From owner-freebsd-hackers Thu Oct 17 13:46:22 1996 Return-Path: owner-hackers Received: (from root@localhost) by freefall.freebsd.org (8.7.5/8.7.3) id NAA28629 for hackers-outgoing; Thu, 17 Oct 1996 13:46:22 -0700 (PDT) Received: from godzilla.zeta.org.au (godzilla.zeta.org.au [203.2.228.19]) by freefall.freebsd.org (8.7.5/8.7.3) with ESMTP id NAA28623 for ; Thu, 17 Oct 1996 13:46:17 -0700 (PDT) Received: (from bde@localhost) by godzilla.zeta.org.au (8.7.6/8.6.9) id GAA11694; Fri, 18 Oct 1996 06:43:21 +1000 Date: Fri, 18 Oct 1996 06:43:21 +1000 From: Bruce Evans Message-Id: <199610172043.GAA11694@godzilla.zeta.org.au> To: ache@nagual.ru, terry@lambert.org Subject: Re: POSIX TEST SUITE Cc: hackers@FreeBSD.org Sender: owner-hackers@FreeBSD.org X-Loop: FreeBSD.org Precedence: bulk >I *know* there will be exceptions in signal and tty handling; I >*expect* other problem areas as well (times updates, etc.). These areas should be fairly conformant, since I have run a test suite on them and fixed and fixed many of the problems. My current list of POSIX conformance problems is: signal handling: - for bogus signal numbers, sigismember() sometimes completes successfully and returns 1 although the number is not in the set. It must either detect the error and return -1 and not detect the error and return 0. - sleep() interfers with alarm() not only during the sleep but for 1 second after. - when sleep() is interrupted by an alarm, 0 is returned instead of the residual time. tty handling: - EOPNOTSUPP instead of EINVAL is returned for attempts to lock a tty. file times: - atimes and mtimes are marked for update upon unsuccessful completion. - atimes and mtimes for devices on at least ufs file systems are marked for update even before i/o is begun. - ctimes in directories are marked for update by unsuccessful renames. file system: - all syscalls involving pathnames are broken because "" is interpreted as "." instead of being invalid. - writing 0 bytes to a file doesn't always have no effect (it extends the file after an lseek() to past the end). - lseek() on a named pipe doesn't return -1/EISPIPE. - chmod doesn't silently ignore attempts to set the setgid bit by the owner of the file when the owner doesn't have permission to set this bit. - many "constant" pathconf limits aren't constant. other: - many "constant" sysconf limits aren't constant. Bruce