> confirm that they fail? If they're failing, I think it can only mean
>>>> we have an OSX-specific wrong-code regression, which we need to solve.
>>>
>>> On tango, they fail on osx, since like forever, I had told you but
>>> probably
>>> I should have made a more formal thing about it.
>>> I sort of submitted a bug in
>>> http://dsource.org/projects/tango/ticket/1958
>>
>> OK, sounds like the unit tests may never have passed on OSX.
>> BTW you can now assume that none of the Phobos devs will look at
>> anything related to Tango. Compiler-related bug reports need to be
>> added to Bugzilla.
>
> I have always tried to reduce compiler bugs to lib independent bugs, and
> submit them

Thanks. Seems like this one just didn't copied into Bugzilla. Which is
a shame, because it's important.
Anyway, from looking at the code, I reckon that it's the
position-dependent code that's the problem. Is it possible that OSX
requires position independent code? I had always assumed that if it
did, D_PIC would always be set, but from looking at the compiler
source, it looks as if D_PIC is set only if it was set from the
command line.

Try going into bigintx86, and replacing "D_PIC" with "all". If the
failing tests pass, that's the problem. And unless OSX has a concept
of PIC, I think OSX should define D_PIC at all times.