CHANGES6

The following list describes the changes made to the original source
code for the Sixth Edition (V6) UNIX Thompson shell and global
command in porting them to run on modern UNIX systems.

-------------------------------------------------------------------------------
[osh-current]: Friday, 2010-05-14:
	* No changes.

-------------------------------------------------------------------------------
[osh-20100430]:
	* No user-visible changes.

-------------------------------------------------------------------------------
[osh-20100228]:
	* No user-visible changes.

-------------------------------------------------------------------------------
[osh-20091218]:
	* No user-visible changes.

-------------------------------------------------------------------------------
[osh-20091127]:
	* No user-visible changes.

-------------------------------------------------------------------------------
[osh-20091122]:
sh6.c:
	* Increased apid[] buffer size to 33 via DOLMAX; this allows $$
	  to correctly represent shell's process ID on systems where
	  process IDs > 99999 are possible.

glob6.c:
	* Changed ERR_ALTOOLONG to ERR_E2BIG.

*6.1:
	* Added COPYRIGHT section headings, and added copyright notices.

-------------------------------------------------------------------------------
[osh-20091029]:
sh6.c:
	* Wrap WCOREDUMP(s) w/ #ifdef WCOREDUMP ... #endif .

sh6.1:
	* Change example in "I/O redirection" subsection that could
	  confuse new users.

-------------------------------------------------------------------------------
[osh-20090527]:
sh6.c:
	* Replaced err(), prn(), prs(), and xputc() w/ err() and fd_print()
	  from err.c .

	* Added sh_errexit() to handle all error exit scenarios for the shell.
	  This function is called by err().

	* Copied prsig() from osh.c; changed pwait() to use it for printing
	  termination messages.

glob6.c:
	* Replaced gerr() w/ err() and ut_errexit() from err.c .

-------------------------------------------------------------------------------
[osh-20081213]:
	* No changes.

-------------------------------------------------------------------------------
[osh-20081122]:
sh6.c:
	* Fixed a bug introduced in osh-20081026 where the shell incorrectly
	  prints a process ID w/ an empty termination message upon receipt
	  of the SIGINT signal in certain cases.  For example:

		% sleep 10 | sleep 20 | sleep 30 ; wait
		^C
		2845:
		2846:

	  This example should only print three newlines to the standard error
	  on SIGINT, not one newline and two process IDs w/ empty messages.
	  Now, it behaves as expected.

-------------------------------------------------------------------------------
[osh-20081026]:
	* Moved common definitions for these programs to "defs.h".

sh6.c:
	* Included "sh.h" for needed declarations.

	* Synced `struct tnode' w/ osh.c by changing the order of the `nav'
	  and `nflags' members.

	* Changed calls to err() to use the diagnostics defined in "defs.h".

glob6.c:
	* Added a base reallocation multiplier to gavnew().  This doubles the
	  number of new arguments for which to reallocate memory when the
	  current allocation will not be large enough for the resulting
	  argument vector.  This reduces the number of realloc()s required
	  for very large argument vectors while allowing the first malloc()
	  to be a relatively small allocation for up to 126 matching
	  file-name arguments.

	* Changed calls to gerr() to use the diagnostics defined in "defs.h".

-------------------------------------------------------------------------------
[osh-20080629]:
sh6.c:
	* Changed the way that the shell deals w/ the SIGTERM signal so that
	  it only sets SIGTERM to its default action for child processes if
	  it was set to its default action when the shell started.

	* Changed the code to use sasignal() instead of signal(3).

sh6.1.in:
	* Updated the documentation as needed to reflect changes to the code.

-------------------------------------------------------------------------------
[osh-20080109]:
	* No user-visible changes.

-------------------------------------------------------------------------------
[osh-20070707]:
	* No user-visible changes.

-------------------------------------------------------------------------------
[osh-20070324]:
	* No changes.

-------------------------------------------------------------------------------
[osh-20070321]:
	* Imported the project into a local subversion repository.

	* Changed the all of the code to "#include <stdbool.h>" and to use
	  the "bool" data-type macro where appropriate.

	* User-visible changes?  None.

-------------------------------------------------------------------------------
[osh-20070131]:
sh6.c:
	* Changed the way SIGINT and SIGQUIT are handled so that the shell's
	  child processes also ignore these signals when the shell and its
	  parent process do.  See the "Signals" subsection of sh6(1) for a
	  complete description of the new behavior.  Notice that this is the
	  same behavior found in osh(1).

glob6.c:
	* Changed the glob6 non-zero exit status so that it is always 2 when
	  any error is encountered.  This preserves compatibility w/ the
	  original, which never caused the shell to terminate a command file
	  when a globbing error occurred.

