developer is officially maintaining the package.
The rationale for changing this from "tech-pkg" to "pkgsrc-users" is
that it implies that any user can try to maintain the package (by
submitting patches to the mailing list). Since the folks most likely
to care about the package are the folks that want to use it or are
already using it, this would leverage the energy of users who aren't
developers.
+SHELL.
* Turn PKG_REGISTER_SHELLS into a variable that can be set in the shell
environment so that admins can make a choice when installing from
binary packages.
* PKG_SHELL is now a list of paths, and if the path is relative, then it
is taken to be relative to ${PREFIX}. Convert packages that set
PKG_SHELL to take advantage of this new feature by changing the full
paths to the shells into relative paths.
examples:
* Updated the example initialization files.
* Added a few command files to give users a head start
in scripting the shell.
osh.1 and sh6.1:
* Made a few more minimal revisions.
osh and sh6:
* Did a little more code cleanup.
Changes that improve compatibility w/ the Sixth Edition Unix shell
are marked w/ a `C:' in the details below.
-------------------------------------------------------------------------------
[osh-041028]:
*.1:
* Fixed a few typos in osh.1 and sh6.1.
* Did a little fine tuning of osh.1 and sh6.1 to hopefully
eliminate some incomplete and/or unclear explanations.
* Did the same for both if.1 and goto.1.
osh and sh6:
* Fixed an annoying bug introduced in the previous release...
The way error messages were printed in error() was not accounting
for the fact that the standard error stream is quite often (if not
always) unbuffered by default. This could make some error messages
difficult to read when a pipeline was involved. A little example:
Before (unfixed):
% foo|bar|baz
foo: not foundbar: not foundbaz: not found
After (fixed):
% foo|bar|baz
foo: not found
bar: not found
baz: not found
-------------------------------------------------------------------------------
[osh-041018]:
This release includes sh6 in addition to osh, if, and goto.
Sh6 is simply osh without the enhancements. It is built from the
same sources as osh, but it is not built or installed by default.
Makefile:
* Added targets for sh6 so that it can optionally be built, tested,
and installed if desired.
osh and sh6:
* Split the code into modules.
This is primarily for developer sanity... ;)
The new files are osh.h, main.c, parse.c, and exec.c.
* If (geteuid() != getuid() || getegid() != getgid()) is true,
print a nice error message and exit with a status of 2.
* Strip all NUL characters from the shell's standard input as it is
being read into the command line buffer. Input to the shell is
expected to be text.
* Changed the way the shell handles non-seekable files.
This is for both initialization files (osh only) and command files.
Do not block on open(2); open it and determine if it is a regular
file (or seekable). If it is not a regular file or is not seekable,
exit with an error. If seekable, reset the file for blocking I/O
and continue as normal.
Note that you can still read commands from FIFOs if you want.
Instead of doing `osh myfifo', you can either do `osh <myfifo'
or `osh - my list of positional parameters <myfifo'.
* Changed the error handling to use stdarg(3).
This allows for more code consistency and makes it easier
to handle all errors with just one line of code.
osh only:
* Added another possible initialization file for osh: $HOME/.oshrc
Osh only attempts to execute commands from this file if it is an
interactive shell. In the case of a login shell, osh tries this
file only after it tries both /etc/osh.login and $HOME/.osh.login.
* Made osh less strict about errors in initialization files.
Previously, common shell-detected errors in any of the files
were generally treated in the same way as they would have been
treated in a command file (i.e., the error was fatal). Realizing
that this potentially caused difficulty and annoyance for the user,
I opted to change it so that these types of errors are handled as
they are when osh is interactive.
This should make it easier for the user to debug
initialization files if needed.
* Added a `source' special command.
It is functionally similar to the way this command works in csh(1).
See osh(1) for details.
if:
* Changed ARGMAX from 50 to 256.
* If (geteuid() != getuid() || getegid() != getgid()) is true,
print a nice error message and exit with a status of 2.
