path: root/ChangeLog

Changes since 1.2: (Too many to count!)

A memory stomping bug was fixed (provoked by assigning a variable
to its old value plus something else).

Better signal handling; a real signal handler which manages a queue
of pending signals was added.

rc now ignores SIGQUIT and traps SIGINT even in non-interactive
mode. Thus,

	rc ed.sh

will not do mysterious things if the shell script "ed.sh" calls a
command like "ed" and then you hit ^C.

rc now opens 0, 1, 2 on /dev/null if they are inherited closed.
rc -o prevents this.

A couple of stupid O(n^2) string appends were replaced with O(n)
loops. This should make foo=`{cat /etc/termcap} bar=$^foo a little
faster :-)

Returning a list of signals from a function is now legal, so "return
$status" should always work.

The code has been revised, new printing routines have been added.

rc no longer uses redundant braces when exporting functions.

A first stab at a verification suite has been added (trip.rc).
(someone, please help me make this comprehensive!)

rc -p now does not initialize functions from the environment. This
should make it easier to write shell scripts that don't need to
assume anything about the environment.

Inherited ignored signals are now ignored in the current shell and
passed on ignored to the child processes. whatis -s also reflects
this information.

A file descriptor leak in the /dev/fd implementation of >{} was
fixed.

A variable set to '' was not imported from the environment; this
has been fixed.

Changes since 1.3beta:

New Makefile/config.h setup.

builtin echo may now be conditionally included out, to use a Goldwynism.

builtin exit takes any legal exit status. If the status is not all zeros,
rc exits with 1. (having "exit sigiot" produce a core dump would be going
a little far, I think.)

limit does not append a unit after a zero limit; 0g was too confusing.

exec > /nonexistentfile does not cause rc to exit any more.

If a noninteractive rc is started with sigint ignored, rc does not install
its own signal handler.

error messages produced by rc in a subshell were cleaned up. (rc erroneously
reset the 'interactive' flag after a fork)

print.c was cleaned up a little; no functionality was changed, but should
be more portable now.

a bug in rc-1.3beta (not previous versions) was fixed: setting the first
element of $path to '' caused PATH to be exported as '':etc..

getopt's "illegal option" message was gratuitously changed to something
less abrupt.

some dead code was removed from input.c

%term was changed to %token in parse.y; apparently newer yacc's don't grok
%term any more.

a race condition in the signal handler was fixed.

the variable in for() was getting evaluated each time through the loop
(e.g., for (`{echo i;date>[1=2]} in 1 2 3)echo $i would print the date
three times). This was cleaned up.

a redundant fork() was removed from walk.c; this showed up when running
a braced command with a redirection in the background. e.g., {a;b}>c&

man pages for history and rc were cleaned up by david (thanks).

rc set SIGQUIT and SIGTERM to SIG_DFL on background jobs---even when
trying to do old-style backgrounding (i.e., don't use process groups,
just ignore SIGINT & SIGQUIT & SIGTERM).

$0 is now changed to the name of the signal when entering a signal
handler. Thus it's possible to write code like

	fn sigint sigterm sigquit {
		switch ($0) {
		case sigint
			...
		case sigterm
			...

wait with no arguments now prints the pid of any and all children
that died with a signal. e.g.,

	; wait
	25321: terminated
	25325: terminated

as opposed to

	; wait
	terminated

An error saving/restoring state in the input stream code would
cause rc to exit with the (erroneous) command:

	eval '|[a'

FIFO's were not removed in a backgrounded command, e.g.,

	cat <{echo hi}&

Changes since rc-1.4beta:

getopt was renamed to rc_getopt to avoid libc collisions.

