2014-01-18 18:24:01 -05:00
|
|
|
POSIX Divergence
|
|
|
|
|
================
|
|
|
|
|
|
|
|
|
|
The Sortix operating system cares about compatibility with existing software and
|
|
|
|
|
maintaining such software by purging obsolete stuff from the system. Sortix
|
|
|
|
|
recognizes the POSIX standard as valuable and embraces it rather than fighting
|
|
|
|
|
it. Nonetheless, we don't implement standards for the sake of standards, but for
|
|
|
|
|
the benefits we get from complying. Sometimes the standard mandates something
|
|
|
|
|
that is broken or we can do considerably better. In those cases it's best to
|
|
|
|
|
diverge and fix the design mistakes properly - but we must not forget there is
|
|
|
|
|
a considerable compatibility cost we must pay. Don't diverge unless there is a
|
|
|
|
|
good reason to and compatibility and an upgrade path must be considered.
|
|
|
|
|
|
|
|
|
|
That said, Sortix has diverged from full POSIX compliance in the hope such
|
|
|
|
|
decisions would be for the better. This document attempts to list where Sortix
|
|
|
|
|
has intentionally diverged from POSIX 2008. See doc/obsolete-stuff for a more
|
|
|
|
|
detailed rationale of why the interfaces are absent or scheduled for removal.
|
|
|
|
|
Mind that Sortix doesn't fully intend to implement the XSI option and violations
|
|
|
|
|
of that isn't listed here.
|
|
|
|
|
|
|
|
|
|
It is worth noting that some particular interfaces scheduled for removal will
|
|
|
|
|
likely take a very long time to phase out and wholly remove.
|
|
|
|
|
|
|
|
|
|
Mandated by POSIX but not implemented in Sortix
|
|
|
|
|
----
|
2014-01-22 20:19:58 -05:00
|
|
|
* POSIX advisory locks (fcntl, lockf) are not implemented.
|
2014-01-18 18:24:01 -05:00
|
|
|
* getpgrp is not implemented.
|
|
|
|
|
* <strings.h> has been merged into <string.h>.
|
|
|
|
|
* Numerous namespace violations (will be fixed or documented here).
|
|
|
|
|
* Numerous missing features (will be fixed or documented here).
|
|
|
|
|
|
|
|
|
|
Obsolescent in POSIX but not implemented in Sortix
|
|
|
|
|
----
|
|
|
|
|
* gets is not implemented (obsolescent in POSIX).
|
|
|
|
|
* tmpnam is not implemented (obsolescent in POSIX).
|
|
|
|
|
|
|
|
|
|
Mandated by POSIX and scheduled for Sortix removal
|
|
|
|
|
----
|
|
|
|
|
* select is scheduled for removal.
|
|
|
|
|
* struct timeval is scheduled for removal.
|
|
|
|
|
* times, <sys/times.h> is scheduled for removal.
|
|
|
|
|
|
|
|
|
|
Obsolescent in POSIX and scheduled for Sortix removal
|
|
|
|
|
---
|
|
|
|
|
* asctime and asctime_r are scheduled for removal (obsolescent in POSIX).
|
|
|
|
|
* ctime, ctime_r are scheduled for removal (obsolescent in POSIX).
|
|
|
|
|
* utime, <utime.h> are scheduled for removal (obsolescent in POSIX).
|
|
|
|
|
|
2013-08-04 14:24:59 -04:00
|
|
|
Signal Stacks
|
|
|
|
|
-------------
|
|
|
|
|
|
|
|
|
|
Threads are able to set a recursive signal handling stack using sigaltstack(2)
|
|
|
|
|
even if SS_ONSTACK is currently set - while POSIX mandates EPERM in this case.
|
|
|
|
|
Such a stack will be used for recursive signals (with SA_ONSTACK set) for the
|
|
|
|
|
duration of the signal handler. The original signal stack state will be restored
|
|
|
|
|
when the signal handler returns, any edit with sigaltstack will be temporary
|
|
|
|
|
(unless the saved ucontext is modified).
|
|
|
|
|
|
2014-01-18 18:24:01 -05:00
|
|
|
Timestamps
|
|
|
|
|
----------
|
|
|
|
|
|
|
|
|
|
The time_t values given by clock_gettime(CLOCK_REALTIME, ...) (and other system
|
|
|
|
|
interfaces built upon that) are the number of actual seconds that has passed
|
|
|
|
|
since 1970-01-01 00:00:00 UTC. This includes leap seconds, unlike the timestamps
|
|
|
|
|
mandated by POSIX. This has the advantage that time_t values are unambiguously
|
|
|
|
|
translate to a particular date and time and back and that the values are
|
|
|
|
|
continuous and adding an internal to a time_t value actually delays by that
|
|
|
|
|
interval. This has the unfortunate consequence that translating to and from
|
2014-05-28 11:29:40 -04:00
|
|
|
POSIX time requires subtracting the number of leap seconds that has occurred for
|
2014-01-18 18:24:01 -05:00
|
|
|
that time value. The standard library needs to get some utility functions for
|
|
|
|
|
doing such conversion.
|
