LWP::Protocol:http does not check for temporary errors during upload

I had much difficulty in attempting to use 'trickle', a bandwidth
constraining utility, to limit the upload speed of flickr_upload,
which uses the Flickr::API modules, which in turn use
LWP::Protocol::http. An error response from LWP::Protocol::http was
being propogated back upward and causing the upload to abort.

I traced the problem to this segment of code in LWP/Protocol/http.pm:

if (defined($wbits) && $wbits =~ /[^\0]/) {
my $n = $socket->syswrite($$wbuf, length($$wbuf), $woffset);
unless ($n) {
die "syswrite: $!" unless defined $n;
die "syswrite: no bytes written";
}
$woffset += $n;

After the syswrite, the error response is not checked for a temporary
error like "Resource temporarily unavailable" and just fails outright.
I modified the code in my personal copy to:
- retry if the error response is "Resource temporarily unavailable"
- make smaller writes (makes for smoother rate shaping with 'trickle')
- use Time::HiRes and sleep for a short period of time between attempts

if (defined($wbits) && $wbits =~ /[^\0]/) {
my $n;
do {
$n = $socket->syswrite($$wbuf, 2048, $woffset);
} while ($! eq 'Resource temporarily unavailable' &&
sleep(0.005));
unless ($n) {
die "syswrite: $!" unless defined $n;
die "syswrite: no bytes written";
}
$woffset += $n;

That's probably not the most ideal general-purpose remedy, but it
worked great in getting it to not fail outright when run with trickle.

In any case, I thought I would point out this problem so that it could
perhaps be addressed in a future release of LWP.

Thanks!
-jrs

Re: LWP::Protocol:http does not check for temporary errors during upload

I'd call that a bug in trickle. If the select system call reports that
the file descriptor is writable, it's perfectly reasonable to assume that
the write system call will write at least one byte rather than fail
with EAGAIN.

/Bo Lindbergh