Re: [PATCH 2/5] Improve portability (particularly for Woe32)

[Jose E. Marchesi]

> On Sunday, March 28, 2021 1:00 PM, Bruno Haible <bruno@clisp.org> wrote:
>
>> Georgiy Tugai wrote:
>>
>> > Should I leave the HAVE_FCNTL ifdefs in, or no?
>>
>> defined HAVE_FCNTL is equivalent to !(defined _WIN32 && !defined CYGWIN).
>
> Sure, but "HAVE_FCNTL" expresses the intent better than a WIN32 test, right? 
> :)
>
> e.g. if someone later figures out how to get full fcntl on Windows, or
> implements a new abstraction layer comparable to Cygwin, for that matter,
> that macro will still be correct.
>
> My point is that from a Woe32 portability perspective, disabling the MI by
> default would be enough, if we're planning on rewriting the MI I/O layer
> anyway. I don't have enough experience with multithreaded C code to be
> confident in doing that myself.

I agree.  It is perfectly ok to disable the MI interface when building
for Windows for the time being.

At some point someone may provide an alternate implementation of the MI
select loop that also works reliably in Windows, but that is a little
project in itself, and right now I would say it is more important to
have a working poke in Windows, albeit with limitations.