This is a list of things that still need to be done:

* Fix the totally bogus _atomic_lock in arch/mips.
  This is a nightmare because you really need to incur a ctxt switch
  for pmaxen. R4000 has some instructions to do an atomic test-and-set
  but not mips1. fuck fuck fuck. the ctx switch has to be sigsetmask

* Add thread stuff to the other archs in libc/arch/; ie change
  some usages of ENTRY to SYSENTRY in some .S files and add
  the new macros to their SYS.h. Only the i386 arch is working.

* Finish/verify the md stuff for setjmp and atomic_lock for sparc and
  alpha. Boy do they look dodgy.

* Find a better way of doing the _sys_aliases stuff; ideally by
  using the linker. jmp's (current implementation) are ugly, but
  I don't want to have to hack ../libc/sys/Makefile.inc just to
  partition the syscalls into wrapped and non-wrapped. The `.set'
  assembler directive just doesn't work as I thought it would.

* Fix the bug where _thread_init() isn't called when libc_r.so is used
  instead of libc_r.a ... some kind of shlib common problem? may
  need to use crt0 instead? __CTOR_LIST__ is just not being updated by ld.
  Frankly this one is driving me nuts. Maybe <bsd.lib.mk> is wrong.
  This varies on different architectures and exec format.

* Test the objective-C compiler to see if it can use threads.
  Really this shows up a nasty problem with needing to say -threads
  whenever you want to use threads. Ideally, linking with -lc_r library
  would override -lc; but the _POSIX_THREADS define would alter
  some optimisations done in <stdio.h>... What to do!?! FreeBSD looks
  like they are going to have an always-threaded libc with optimisations
  for the 1-thread case and use weak symbols to override functions when
  someone uses -lc_r. Do we have weak symbols everywhere? I wonder
  what the performance hit is if #define LOCK() if(nthreads>1){lock();}
  is used... need to profile this.

* Find all the static buffers in libc and make them per-thread. This
  is not absolutely necessary but incurs no penalty for single-threaded
  operation, and makes life easier & safer for when threads are used.
  On the other hand, it adds a bit more code bloat when you use threads
  and makes openbsd programs non-portable to other OSs with less-nicer
  libraries. If we do the bare posix minimum, we promote portable code.
  
* Look into how netbsd are going with their kernel threads

* Look into how asynchronous I/O can help us. In particular, the
  (unimplemented) aio*() functions.

* Update the libc manual pages to describe the posix re-entrant functions.
  Although this is actually trivial to do, I have to decide on a
  consistent way of adding them - maybe `.Sh THREAD-SAFE FUNCTIONS' ?
  Should look into standards to see what they suggest/did..
  May also need to document "This is not thread-safe" for some library
  functions (yet to be identified).

* Find out where freebsd/netbsd use pread() and pwrite().. i think
  its in the database routines mostly. We will need to rip their code.

* Figure out what to do with the configuration system variables (_SC_*)

$OpenBSD: TODO,v 1.3 1998/11/20 11:18:57 d Exp $