* Added a few new primaries: `-h', `-s', `-t', and `-x'
See if(1) for details.
goto:
* Changed the size of the label buffer from 128 to 1024.
* Do not require the `:' of a labelled line to appear in column 1.
Instead, allow the `:' to optionally be preceded by blanks so that
labelled lines can be indented in command files. See goto(1) for
more details if needed.
* Eliminated unnecessary calls to strcmp(3) whenever a possible label
cannot possibly match the label argument given on the command line.
* Give an error if any NUL character is encountered in the input.
* Give an error if a zero-length string is given as the label argument.
-------------------------------------------------------------------------------
[osh-040812]:
osh.c:
* Changed the way unused pipe descriptors are handled in the child
process after fork(). This fixes a bug where the pipe in a pipeline
such as `( cat /dev/zero ) | sleep 1' would never enter an EOF state.
Previously, the close-on-exec flag was being set for the descriptors
in question. Of course, this did not work for the above and similar
cases. So, the descriptors in question are now close()d explicitly.
* Made some final changes to globbing to allow for more sensible
behaviour WRT quoting. The functions affected are: globargs(),
globchar(), and striparg(). Read and/or run `tests/glob_test.osh'
for details of the user-visible changes. This script may cause
previous versions of the shell to dump core.
Basically, since striparg() had always been called *after* glob(3)
and since the path names generated by glob(3) cannot be trusted, it
ends up that the best course of action is to call striparg() *before*
glob(3). This relatively simple change fixes a variety of *possible*
problems related to globbing.
-------------------------------------------------------------------------------
[osh-040731]:
osh.c:
* Added bounds checking to striparg() to protect against a possible
buffer overflow. Though this is unlikely to happen in the general
case, it is certainly *not* impossible. This function had been
unprotected since at least osh-020214.
* Added a new diagnostic, `Arg too long', to go with the
above-mentioned change.
* Removed the `No directory' diagnostic from globargs().
It was simply not worth the trouble. In compatible mode,
the `No match' diagnostic provides sufficient compatibility
in my opinion...
* Changed the `chdir' command so that it only attempts to change to
the previous working directory when the `-' argument is *not* quoted.
This alows the shell to change to a directory by that name.
For example:
% mkdir -; chdir -; pwd; chdir \-; pwd; chdir -; pwd; rmdir -
chdir: no old directory
/home/jneitzel/osh_stuff/osh-040731
/home/jneitzel/osh_stuff/osh-040731/-
/home/jneitzel/osh_stuff/osh-040731
Remember that "-" or '-' has the same effect as the \- used above.
This is perhaps a little silly, but I figure if a directory *can*
exist then it should be possible to change to it.
* Fixed the `<-' redirection argument so that it adheres to the
documentation. This fixes a file descriptor leak in addition
to the incorrect behaviour. It should be silently ignored in
the following case: `echo hello | grep h <-'; now it is.
-------------------------------------------------------------------------------
[osh-040723]:
osh.c:
C: Fixed a small idiosyncrasy with the `No directory' and `No match'
diagnostics when running in compatible mode. For example:
Before:
% if -d . -a -r . echo "`.' is a readable directory.";\
echo foo*; echo ?; echo []
`.' is a readable directory.
No match
No match
No directory
After:
% if -d . -a -r . echo "`.' is a readable directory.";\
echo foo*; echo ?; echo []
`.' is a readable directory.
No match
No match
No match
The shell should only print `No directory' when a directory
does not exist (ENOENT) or cannot be read (EACCES). Yes, it is
expected that invalid patterns such as `[' and `[]' result in
the shell printing a `No match' diagnostic.
* Changed the `<--' input redirection argument to `<-' instead.
This seems more consistent and will allow for possibly clearer
documentation in the future (if and when I add another feature
I've been thinking about).
fd2.[1c]:
* Removed the fd2 utility and its manual page because of
possible licensing issues.
-------------------------------------------------------------------------------
[osh-040718]:
This release is made primarily to synchronize with the new branch
of the shell which is named `sh6'.
osh.1:
* More revisions and clarifications...