$cdpath with a / in it caused a cd to sometimes have two //'s at the
front of the path. This is reserved by POSIX, so I changed it to skip
one of the /'s.

signal handling now emulates sh in the way I described in a previous
message: the race condition present in older rc's whereby some SIGINTs
got lost is now gone; any SIGINT received during a wait() is acted upon
at the end of the wait(), unless of course SIGINT is being deliberately
ignored.

getopt was renamed to avoid naming conflicts with libc. Also a sound
move since rc_getopt is no longer quite libc-getopt compatible; I had
to add in a mechanism for resetting getopt.

signal handler code in fn.c was cleaned up; there were several bugs
in rc-1.4beta, notably the shell sometimes spawned background jobs
with SIGTERM ignored. I took the opportunity to make things a little
cleaner here.

a quasi-memory leak in the code for while() was fixed: a long-running
while that had rc commands allocating memory in it could cause the
shell to grow without bounds. I fixed this by placing the while loop
(*and* test!) inside a new allocation arena each time through the loop.

A new configuration parameter, NOJOB, was added to allow you to force
v7-style backgrounding (no setpgrp, ignore SIGINT and SIGTERM).

The FIFO code was reworked a little. It should be more robust now---
FIFOs get removed at the end of the command of the argument list
that they were on:

		fn foo {echo $*; cat $*}
		foo<{echo hi}

now works as expected. Also FIFO names are pushed onto the exception
stack so that their removal occurs in the face of exceptions and so
on.

A memory leak in treefree() was plugged up --- the root node of a
function was not getting freed.

Changes since rc-1.4:

General changes:

        Some small memory leaks/uninit references revealed by Purify.

        $bqstatus for querying the exit status of a backquote.

        Globbing through unreadable directories.

        More options to whatis.

        History append which always re-opens the file (avoids
        inconsistencies over multiple NFS-mounted accesses to
        $history).

        Support "rc -s".

---------

Makefile:	Added comment referring to README for yacc/malloc problem.

uiltins.c:	Added more options to whatis, protected rlimit prototypes
		with #ifndef SYSVR4, renamed SIGCHK to sigchk.

except.c:	Patched nl_on_intr printing to happen only in interactive
		shells.

exec.c:		Added comment explaining nl_on_intr variable, renamed SIGCHK
		to sigchk.

fn.c:		Replaced by-hand consing of exception stack etc. for signal
		handler execution with call to funcall(). Replaced fun2str
		call with call on print routines.

footobar.c:	Got rid of memory leak in get_name(), parenthesize count,
		flat and var nodes for correctness in unparsing, removed
		list2str, made get_name use nalloc space, merge in a
		better parse_var from es.

glob.c:		Split out a test so that closedir is called correctly, renamed
		SIGCHK to sigchk.

glom.c:		Added bqstatus, renamed SIGCHK to sigchk, removed spurious
		setsigdefaults, patched uninit memory reference, rename
		"clear" to "memzero", wait for bq subproc to finish on EINTR.

hash.c:		Added options to function/variable print code.

history/history.c: Added '@' to the list of chars which can precede an
		ignored '-'.

input.c:	Got rid of tiny memory leak in closefds, got rid of uninit
		memory reference in popinput, moved nul ignored check into
		realgchar(), changed history append to always reopen the
		history file, replaced SIGCHK with sigchk. Freed memory
		returned by call to readline().

lex.c:		Corrected typo in comment, moved nul character ignore code
		to input routines.

main.c:		Added -s flag.

nalloc.c:	Added convenience feature to erealloc. (Allow NULL parameter)

print.c:	Changed use of va_start so broken compilers can compile
		print code.

proto.h:	Added fake memset.

rc.h:		Replaced clear() function prototype with a macro call on
		memset(), removed SIGCHK, fixed prototypes.

signal.c:	Removed unconditional exit in catcher code, renamed SIGCHK
		to sigchk.

status.c:	Rewrite sgetstatus breaking out strstatus so that bqstatus
		can be set from glom routines.

utils.c:	Got rid of clear() (use memset from C library), rename SIGCHK
		to sigchk.

var.c:	 	Got rid of memory leak in listassign(), converted list2str()
		call to something using the print routines.

version.c:	New version string.

wait.c:		Got rid of memory leak in rc_fork, renamed SIGCHK to sigchk.

walk.c:		Fixed pre-redirection bug, removed spurious setsigdefaults(),
		renamed SIGCHK to sigchk.


Changes since rc-1.5beta1

  Configuration: rc now uses GNU autoconf.

  Portability: mksignal works on HPUX 10.

