On Sat, 10 May 2003, Marc Wetzel wrote:
:)Hi Ted,
:)
:)I found the select bug also and tried to fix it.
:)
:)The following is a "fixed" routine that works for me. The bug was in
:)using getdtablesize() instead of FD_SETSIZE.
:)You should also add #include <errno.h> in the include part as I
:)introduced some retries if EINTR was found (as in avrdude).
:)
:)I also added some compiled version for the website.
:)
:)int TSerial::Rx(unsigned char* queue, int queue_size, timeval* timeout){
:) int ret;
:) fd_set rfds;
:) FD_ZERO(&rfds); FD_SET(serline,&rfds);
:)
:)retry:
:) if ((ret=select(/*FIXME ancient getdtablesize()*/
:)FD_SETSIZE,&rfds,NULL,NULL, timeout))==-1){
:) Info(4, "Select on %d returned retval:%d
:)errno:%d\n",serline,ret,errno);
:) if (errno == EINTR) {
:) goto retry;
:) }
:) throw Error_C("Select failed");
:) }
:) if (ret==0){throw Error_Device("Programmer is not responding.");}
:) int size = read(serline, queue, queue_size);
:) Info(4, "Receive: { ");
:) for (int n=0; n<size; n++) {
:) Info(4, "%c [%02x] ", isprint(queue[n])?(char)queue[n]:'.',
:)queue[n]);
:) }
:) Info(4, "}\n");
:) return size;
:)}
:)
:)
Ok, I've committed a fix as such:
http://savannah.nongnu.org/cgi-bin/viewcvs/uisp/uisp/src/Serial.C.diff?r1=1.4&r2=1.5
Using FD_SETSIZE is serious overkill since we only care about a single file
descriptor. I also used a while loop with a stop-gap to avoid a possible
(but very unlikely) infinite loop instead of the goto-label.
Could I get you guys to give it one more round of testing under cygwin?