bug#49524: 28.0.50; make-serial-process is not portable

[Ken Brown <kbrown@cornell.edu>]

On 7/12/2021 9:27 AM, Ken Brown wrote:
> On 7/11/2021 12:24 PM, Eli Zaretskii wrote:
>>> From: Ken Brown <kbrown@cornell.edu>
>>> Date: Sun, 11 Jul 2021 11:24:58 -0400
>>>
>>> Fmake_serial_process calls Fserial_process_configure, which calls
>>> serial_configure, which calls cfsetspeed with the speed argument equal to the
>>> numerical baud rate (e.g., 9600).  But the documentation of cfsetspeed says that
>>> the speed argument must be one of the Bnnn constants defined in termios.h (e.g.,
>>> B9600).  See, for example,
>>>
>>>     https://man7.org/linux/man-pages/man3/termios.3.html
>>>
>>> This incorrect call of cfsetspeed happens to succeed on GNU/Linux because
>>> glibc's cfsetspeed allows the argument to be the numerical baud rate, which it
>>> converts to the appropriate Bnnn constant.  But I don't think emacs should be
>>> relying on this undocumented behavior.  In particular, this doesn't work on
>>> Cygwin.  And it wouldn't even work on GNU/Linux if emacs used the cfsetspeed
>>> replacement defined in sysdep.c instead of glibc's cfsetspeed.
>>>
>>> I think the way to fix this is to imitate the glibc code that converts the baud
>>> rate to a Bnnn constant, but maybe someone has a better idea.
>>
>> Converting in sysdep.c:serial_configure sounds TRT to me.
> 
> Patch attached.

BTW, we've decided to change Cygwin's cfsetspeed to be compatible with glibc's, 
so the problem fixed by my patch won't exist on Cygwin going forward.  And I 
checked FreeBSD out of curiosity and found that there's no issue there either 
because of the way they define the Bnnn constants:

#define	B0	0
#define	B50	50
#define	B75	75
...

So I should probably remove the reference to non-glibc platforms in my commit 
message.

Ken