[Openmcl-devel] Raw-mode terminal weirdness redux

[R. Matthew Emerson]

> On Mar 27, 2018, at 10:05 AM, David Brown <lisp at davidb.org> wrote:
> 
> On Tue, Mar 27, 2018 at 09:32:54AM -0700, Ron Garret wrote:
>> Found a reliable workaround!
>> 
>> ...
>> This version is broken.  If you comment out the line that says “This
>> breaks it” it will start working properly.  Both working and
>> non-working versions are 100% reliable.  Using a single RLET to
>> allocate both termios structures also works.  It’s only when there is
>> a call to tcgetattr in between the allocations that it breaks.  (Note
>> that the “broken” version works in 1.8.  But for the fact that C code
>> can reproduce this problem intermittently I would call this slam-dunk
>> evidence that this is a bug in the CCL compiler, most likely in RLET.
>> But since the C code does what it does, I’m not willing to go quite
>> that far.)
> 
> The common thread in all of these, is that "raw" is only partially
> initialized (cfmakeraw only initializes c_iflag).  Subtle changes in
> stack positioning will make the initial values different, which
> explains both the inconsistent behavior with the C program, and the
> behavior of the lisp programs (and why it matters how you declare it).
> 
> You need to either tcgetattr again to fill in the 'raw' with the
> current values, or copy the data from the 'normal' before calling
> cfmakeraw on it.
> 
> My guess is that, with c_oflag being set to an uninitialized value, it
> has flags set such that it generates no output.

I was under the faulty impression that #_cfmakeraw would fully initialize the termios structure.  Nice catch.

So, we'd want something like

(defun set-tty-raw (tty)
  (rlet ((attr :termios))
    (#_tcgetattr tty attr)
    (#_cfmakeraw attr)
    (eql 0 (#_tcsetattr tty #$TCSAFLUSH attr))))