On 01/18/2013 11:34 PM, Eli Zaretskii wrote:

> I'm not sure I see how the gap size fails to be updated.  There's a
> call to make_gap just before read_non_regular is called.  Or did you
> mean GAP_SIZE?  If the latter, then the comments there explain why
> this is not done.

Argh. We fool itself with gap_size (src/fileio.c, 24.2.92):

   4098              /* Maybe make more room.  */
   4099              if (gap_size < trytry)
   4100                {
   4101                  make_gap (total - gap_size);
   4102                  gap_size = GAP_SIZE;           /* !!! here */
   4103                }

After that, local gap_size (e.g. amount of bytes which may be used to
read next chunk from the file) is GAP_SIZE - inserted, not GAP_SIZE.

> But until GAP_SIZE and ZV are updated, the inserted text is not
> really part of the buffer, right?  So what is the problem here?

Here is the original code (src/fileio.c, 24.2.92):

   4169    /* Make the text read part of the buffer.  */
   4170    GAP_SIZE -= inserted;
   4171    GPT      += inserted;
   4172    GPT_BYTE += inserted;
   4173    ZV       += inserted;
   4174    ZV_BYTE  += inserted;
   4175    Z        += inserted;
   4176    Z_BYTE   += inserted;
   4177                                                                    /* !!! `inserted' bytes becomes "really inserted" */
   4178    if (GAP_SIZE > 0)
   4179      /* Put an anchor to ensure multi-byte form ends at gap.  */
   4180      *GPT_ADDR = 0;
   4181
   4182    emacs_close (fd);
   4183
   4184    /* Discard the unwind protect for closing the file.  */
   4185    specpdl_ptr--;
   4186
   4187    if (how_much < 0)
   4188      error ("IO error reading %s: %s",                             /* error leaves `inserted' bytes not decoded !!! */
   4189             SDATA (orig_filename), emacs_strerror (errno));

Attached is the fix for 24.2.92, and I believe that this is important
for the next pretest. For trunk, I'll revert 111547 and do the similar fix.

Dmitry