  Portability: resources can be (quad_t).

  Bug: if rc was started with SIGCLD == SIG_IGN (e.g. by Solaris's
  rshd) it would loop.  Fixed by resetting SIGCLD to SIG_DFL if it
  was SIG_IGN when rc was started.

  Portability: POSIXish systems don't have NGROUPS.

Changes since rc-1.5b2

  Configuration: rc now uses GNU automake.

  Portability: use sigsetjmp() when available.

  Portability: improve tests for quad_t.

  Configuration: don't leave FIFOs in /tmp when configuring.

  Configuration: let configure find `-ltermcap' when using
  readline.

  Configuration: pick up default path from config.cache.

  Bug: sense of most HAVE_RESTARTABLE_SYSCALLS tests was wrong.

  Bug: print prompts with `-i', even with editline / readline.

  Testing: just say `false' (not `/bin/false').

  Bug: confusion over gid_t versus GETGROUPS_T in which.c.

  Feature: `-V' option added to report version number.

  Configuration: remove version.c; `id' is defined in main.c now.

  Bug: clear list of living children after fork(); `{ ls & wait }
  |cat' no longer hangs or panics.

  Testing: add regression test for above.

  Tidiness: all the system call wrappers to prevent calls being
  restarted now live in system-bsd.c.  The configure script
  decides whether to build system.c or system-bsd.c.  Also,
  signal.c is more careful to only declare slowbuf if it will be
  needed.

  Tidiness: similarly, configure decides whether to build execve.c
  or not.

  Portability: test for ssize_t and use it where available; use
  long if there's no ssize_t.

  Portability: use sigsetjmp where it's available and appropriate.
  If no sigsetjmp, just use sigjmp; this probably fails in a
  traditional SysV environment.

  Portability: test explicitly for SysV SIGCLD semantics.  Main
  (and dubious) benefit is that you can now define a function
  called `sigcld' on systems where there is no signal called
  SIGCLD!

  Bug: rc has its own memory allocator; don't use malloc directly.

  Bug: the rc_wait() wrapper is only needed on systems which
  restart system calls.  On Linux, in particular, the wrapper
  leads to a race which causes rc to hang (or panic in later
  versions).  Other systems apparently don't exercise this race.

  Bug: waitforall() must return if rc_wait4() returns with errno
  == EINTR.  Otherwise, the rc builtin `wait' cannot be interrupted
  if there are background processes (although apparently only if
  there is a handler for SIGINT).

  Portability: dreadful hack to track down the real locations
  of signal.h to find signal names.  rc now builds under CygWin32!

  Portability: replace above dreadful hack with mksignal.c, contributed
  by Vincent Broman.

  Portability: use POSIX wait() macros (WIFEXITED and friends).
  Unfortunately, POSIX omitted to supply WIFDUMPED, so this doesn't buy a
  great deal.

  Distribution: remove the dependencies of y.tab.[ch] on parse.y from Makefile.am.
  The justification for this is that, unless you're hacking on rc's grammar, there's
  no reason to use anything other than the distributed files (which were generated
  with byacc and very lightly edited to silence a few gcc warnings).

  Enhancement: the example in addon.c wasn't very useful, since it
  depended on files which weren't included with the distribution.  There
  is now a working, if trivial, example.

  Tidiness: the code was compiled with as many gcc warnings enabled as
  I could find.  Very few problems turned up.  Some unused function
  arguments were removed.  Where unused arguments could not be removed
  (because the function is one of a set that must all have the same
  prototype), they were renamed to `ignore'.

  Portability: some versions of readline define the name `Function',
  which rc also uses.  Its use in rc has been renamed to `rc_Function'.

  Documentation: the `history' man page didn't explain what end of line
  does in editing mode.  It now does.

  Bug: the interaction with readline was broken, particularly with
  respect to signal handling.  I've incorporated some changes from Tom
  Culliton which should sort this out.  Unfortunately, we now have 3 different
  code paths for readline 2.1, readline 2.2, and editline :-(.

  Configuration: if you say `--with-readline' or `--with-editline' it is now an
  error if the appropriate library is not found.