-------------------------------------------------------------------------------
[osh-20061230]:
sh6.c	1.3 (jneitzel) 2006/09/15:
	* The shell now ignores any `#!shell' sequence which occurs as the
	  first line of a regular file.  For example, the first line might
	  be `#!/usr/local/bin/sh6', `#!/usr/bin/env sh6', or similar.
	  However, it is important to notice that the original Thompson shell
	  did not do anything of the sort.  This change simply adds a degree
	  of convenience and flexibility for the user.

	* After the shell is invoked and initializes itself, it closes
	  unneeded file descriptors.  Changed when this is done so that it
	  does not break x-only scripts if they are supported on a particular
	  UNIX system.

	* The shell no longer exits w/ an explicit non-zero status if a read(2)
	  from the input file descriptor fails.  It now exits silently w/ the
	  current value of the global variable status in such a case.

	* The shell now exits silently w/ a zero status if the input file
	  descriptor refers to a directory.  The reason for this change is
	  because invoking the shell as `sh6 file', `sh6 <file', `sh6 - <file',
	  or similar can possibly cause the user's terminal to freak out on
	  some systems if `file' is a directory.  This change simply adds a
	  degree of predictability and eliminates the possible need for a
	  terminal reset in such cases.

	* Added explicit checks to ensure that file descriptors 0, 1, and 2
	  are open.  If any of them is not open, the shell exits w/ a non-zero
	  status and prints a diagnostic if possible.

	* Changed how SIGINT and SIGQUIT are ignored for `&' commands.
	  This change causes a `&' command to ignore these signals even
	  in those cases where the main shell process ($$) does not, such
	  as when executing a command file.

	* Added an additional `: cannot open' diagnostic for those cases where
	  `/dev/null' must be opened for a `&' command, but the shell's attempt
	  to do so fails.

	* Changed the `try again' diagnostic to `Cannot fork - try again'
	  since it more accurately describes the error condition.  The
	  "try again" portion of the diagnostic indicates that the
	  inability to fork(2) may be a temporary condition.

	* Trim the directory name used for `chdir dir' so that the user can
	  quote it if needed.

	* Trim the file name used for `<file', `>file', and `>>file' so that
	  the user can quote it if needed.

	* Parse each command line into a dynamically allocated tree of tnode
	  structures rather than using a statically allocated intptr_t buffer.

	  Remember that malloc(3) did not exist in Sixth Edition UNIX;
	  alloc(III) did exist, but /bin/sh did not use it.

glob6.c	1.3 (jneitzel) 2006/09/23:
	* Improved the quoting behavior for the sake of convenience,
	  flexibility, and usability.

	* Improved the behavior of `[...]' range matching (e.g., `[a-z]') so
	  that it adheres to the documentation (see glob6(1)).  This means
	  that an invalid range such as `[z-a]' matches nothing and that
	  `[a-m-z]' is equivalent to `[a-mz-]' instead of `[a-z]'.

	* Dynamically allocate memory for the generated argument list rather
	  than using statically allocated buffers.  The total length of the
	  generated argument list is limited to a maximum of ARG_MAX bytes.
	  An `Arg list too long' diagnostic is printed if this limit is
	  exceeded or if pexec() fails and sets errno to E2BIG.  Of course,
	  `Out of memory' is printed if glob6 runs out of memory.

	  Remember that malloc(3) did not exist in Sixth Edition UNIX;
	  alloc(III) did exist, but /etc/glob did not use it.

-------------------------------------------------------------------------------
[osh-060124]:
	This is the initial public release for the following ports of the
	original Sixth Edition UNIX source code.

sh6.c	1.2 (jneitzel) 2006/01/09:
	* Deny set-ID execution.  If the effective user (group) ID of the sh6
	  process is not equal to its real user (group) ID, it shall now print
	  a diagnostic and exit with a non-zero status.

	* Use geteuid(2) instead of getuid(2) to determine whether to print
	  the `% ' prompt for a regular user or the `# ' prompt for the
	  superuser.

	* Cause the shell to ignore SIGINT, SIGQUIT, and SIGTERM when invoked
	  as an interactive shell.  The original Thompson shell did not do
	  this.  In Sixth Edition UNIX, the login(1) utility invoked the shell
	  in the following way:

		execl("/bin/sh", "-", (char *)0);

	  ...
	  This caused the invoked login shell to ignore SIGINT and SIGQUIT.
	  However, this was not the case for a non-login interactive shell
	  invoked by a user.

	* When invoked as `cat <runcom | sh6' or `sh6 <runcom', cause the
	  shell to not print prompts to file descriptor 2.

	* Fixed a file descriptor leak in the creation of pipelines where any
	  of the commands is a shell special command.  A pipeline such as
	  `echo hello | : | grep h' is one way to demonstrate this leak in
	  an unfixed copy of the shell.

	* The shell no longer attempts to open `file' in a pipeline of the
	  following form:

		% echo hello | cat <file | grep h

	  ...
	  since `cat' will never read from the opened file.  This fix causes
	  the shell to silently ignore the ambiguous redirect, adhering to
	  the behavior originally documented in sh(1) (see sh6(1)).

