chip@tct.uucp (Chip Salzenberg) (08/22/90)
Attempting to make Bash 1.05 work with the POSIX job control under SCO
Unix 3.2.0, I've run into two bugs in SCO's POSIX implementation. I
thought those SCO Unix users out there would like to know about them.
(Perhaps SCO will also notice this report and fix the bugs. To dream
the impossible dream... :-))
1. The sigsuspend() function doesn't work in the presence of a
handler for SIGCLD. Even when sigpending() shows that a SIGCLD is
pending, sigsuspend() with an empty signal set does NOT call the
SIGCLD handler. However, if SIGCLD is unmasked with
sigprocmask(), the handler is called correctly. Therefore, this
bug seems to be specific to sigsuspend(), and not to the general
implementation of SIGCLD.
This bug makes impossible the reliable handling of a single SIGCLD
event, and is therefore quite serious.
2. The setpgid(0, 0) function makes the calling process the head of
its own process group; this is as it should be. However, it also
causes open("/dev/tty", O_RDWR) to fail, which is NOT as it should
be. In addition, attempting to use a file descriptor still open
to /dev/tty doesn't work. I only got job control to work somewhat
by doing the tcgetpgrp() and tcsetpgrp() calls on stdin instead of
/dev/tty.
Oddly enough, system calls that failed on /dev/tty returned EIO,
which is not in the list of error codes that are supposed to be
returned by tcgetpgrp() and tcsetpgrp().
This bug is serious; separate control of process group and
controlling tty is one of the most important features of POSIX job
control.
Because of these bugs, I did not succeed in getting Bash 1.05 to work
with job control under SCO Unix 3.2.0.
I understand that SCO Unix 3.2.2 includes ksh with job control. If
this is true, perhaps the port of ksh revealed the above bugs, which
means they're already fixed. Here's hoping...
--
Chip Salzenberg at Teltronics/TCT <chip@tct.uucp>, <uunet!pdn!tct!chip>