quasar@ctt.bellcore.com (Laurence R. Brothers) (03/31/89)
I've written a few hacks to draw stuff on the root window (I'm on a
Sun-4/260, OS 4.0, by the way). I open /dev/win0 and then pw_open the fd I
get. I now want to retain the pixwin I'm drawing to so as to automatically
handle damage. I presume the original root pixwin is retained, since it
does in fact handle damage. I've fiddled with the pw_prretained pixrect
field of the root pixwin without success. I've tried a number of different
tacks, already, so I'd appreciate it if any suggestions anyone makes are
tested first...
As a digression, I should mention that there is approximately a 25%
failure chance for a program whose first two executable lines are:
root_fd = open("/dev/win0");
root_pixwin = pw_open(root_fd);
If it doesn't work, either root_fd is set to -1, (error "file already
exists") or pw_open fails to work and returns NULL. Somehow, the success
of a program is dependent on the structure of the rest of the source code
-- i.e., a given source file will either always work, or always not work.
Changing the file may randomly change this state -- any ideas? Giving open
its second argument seems to have no effect on this behavior.
Incidentally, using dbx on one of these programs causes it automatically
to fail to open /dev/win0, returning -1, so it is kind of hard to debug....
Laurence R. Brothers (quasar@ctt.bellcore.com)
Bellcore -- Computer Technology Transfer -- Knowledge-Based Systems Developmentreg@lti.com (Rick Genter x18) (04/26/89)
> As a digression, I should mention that there is approximately a 25% > failure chance for a program whose first two executable lines are: > root_fd = open("/dev/win0"); > root_pixwin = pw_open(root_fd); If these are the two lines verbatim, then no kidding it fails at random. open(2) takes 2 or 3 arguments, but never 1. You want to do: root_fd = open("/dev/win0",O_RDONLY); as it is, you're passing garbage as your second argument. The third argument is only used for O_CREAT. By the way, O_* are defined in <sys/file.h>. (Actually in <sys/fcntlcom.h>, but that's included by <sys/file.h> [SunOS 4.0*].) If you're going to hack, at least use lint(1). - reg -- Rick Genter ...!{buita,bbn}!lti!reg Language Technology, Inc. reg%lti.uucp@bu-it.bu.edu 27 Congress St., Salem, MA 01970 (508) 741-1507
guy@uunet.uu.net (Guy Harris) (05/09/89)
>By the way, O_* are defined in <sys/file.h>. (Actually in ><sys/fcntlcom.h>, but that's included by <sys/file.h> [SunOS 4.0*].) They are properly defined by including <fcntl.h> (note that the 4.0 "open" and "fcntl" man pages indicate this - it's true of pre-4.0 releases as well); Berkeley screwed up by documenting <sys/file.h> as the place from which to get them: 1) they first appeared in System III, and the documentation there said to get them from <fcntl.h>, so there was prior art; 2) POSIX says to get them from <fcntl.h>; 3) getting them from <fcntl.h> works on 4.[23]BSD as well; getting them from <sys/file.h> will probably *not* work on S3 or S5. O_NDELAY is defined in SunOS 4.0 *only* by including <fcntl.h>, since there are two flavors of no-delay mode, and thus two versions of <fcntl.h> - the 4BSD-environment one ("/usr/include/fcntl.h") defines it one way, and the S5-environment one ("/usr/5include/fcntl.h") defines it another way.