jr@bbncc5.UUCP (John Robinson) (10/29/85)
In debugging a redisplay problem with GNU emacs, I found that the
write() system call will sometimes (apparently due to caught signals)
return before having written all it was asked. Seems the auto-resume
feature of 4.2 slow-device output is not completed.
At any rate, this causes stdio problems (turns on the error bit in the
file block), and makes for a bizarre-looking screen. I modified
flsbuf.c to cause stdio to restart the failed write(), and now all
seems to work fine.
This is the RCS line from the file I used:
/* @(#)flsbuf.c 4.6 (Berkeley) 6/30/83 */
Here are the changes. The first is to _flsbuf() and the second to
fflush(). The latter is really the one that emacs seems to need:
70,72c70,77
< if (rn != n) {
< iop->_flag |= _IOERR;
< return(EOF);
---
> for (;rn != n;) {
> /* If write sent nothing or failed, give up */
> if (n <= 0) {
> iop->_flag |= _IOERR;
> return(EOF);
> }
> base += n; rn -= n;
> n = write(fileno(iop), base, rn);
81a87
> register int wn;
87,89c93,102
< if (write(fileno(iop), base, n) != n) {
< iop->_flag |= _IOERR;
< return(EOF);
---
> wn = write(fileno(iop), base, n);
> /* This retries the write until it decides it's a lost cause */
> for (;wn != n;) {
> if (wn <= 0) {
> iop->_flag |= _IOERR;
> return(EOF);
> }
> base += wn;
> n -= wn;
> wn = write(fileno(iop), base, n);