-------------------------------------------------------------------------------
[osh-040714]:
Makefile:
* Refined the description for _XOPEN_SOURCE a little.
osh.1:
* General improvements...
This includes documenting some things that have never been very well
documented in this shell.
osh.c:
C: Made changes to globbing which affect the shell in compatible mode.
This includes the addition of the `No directory' diagnostic which
was present in /etc/glob from Sixth Edition Unix. Also, it looks
like I had previously misinterpreted exactly when the `No match'
diagnostic was supposed to be printed. Now, when running in
compatible mode the shell really is compatible. Yay =)
* Added a `umask' special command.
* Disallow SIGCHLD from being trapped.
When this signal is requested in a `trap' command, it is quietly
disallowed. This is the same behaviour seen with both SIGKILL and
SIGSTOP. Thus, doing a `trap + 9 17 20' quietly has no effect.
* Made changes to how the shell builds a command's argument vector.
Previously, malloc(3) was used. Now, each command in the command
line is simply split into `\0'-terminated words. Each argument is
actually a pointer to the corresponding word in the command line.
The changes to parameter substitution in osh-040628 made this
a perfectly sensible course of action. This change also gives
a microscopic improvement in run-time performance (as judged
by time(1)).
The only remaining use of malloc(3) in the shell's execution stage
can be found in globargs().
-------------------------------------------------------------------------------
[osh-040628]:
Thanks to Stephen M. Jones for suggesting that osh should be able
to read a global rc file. Thanks to Stephen C. VanDahm for assisting
with some portability issues found in osh-040421. Thanks also to
Josep Portella Florit for reviewing osh-040421, making several useful
suggestions, and sending patches.
BTW, many changes were made to the manual pages.
I hope they are clear, but I trust that if they are not then someone
might be kind enough to tell me so and/or make suggestions.
Some new files are included:
examples/*: initialization file examples
fd2.[1c]: the fd2 utility
Makefile:
* Added some notes about _XOPEN_SOURCE.
* Added a target to optionally build and install fd2.
* Removed the compile-time definition of `CLONE'.
This is now a run-time option which can be toggled in order to
enable or disable enhancements to the shell.
osh.c:
C: Changed how the shell does parameter substitution.
This was the last major incompatibility w/ the Thompson shell.
Now, substitution is done *before* any command-line parsing
takes place.
* Added the ability for login shells to read the initialization files
/etc/osh.login and/or $HOME/.osh.login if they exist. A shell is
considered to be a login shell if its first argument starts w/ a
`-' character (e.g., -osh).
* Added a `set' command to allow shell compatibility to be toggled
at run time. In addition, the shell now checks for `OSH_COMPAT'
in the environment to tell future invocations of the shell which
mode the user wishes to run in.
* In addition to the `set' command mentioned above, the following
special built-in commands have been added and are available when
the shell is in "noclone" mode:
exec, setenv, trap, unsetenv
* In globargs(), use `gl_pathc == 0' to detect an unmatched pattern
instead of checking if glob(3) returned `GLOB_NOMATCH'. This allows
for those cases where glob(3) may not be POSIX-compliant.
* Made osh command files that are run asynchronously ignore interrupts.
For example, `osh runcom&' should ignore SIGINT and SIGQUIT, and now
it does.
* Reverted a so-called compatibility fix made in osh-040421.
Now, ignore SIGINT and SIGQUIT for asynchronous commands invoked
from a command file. From a usability perspective, it is simply
too annoying to not do it this way.
* Changed the `exit' command so that it always terminates a shell when
reading commands from a file. Previously, it only terminated a shell
when invoked as `osh file'. Note that `exit' still has no effect for
interactive shells or `osh -c command'; this is intentional as it is
compatible w/ the behaviour of `exit' under the Thompson shell.
if.c:
* Include stdlib.h for exit(3) so that OS X doesn't complain.
* Rename exp() to expr() to avoid conflicts w/ exp(3) on OS X.
Strange, as math.h is not included there should not have been
any conflict. Oh well, it is fixed now.
