Re: [RFC, fortran]: Updated TRANSFER patch, now with a proper target-memory implementation.

[Brooks Moses <bmoses at stanford dot edu>]

At 03:33 PM 4/6/2007, FX Coudert wrote:
> Many thanks to Paul and yourself for working on this, it looks very
> good and reasonable code (which is not easy to write with the host/ target 
> memory representation mismatch). I intend to look into it a
> bit over the week-end. My few suggestions for the near future would
> include:
>   a- writing testcases that systematically compare parameter to non- 
> parameter TRANSFER, on all possible objects

I believe Paul is working on something like this, though I'm not sure of 
the details of his plans.

>   b- find a way to run these testcases in a cross-compiler setup
> where host and target have different endianness. Maybe we could get
> some help about setting this up? (the only cross I ever worked with
> is linux-mingw)

I am hopefully acquiring a G4 Mac for GFortran development (and general 
playing) sometime soon; potentially this weekend.  This sounds like a good 
use for it. :)

> Then, I have one more question: did you try this patch on real (kind=10) 
> variables, including arrays and derived type components? I
> think it's one of the difficult corner cases...

I haven't done much of that, but there are some in one of the test programs 
I was using, which I got from Paul a couple of weeks ago.  It at the least 
compiles without errors, and claims that the real(kind=10) variables are 
being stored in 12 bytes of space, which I gather is the correct number.

- Brooks