	* The shell no longer attempts to create `file' in a pipeline of the
	  following form:

		% echo hello | cat >file | grep h

	  ...
	  since `cat' will never write to the created file.  This fix causes
	  the shell to silently ignore the ambiguous redirect, adhering to
	  the behavior originally documented in sh(1) (see sh6(1)).

	* The shell no longer breaks pipelines of the following form:

		% ( ( echo hello ) | ( cat ) | ( grep h ) ) >file &

	  ...
	  This fix further limits when `/dev/null' is opened as input for `&'
	  commands so that doing so cannot replace the read end of a pipe
	  previously set up by the shell, adhering to the behavior originally
	  documented in sh(1) (see sh6(1)).

	* Added explicit error detection in those cases where the original
	  Thompson shell did not do so.  For example, the original Thompson
	  shell did not check the value returned by dup(2) or pipe(2).

	* Changed all of the close(2)/dup(2) sequences for file descriptors
	  0 and 1 to simply call dup2(2) instead.

	* Changed all of the calls to exit(3) which happen in the child process
	  after a successful call to fork(2) to _exit(2) instead.

	* Added `exit' as a special command which is built into the shell.
	  Its usage is the same as that of the external `exit' utility found
	  in Sixth Edition UNIX.

	* Define explicit exit status for special commands, external commands,
	  and shell errors.

	* Removed the `: too large' diagnostic, and added a `: cannot execute'
	  diagnostic for external-command execution errors.  The `No shell!'
	  and `: not found' diagnostics, however, are unchanged.

	* Use pexec() (see pexec.h) for external command execution to
	  give more flexible and predictable command search behavior.

glob6.c	1.2 (jneitzel) 2006/01/09:
	* Deny set-ID execution.  If the effective user (group) ID of the
	  glob6 process is not equal to its real user (group) ID, it shall
	  now print a diagnostic and exit with a non-zero status.

	* Increased the size of the argument-list buffers to allow generation
	  of longer argument lists.

	* Added a `Pattern too long' diagnostic for those cases where the
	  total length of a pattern argument is >= PATH_MAX (or 1024).  If
	  such an error is encountered, glob6 exits with a non-zero status.

	* Handle quoted arguments specially so that path names containing
	  blanks can be successfully globbed.

	* Use pexec() (see pexec.h) for command execution to give
	  more flexible and predictable command search behavior.

-------------------------------------------------------------------------------
[Unreleased]:
	The following ports of the original Sixth Edition UNIX source code
	were never released to the public.

sh6.c	1.1 (jneitzel) 2004/01/26:
	* Changed the source file name from `sh.c' to `sh6.c' and the
	  resulting binary name from `sh' to `sh6' to avoid name conflict
	  and confusion with the system shell, sh(1), on modern UNIX systems.

	* Transformed the source code from K&R C into ANSI C.

	* This is nothing more than a simple port to modern UNIX systems.
	  It is now 64-bit clean.  At least, this is true for the 64-bit
	  systems to which the porter has access.

	* Removed support for the shell's accounting of process times (see
	  times(3)) which allowed the shell to record command execution data
	  in a log file.

	* Fixed a buffer overflow in the word() function.

glob6.c	1.1 (jneitzel) 2004/01/26:
	* Changed the source file name from `glob.c' to `glob6.c' and
	  the resulting binary name from `glob' to `glob6', matching
	  the similar change to the name of the shell (see above).

	* Transformed the source code from K&R C into ANSI C.

	* This is nothing more than a simple port to modern UNIX systems.

	* Fixed possible overflow of the argument-list buffers
	  in the cat() function.

-------------------------------------------------------------------------------
[Original]:
	Here is some pertinent information about the original files.

The following description refers to Sixth Edition (V6) UNIX obtained from:

	http://www.tuhs.org/Archive/PDP-11/Distributions/research/Dennis_v6/

...
A portion of the `http://www.tuhs.org/Archive/PDP-11/Distributions/DETAILS'
file reads as follows:

Dennis_v6
---------

v6root.gz, v6src.gz and v6doc.gz are a set of three RK05 images of Sixth
Edition with root, /usr and documentation, from Dennis Ritchie. The tape
bootstraps are not included. The most recent timestamp on any file is
July 19th 1975.

v6root.tar.gz, v6src.tar.gz and v6doc.tar.gz are tar archives of the
three RK05 images.

...
The following files and the corresponding manual pages were all
obtained from the tar archives described above:

sh.c:
    What: Original source code for the Sixth Edition (V6) UNIX Thompson shell
    Date: July 17, 1975
    From: Sixth Edition UNIX /usr/source/s2/sh.c

glob.c:
    What: Original source code for the Sixth Edition (V6) UNIX global command
    Date: May 14, 1975
    From: Sixth Edition UNIX /usr/source/s1/glob.c

Jeffrey Allen Neitzel						2010/05/14

@(#)$Id$