[m-users.] Seeking some feedback on FFI style...

[Julien Fischer]

Hi,

On Sat, 6 Aug 2022, Sean Charles (emacstheviking) wrote:

> In an ongoing effort to not forget things as I learn, I just added to
> my collection of mercury samples with this:
> 
> https://github.com/emacstheviking/mercury-library-samples/blob/main/FFI/sample.m
> 
> I just wondered if, by the final version, if the code could be shorter
> / cleaner / mopre elegant etc and also, more importantly, is there
> anything I've done wrong, not done as well as it could have been in
> terms of efficiency ?

1. The will_not_throw_exception attribute is meaningless on
foreign_procs that also set the will_not_call_mercury attribute; you can
omit it.

2. Speaking of which, you have set the attribute will_not_call_mercury
for the precdicate filestat4/4. However, the foreign_proc for that
predicate *does* call Mercury, e.g. via the exported predicate
statResponse().

3. Ditto for filestat3/4, which calls Mercury via makeError() but has
the will_not_call_mercury attribute set.

4. I would handle the I/O state in foreign_procs via don't-care
variables instead of having an assignment in the body of the
foreign_proc.  E.g. do this.

     :- pragma foreign_proc("C",
         foo(_IO0::di, _IO::uo),
         [promise_pure, will_not_call_mercury, etc etc],
     "
       ...
     ").

rather than this:

     :- pragma foreign_proc("C",
         foo(IO0::di, IO::uo),
         [promise_pure, will_not_call_mercury, etc etc],
     "
       ...
       IO = IO0;
     ").

The former approach saves a line. In the past the latter approach
has also triggered warnings from some C compilers.

5. You may want to look into the settings the tabled_for_io attribute
on foreign_procs that do I/O if you want to be able to use the debugger
with such code.

Julien.