[MLton] Newer cygwin build of MLton?

[Lars Bergstrom]

On Jul 12, 2008, at 2:40 PM, Matthew Fluet wrote:
> And be prepared to wait a while for the SML/NJ compiled MLton to  
> compile MLton.  Of course, if you are interested in compiling a  
> program other than MLton, after 'make nj-mlton', you'll have a  
> '<src>/build/bin/mlton' that can be used to compile programs.  This  
> SML/NJ compiled MLton will be slower than a MLton compiled MLton,  
> but possibly a faster route to your final goal than bootstrapping.

Thanks for the tip! This worked well, and I was able to use the output  
of nj-mlton to attempt some compilation, though I was never able to  
get a full compilation of mlton to work. SML/NJ ran out of heap on my  
machine with its paltry 2 GB of RAM :-)

> I don't think there have been any fixes to the Windows FFI since  
> 20070826 that would affect the treatment of stdcall functions.
>
> But, it looks like the 16-byte alignment is bumping the size of the  
> actual argument area, but also being used in the stdcall name  
> mangling.  You probably need to patch like the following:

Indeed, this fix resolved most of my linker issues! With just a little  
bit more hacking, I got it compiling programs - there were a few small  
issues around objects being emitting that called usleep (which appears  
to have been deprecated and removed from the cygwin libc), though that  
was easily kludged with a thunk.

Sadly, the generated binary is not valid. While I frankly am  
suspicious of either the cygwin tool stack or my own mistakes, I'll  
pass along anything I discover related to MLton.

Thanks again for the help!
- Lars