From mboxrd@z Thu Jan  1 00:00:00 1970
Received: by archone.tamu.edu id <22543>; Mon, 1 Jul 1991 18:22:08 -0500
From: Byron Rakitzis <byron>
To: rc
Subject: rc-1.1beta
Message-Id: <91Jul1.182208cdt.22543@archone.tamu.edu>
Date: Mon, 1 Jul 1991 18:21:53 -0500


It's available for ftp from archone. I would like the members of this list to
try it out before I release it to the General Public. I would also like to
incorporate a few more additions before I release the "real" 1.1, e.g., support
for /dev/fd and also John Mackin's scripts for non-ANSI build environments.

Enclosed is the file CHANGES from rc-1.1beta

I forgot to mention in CHANGES that David Sanderson did a great job of reworking
the man page.

Anyway, here goes.

------

Changes since rc-1.0:

builtin now forces a path-search for non-builtin commands. e.g.,

	builtin ls

forces a path search for /bin/ls, rather than running a function
called "ls".

---

A bug in the builtin wait was fixed.

---

The builtin whatis how takes a -s flag, for displaying available signals
and their handlers, and also prints correct code for reinterpretation.
(metacharacters inside variable names were not correctly quoted)

---

An extra field was added to the "limit" builtin for SunOS systems, which
apparently allow a limit on file descriptors.

---

A home-rolled execve() was written for people with geriatric unices that don't do
#! in the kernel.

---

SIGTERM is now ignored by rc along with SIGQUIT. (This is not how Duff's shell behaves,
but it is standard practise for Unix shells to ignore SIGTERM)

---

If $history was set to a bad file, then rc would print far too many error messages.
Now rc prints one error message and assigns null to $history, if $history names
an invalid file.

---

rc now explicitly refuses to trap SIGCLD on System V machines, because of the weird
way in which SIGCLD work. (really because I didn't want to get into the job control
business)

---

A bug was fixed in the parser so that functions with metacharacters in their names
are correctly imported from the environment.

---

The globber was changed to call stat() before calling opendir(); apparently opendir()
succeeds on some Unices even when invoked on regular files.

---

It is now illegal to have '=' be part of a variable name. This should be the only illegal
character inside a variable name!

---

backquote rescanning was rewritten from scratch; now it no longer assigns null list entries
to represent consecutive occurences of $ifs characters in the input.

e.g., now `{echo '   '} returns a null list, and not a 4-element list of null characters.

---

rc used to export all handlers but those in {sighup, sigint, sigquit, sigterm, sigexit}.
Now rc does not export ANY signal handlers.

---

rc's lexer was cleaned up to accept 8-bit data. rc is now presumed to be 8-bit clean.
I would like to hear of evidence to the contrary.

---

rc now traps EOF when scanning a variable name inside a heredoc.

---

rc's fdgchar() was cleaned up so that it did not assign negative values to unsigned objects.
(this one is truly for the pedants)

---

rc's lexer was fixed so that multiple backslashes at the end of a line are interpreted
correctly.

---

rc's parser now allows a newline to appear after "else" (via skipnl()).

---

main() was exiting in certain situations with an exit status of 0 instead of
rc's real exit status.

---

rc's yacc file was fixed so that yaccs which do "magic" can clean up after themselves
via the tokens YYACCEPT and YYABORT. Most notably, Sun's yyparse() calls malloc.

---

A mistake in the argument list of treecpy() was cleaned up.

---

Calls to write() are now checked for failure.

---

walk() was optimized for space in two ways: rPIPE was removed and placed in a separate function
body, and a few obvious chances for employing tail-recursion via goto's was taken. This should
result in a slightly smaller use of stackspace.

---