psh - Perl SHell
psh [options] [file]
Copyright (C) 1999 Gregor N. Purdy. All rights reserved. This script is free software. It may be copied or modified according to the same terms as Perl itself.
psh is a Perl program which executes a read-eval loop with enough options so that general behavior reasonably similar to more traditional shells like 'sh' or 'bash' can be achieved, while still allowing arbitrary perl expressions to be evaluated.
By default within psh, the Perl -w flag and 'use strict' are not employed so that the user is not bound by their stipulations.
They can both be turned on via a command-line flag; or setting
$^W = 1 will turn on warnings, and calling 'use strict' will (almost) do the usual thing if called by the user (see LIMITATIONS,
below).
The command-line arguments to psh are:
psh [-d] [-w] [-r RC_FILE] [-c STRING ] [FILE1 FILE2 ....]
They are processed in the following order, regardless of what order they are specified in:
Enables Perl's warning mode. The -w switch runs perl with the -w switch and ``use strict;''.
The -d option puts psh into ``debugging'' mode, which prints diagnostic output, including
primarily the result of the ``which'' builtin before every execution. Note
that you can also enter/leave this debugging mode in a running psh via the $Psh::debugging variable.
The -r option specifies a file of commands to be read in and evaluated before
processing begins. If it is not set, and
$ENV{HOME} is set, and the file $ENV{HOME}/.pshrc is present, it will be used. If -r is not specified and the current directory is different from $ENV{HOME} and it contains a .pshrc file, that file will be read and executed in addition to
$ENV{HOME}/.pshrc.
If the -c flag is present, then commands are read from
string, and then psh exits. In particular, any FILE1 ... arguments will be ignored.
If any FILE1 ... arguments are specified on the command line, they will be read and executed and then psh will exit. Otherwise, psh will enter an interactive command loop.
Each line of input is read. psh knows a number of possible strategies for evaluating the line, such as
``send it to system() if it starts with the name of an executable visible in $ENV{PATH}``. (See below for a complete list.) Each strategy in turn (from a
user-definable list) examines the command line to see if it can apply, and
the first matching strategy evaluates the line. There is a psh configuration variable (see below) which controls whether the perl value of
the evaluation is saved and printed after each command.
psh automatically collects several lines of input into a unit processed as a
single line if there are unfinished Perl constructs on the line. In
particular, if there is an unmatched quote, paren, brace, or square
bracket, input is read until these characters match. If an input line
contains the Perl ``here document'' construct as in
<<XXX, (anywhere on the line), then input is read and accumulated until XXX occurs on a line by itself. Then the accumulated input is processed as if
it were a single line.
Some evaluation strategies examine the ``words'' of the input. These are produced by a tokenizer which behaves very similarly to traditional shells: words are broken at whitespace, '&' is a metacharacter which means that it always forms its own word, and backslash and double and single quotes act as quoting characters, preventing word breaks at whitespace and the ``meta-ness'' of &.
If the description of the strategy does not mention the ``words'', then the tokenization is irrelevant to that strategy.
psh includes the following evaluation strategies, sorted by the order we suggest for @Psh::strategies.
If the first word of the input line begins with a '#' character, ignore the line.
If the first word of the input line begins with a '!' character, send
everything after the '!' to system().
If the first word of the input line begins with a '{' character, evaluate the entire line as a Perl expression (including the brace).
If the first word of the input line matches a psh ``built-in'' function, call the subroutine associated with that built-in; the subroutine receives a single argument, which is the remainder of the input line exactly as entered.
If the first word of the input line matches either a built-in Perl function
(as determined by the %Psh::perl_builtins hash), or the name of defined
Perl subroutine, pass the line to eval. If $Psh::perlfunc_expand_arguments is true and the line contains no parens, or braces or commas (except for
{a,b,c} as in shell brace-expansion), then this strategy tries to interpret
the arguments on the command line in a ``shell-like'' manner: strings are
literal except for variable expansion, brace expansion, and glob expansion.
The idea of this strategy is to allow perl functions, especially subroutines in main, to be called like ``ordinary commands'' (i.e., executables on disk). Or put another way, the idea is to replace bash's ``shell function'' capacity with ordinary Perl subroutines. The slogan is, ``If the command line looks like an ordinary shell command, interpret it like one, even if the first word is a Perl subroutine.''
If the input line matches the name of a stopped job then brings that job to the foreground instead of starting a new programm with that name.
