bug#22096: 25.0.50; reading from fifo breaks display

[Eli Zaretskii]

> From: Mark Oteiza <mvoteiza@udel.edu>
> Date: Sat, 05 Dec 2015 13:29:14 -0500
> 
> Simpler recipe:
> 
> 1. mkfifo foo
> 2. echo "bar" > foo
> 
> In emacs -Q:
> 
> 3. Evaluate (insert-file-contents-literally "foo" t)
> 
> The buffer display is now broken.

Thanks.  My Emacs is configured with --enable-checking, so it aborted
due to assertion violation.  The patch below fixes that for me; please
see if it fixes the display problem for you.

Does anyone understand how come we used just
bset_enable_multibyte_characters, bypassing all the rest that is
required to make a non-empty buffer unibyte?  It looks simply
blatantly wrong to me.  Compare, for example, with what
decide_coding_unwind does.  Did we never execute this particular code
since it was written, at least not for non-empty buffers?

I hope Handa-san (CC'ed) could comment on this.

diff --git a/src/fileio.c b/src/fileio.c
index 6cda1e3..8e44eb0 100644
--- a/src/fileio.c
+++ b/src/fileio.c
@@ -4265,7 +4265,7 @@ by calling `format-decode', which see.  */)
          && NILP (replace))
        /* Visiting a file with these coding system makes the buffer
           unibyte.  */
-       bset_enable_multibyte_characters (current_buffer, Qnil);
+       Fset_buffer_multibyte (Qnil);
     }

   coding.dst_multibyte = ! NILP (BVAR (current_buffer, enable_multibyte_charac\
ters));