[MLton] commit r4725: Statically-linked wrapper functions for Intel Mac

[John Reppy]

I posted a question on the XCode mailing list and one of the Apple  
engineers
supplied the answer.  You should generate

	.section __IMPORT,__jump_table,symbol_stubs,self_modifying_code 
+pure_instructions,5

instead of

	.symbol_stub

The manual claims that these are the same, but apparently not.

	- John

On Oct 28, 2006, at 2:32 PM, Matthew Fluet wrote:

>
>> The changes seem to have addressed the problem with the  
>> regressions. There are a bunch of ~0.0 vs. 0.0 failures in the  
>> real tests, but the other regressions ran fine.  As expected, the  
>> FFI stuff causes the same problems as with the math functions.  I  
>> think that you are going to have to modify the code generator to  
>> produce the dynamic linking stubs on Mac OS X.
>
> Great.  At r4732, I went ahead and modified the code generator to  
> produce dynamic linking stubs on x86-darwin.  If you get a chance  
> to update to the latest SVN sources and re-compile, it should be  
> the case that FFI works properly.
>