If (1) the first word of the input line matches the name of a file found in
one of the directories listed in the path ($ENV{PATH}), and (2) that file starts with #!/.../perl, and (3) that perl is the same as the Perl under which psh is running, psh will fork and run the script using the already-loaded Perl interpreter. The
idea is to save the exec half of the fork-exec that the executable strategy
would do; typically the exec is more expensive. Right now this strategy can
only handle the -w command-line switch on the #! line. Note this strategy only makes sense before the ``executable''
strategy; if it came after, it could never trigger.
If the first word of the input line matches the name of an executable file
in the path given by $ENV{PATH}, then pass the line to system. Perl variable substitution will be done on
the line first if the $Psh::executable_expand_arguments configuration variable is true.
If the first word of the input line is a ``fallback builtin'' provided for operating systems that do not have common binaries -- such as ``ls'', ``env'', etc, then call the associated subroutine like an ordinary builtin. If you want all of these commands to be executed within the shell, you can move this strategy ahead of executable.
Pass the line to eval, regardless of any condition. This is a catch-all strategy; strategies placed after it will never be triggered.
The list of evaluation strategies to try is kept in the configuration
variable @Psh::strategies, which defaults to
qw(comment bang brace built_in executable fallback_builtin eval). If you want to ensure that ``print'' (for example) refers to the Perl
function ``print'' and not ``/usr/bin/print'', try adding the ``perlfunc''
strategy before ``executable''.
The following functions are provided as built-in functions of
psh. You can add an additional built-in called 'foo' by setting the hash $Psh::built_ins{foo} to a reference to a subroutine. When a line like ``foo bar $baz'' is then evaluated by the 'built_in' strategy, the string 'bar $baz' will be passed to your subroutine as its only argument. This is precisely
the difference between making 'foo' a built-in as opposed to simply defining a procedure main::foo and letting the eval strategy call foo: the built-in subroutine will get the precise input string entered,
whereas the ordinary procedure will of course get arguments that have
already been evaluated by the Perl evaluator. Which you want depends on the
nature of the function being defined.
If $Psh::built_ins{foo} does not exist, but
&Psh::Builtins::bi_foo is defined, then foo will still be considered a builtin calling that
function. This is primarily to ensure that the original builtins are always
accessible, for example if they have been aliased and unaliased.
``Fallback builtins'' are provided similarly to this last option: ``foo''
is a fallback builtin if &Psh::Builtins::Fallback::bi_foo is defined.
For a list of the predefined psh builtins, see Psh::Builtins or use the ``help'' command from within psh.
psh makes a number of variables and functions accessible to the user in the Psh:: package for configuration or utility purposes. Their default values are
given in parentheses below. If the variable is also marked ``[late]'', then
it is undefined at the start of the .pshrc file, but any value given to it during that file will be used instead of
the default setting.
The name of the current shell.
The command serial number in the currently-executing processing loop.
The pid of the process psh will currently forward signals to, or 0 if psh will handle the signals internally. Usually 0 unless psh is waiting for a process in the ``foreground''.
Whether psh's internal debugging output should be produced.
Controls whether the processing loop saves and displays the Perl value of executing a line of input. Three cases are distinguished: a false value, in which case no results are displayed; a subroutine reference, in which case the subroutine is called with the results (this may be multiple arguments if the eval returned an array) and should return true to display, false to not; or a true (scalar) value, in which case any non-undef, non-empty value is displayed.
In addition to displaying the value, it is pushed onto the array determined
by $Psh::result_array. Note that scalar values are pushed directly onto this array, but array
values are pushed by reference.
Every Perl expression that psh evaluates as part of its read-eval loop is prepended with this string, intended primarily to set the expected package context.
If this variable is true, then the following procedure is applied to each
argument ARG of the command line before passing the command line on to
``system'': If ARG starts with a single quote, it is untouched. If ARG
starts with a double quote, it is passed to ``eval'', and then double
quotes are placed around the result. Otherwise, double quotes are placed
around ARG, it is passed to eval, and the result is used directly. The
intention of these steps is to do variable substitution of perl variables
that appear in the command line as the user would expect, and in a way
similar to shells such as bash. For example, if @files =
(file1, file2, file3), then 'ls @files' will expand to 'ls file1 file2
file3', whereas 'echo ``@files''' will expand to 'echo ``file1 file2
file3'''.
