Date: Tue, 02 Jul 1996 16:19:27 +0800 From: Peter Wemm <peter@spinner.DIALix.COM> To: joerg_wunsch@uriah.heep.sax.de (Joerg Wunsch) Cc: jmz@freefall.freebsd.org (Jean-Marc Zucconi), CVS-committers@freefall.freebsd.org, cvs-all@freefall.freebsd.org, cvs-usrbin@freefall.freebsd.org Subject: Re: cvs commit: src/usr.bin/fetch main.c Message-ID: <199607020819.QAA08915@spinner.DIALix.COM> In-Reply-To: Your message of "Tue, 02 Jul 1996 08:39:46 %2B0200." <199607020639.IAA10225@uriah.heep.sax.de>
next in thread | previous in thread | raw e-mail | index | archive | help
>>>J Wunsch said:
> As Jean-Marc Zucconi wrote:
> > jmz 96/07/01 18:49:49
> >
> > Modified: usr.bin/fetch main.c
> > Log:
> > Use read(fileno(fp), ...) instead of fread(..., fp) to avoid buffered inp
ut.
>
> I'm still against it, and i consider it bad style that you're
> committing something that has caused opposition in the mailing lists.
>
> You haven't even marked your change with
>
> /* XXX bogus */
>
> as would be the least sign of warning.
>
> If all your concern is avoiding input buffering, why have you never
> considered doing The Right Thing and using setvbuf(3)?
The problem isn't buffering, it's fread() itself. libftpio has broken one
of the fundamental features of ftpget, being the up-to-date status
reporting. read() returns as soon as a tcp push happens, and the loop
checks to see if the next status report is due. If you use fread(), it
waits until the entire 30K buffer has been read. setvbuf() doesn't seem
to have any effect on this situation, as the only way out of the while
loop in fread.c is to get an eof or error.
ie: in filbuf.c:
fp->_r = (*fp->_read)(fp->_cookie, (char *)fp->_p, fp->_bf._size);
fp->_flags &= ~__SMOD; /* buffer contents are again pristine */
if (fp->_r <= 0) {
if (fp->_r == 0)
fp->_flags |= __SEOF;
else {
fp->_r = 0;
fp->_flags |= __SERR;
}
return (EOF);
}
return (0);
and in fread.c:
while (resid > (r = fp->_r)) {
(void)memcpy((void *)p, (void *)fp->_p, (size_t)r);
fp->_p += r;
/* fp->_r = 0 ... done in __srefill */
p += r;
resid -= r;
if (__srefill(fp)) {
/* no more input: return partial result */
return ((total - resid) / size);
}
}
There is no way for fread() to simply return the data that has been read
so far on a tcp connection, which is what's needed here.
The other thing that I've found that seems to work OK is to reduce the 32K
fread() buffer down to 1K or lower still.
-Peter
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199607020819.QAA08915>