* Enable this utility to return a meaningful exit status to the user.
In short, `if foo = foo' returns an exit status of 0; `if foo = bar'
returns an exit status of 1. Previously, exit status was always 0.
* Made the usage less ambiguous; corrected the documentation to reflect
the actual behaviour. In short, usage is (and always has been) as
follows:
if expr [command [arg ...]]
* Added some useful conditional primaries for constructing expressions.
See the manual pages for details.
* Use the stdio(3) functions instead of write(2) for printing
the error messages.
* In addition, added some useful diagnostic messages which were
inspired by the test(1) utility from Seventh Edition Unix.
goto.c:
* Give an error message when standard input is not seekable.
Previously, a label not found error would be produced instead.
fd2.c:
* A new utility and manual page... It is an adaptation of the PWB/Unix
(roughly PWB/1.0 ?) redirect diagnostic output command. The original
source came from the file `spencer_pwb.tar.gz' which can be found at:
http://www.tuhs.org/Archive/PDP-11/Distributions/usdl/
This closes PR pkg/25314.
Changes:
* Made various changes to hopefully improve the clarity.
Added COMPATIBILITY, HISTORY, and NOTES sections.
* Made changes to how the shell handles terminating `\' characters
w/ the `-c' and `-t' flags. This is a simple extension of the
same behaviour exhibited when the shell is interactive or when it
executes a command file, the only difference being that where a
terminating `\' character causes the shell to read the next line
of input in an interactive shell or command file, w/ the `-c' and
`-t' flags the shell terminates w/o executing the command line.
* Allow parameter substitution w/ the `-c' and `-t' flags. This
feature is not documented. For example, invoking the shell as
follows allows parameter substitution to take place:
% osh -t one two three
echo $0 $1 $2 $3
-t one two three
* The shell now ignores SIGINT and SIGQUIT when the `-c' or `-t' flag
is used. Thus, asynchronous commands invoked in this way ignore
interrupts as they should.
* Ignore SIGINT and SIGQUIT for all commands started from asynchronous
subshells. For example, `( sleep 300; some_command ) >outfile&' now
ignores `^C' and `^\' as it should.
* Don't ignore SIGINT and SIGQUIT for asynchronous commands started
in command files. If a command file is terminated by one of these
signals, the asynchronous commands should also terminate.
* Reverted a change made to termination reporting in osh-040216 that
was not actually compatible w/ the V6 shell.
* Always terminate the shell when read(2) fails.
This fixes a possible infinite loop.
* Fixed a bug in the parser that caused syntactically incorrect
subshell commands not to be detected as such when preceded by
redirection arguments (e.g., `<infile >outfile ( | )').
This bug was introduced in osh-040216.
* Fixed possible buffer overflows in substparm(); added a new error
message, "Too many characters", and made other changes necessary to
properly deal w/ the new error condition. This problem had been
present since at least osh-020214.
* Made some changes to how globbing is handled by the shell.
Specifically, glob(3) should only be called when an argument contains
unquoted occurrences of any of the glob characters `*', `?', or `['.
Previously, it was being called for every argument of an external
command. This change improves run-time performance slightly as
judged by time(1) and information returned by getrusage(2).
* The above-mentioned change also allows the following compatibility
feature. Added globbing compatibility when the shell is compiled
w/ -DCLONE so that when no matches are found a diagnostic,
"No match", is printed.
have it be automatically included by bsd.pkg.mk if USE_PKGINSTALL is set
to "YES". This enforces the requirement that bsd.pkg.install.mk be
included at the end of a package Makefile. Idea suggested by Julio M.
Merino Vidal <jmmv at menta.net>.
as shells/osh.
Osh is a re-implementation of the old and obsolete shell version,
which was in standard use up to UNIX 6th Edition and was supplied
as osh with UNIX 7th Edition. Its command language is a sparse
subset of those of modern shells and is mostly common both to sh(1)
and csh(1).
This pkgsrc entry is based on the FreeBSD ports entry for osh.