If set to true, psh will ignore all segfaults and attempt to continue.
If set to true, psh will attempt to continue after internal errors.
The filename psh will use to save the command history in from one invocation to the next, if $Psh::save_history is true.
The number of lines of command history to keep.
The short host name of the machine psh is currently running on.
The fully qualified host name of the machine psh is running on.
A file giving news for the shell.
If this variable is true, then command lines interpreted by the perlfunc strategy which contain no parens, or braces or commas except in bash-like brace-expansion expressions, will not be simply evaluated. Rather, they will be interpreted much like traditional shells do: the line will be spilt into words, which undergo globbing, brace expansion, and variable expansion, and the resulting array of words is passed to the specified function as its arguments.
Controls the prompt string in interactive use, see below.
Just like $Psh::prompt, but for continuation lines when multi-line input is being read.
Controls where the results of Perl evaluations saved via $Psh::echo
will go. It may be a reference to an array, or the string name of an array.
If this is true, the command history is saved in $Psh::history_file
from one invocation of psh to the next.
When Psh::Util::which is asked to locate a filename in the current PATH, it will only look for
filenames which match this regexp. Names that do not match this regexp will
automatically come back ``not found''.
Contains names of a number of net based programs for enabling TAB completion of hostnames/bookmarks
Supposed to contain your most used IP numbers, hostnames or URLs. Those
will be eligible for TAB completion if you use one of the programs in @Psh::Completion::netprograms. psh will initialize this list with your /etc/hosts file
An array of month names used in printing dates, for example in prompt strings.
The list of strategies for executing a line, tried in order.
The default array where psh stores away the results of executing lines, as described in $Psh::echo above.
An array of the weekday names used in printing dates, for example in prompt strings.
Contains a list of environment variables which should be tied to arrays. The key of the hash is the name of the variable, the value is the delimiter of the list (e.g. ':' in PATH). The default value for array_exports currently contains PATH, CLASSPATH, LD_LIBRARY_PATH, FIGNORE and CDPATH.
The keys are command names, the values are subroutine references which are called with the remainder of the line exactly as input.
The perlfunc evaluation strategy uses this hash on the first word of a
command line to determine whether it should handle that command line. If
the first word is present as a key in this hash, then the corresponding
value determines the treatment: 0 means don't handle this line; any
positive value means do handle the line as long as it has at least that
many words. (Since the first word counts, a value of '1' will guarantee
that the perlfunc strategy handles the line.) Useful for example to set $Psh::perl_builtins{grep} = 0 if you don't want ``grep'' to be treated as a perl function.
Note that the perlfunc strategy will also handle the line if the first word is not present as a key in this hash but a Perl subroutine of that name is defined.
The keys of this hash never have their arguments expanded in a shell-like fashion, even if $Psh::perlfunc_expand_arguments is true.
The keys of this hash are single characters, and the values are subroutine references that implement the given escape character in prompt strings. (See PROMPT STRINGS below.)
These hashes implement the named strategies used in psh's evaluation of input lines. It is possible to add new strategies, see below.
Allows you to add own TAB completions. The keys are names of programs to which the TAB completion should apply. The value is either a list reference with at least two values, up to three values, or a sub. If it's a sub it is called with the text to complete, text in front of the text to complete to the last whitespace and the word which triggered the completion.
The first element of the rules list is the character which should be appended after the completion is successful, the second value is again a list reference containing all possible completions. If the optional third value is true, the custom completion will replace all results of standard completions (like filename or username completion), if the value is false, the custom completion just adds additional results.
This is probably complicated but worth the effort.
Example:
%Psh::Completion::custom_completions= ( 'scp'=>['',['server1:/path1','server1:/path2','server2:/path']], 'foo'=>[' ',['bar','bla','fish'],1], 'hmm'=> sub { return ['/',['1','2','3'],1] } );
This function takes a string, evaluates it as if it were a line of
psh input, and returns the value. Useful in loops like:
psh$ for $file (glob $pat) { Psh::evl("ls -ld $file"); }
Returns true if its first argument is a number. Intended for use in filter
subroutines placed in $Psh::echo. For example,
$Psh::echo = \&Psh::is_number; will cause only numeric return values to be printed.
Returns the current news, as a string.
These four functions are called whenever psh wants to produce -d-mode output, error messages, normal output, and warnings, respectively. They could conceivably be redefined to implement logging or similar facilities.
Takes one argument, a package name, and shows all of the symbols in that package.
There are other functions in the psh:: package, but they are probably not useful except internally to psh.
Setting the variable $Psh::prompt to a string will cause that string to be used as the prompt-string. Setting
it to a subroutine reference causes the result of running that subroutine
to be used each time. For example,
$Psh::prompt = sub { $i++; "psh [$i]\$ "; }
will cause the prompt to be psh [1]$ followed by psh [2]$, and so on.
psh uses some of the same ``prompting variables'' as bash. They are accessed by placing a backslash followed by the code in the
prompt string, either hard coded, or as returned by the prompt string
function. The variables supported are:
Custom prompting variables may be added by adding entries to the array
%Psh::prompt_vars keyed by the single character code. The entries should be subroutine
references that return the replacement string.
There are two hashes, %Psh::strategy_which and %Psh::strategy_eval. An evaluation strategy called ``foo'' is implemented by putting a subroutine object in each of these hashes
keyed by ``foo''. The first subroutine should accept a reference to a
string (the exact input line) and a reference to an array of strings (the
array of ``words'' in the input line produced by &Psh::decompose, provided as a convenience so that each individual strategy doesn't have
to recompute this). It should return a string, which should be empty if the
strategy does not apply to that input line, and otherwise should be an
arbitrary non-null string describing how that strategy applies to that
line. It is guaranteed that the string passed in will contain some
non-whitespace, and that the first string in the array of words is
non-empty.
The $strategy_eval{foo} routine accepts the same two first arguments and a
third argument, which is the string returned by $strategy_which{foo}. It
should do the evaluation, and return the result. Note that the
$strategy_eval function will be evaluated in an array context.
Due to limitations of the Win32 type of operating system there's no job control available on those systems.
The loop inside psh will clobber $1 and other variables because it uses matches to implement some of its
special functions.
Very little error checking is done. For example, if you evaluate !blork by the bang strategy and there is no command ``blork'' on your system, you get no feedback whatsoever.
Right now, job control simply assumes that the POSIX interface is fully implemented. There should be a way to turn job control off if this is not the case.
The ``exit status'' of programs invoked in the foreground by the ``executable'' strategy (or even the ``bang'' strategy) isn't available from within psh.
Note that since expressions like 'use foo' return undef when sent to
eval(), it is not possible to use that return value as
indication of an error. Instead, we use the heuristic that there was no
error unless the special Perl variable '$@' is non-empty. Note that the
side effects of 'use foo' as a psh command line appear to be exactly as expected.
Not exactly a psh limitation: Term::ReadLine::Gnu seems to be rather buggy on Linux systems
and leads to continous crashes of psh. Use Term::ReadLine::Perl instead ( set PERL_RL to Perl to disable
Term::ReadLine::Gnu)
psh needs several optional Perl modules to offer full functionality:
Larry Wall exhibits the simple Perl shell while (<>) { eval; print $@; } on page 161 of the Camel Book (2nd Edition).
Rich Graves <rcgraves@brandeis.edu> posted a comment to the original psh-0.001 announcement on http://freshmeat.net, which contained this gem that leverages the Perl debugger: perl -d -e 1;
Hiroo Hayashi <hiroo.hayashi@computer.org> includes perlsh, a ``one-line perl evaluator with line editing function and variable name completion function'' as an example with his Term::ReadLine::Gnu Perl module.
In an example of convergent evolution, at http://jenda.krynicky.cz/
there is a Perl shell module called PSH.pm which is quite similar to this psh. It is designed to provide a command line that can be called inside some
other program via PSH::prompt();, but a small file
psh.pl is also included that uses PSH to provide a standalone shell. Perhaps some
merger of these efforts would be beneficial to all?
Some versions of the Perl faq mention an interactive Perl shell called
SoftList, which can still be found at
http://www.mit.edu/afs/sipb/contrib/perl/SoftList/. It predates Term::Readline and was apparently last touched in 1993, so it
seems to be obsolescent.
Tim Newsome, <nuisance@cmu.edu>, has developed a shell he calls
timtosh (There Is More Than One SHell). Per his web site (http://www.wiw.org/~drz/timtosh), it is a shell written entirely in Perl. The goal is to get a shell which
you can extend in Perl and can do some other niceties related to Perl (like
perl re file matching). As of 1999-12-13 (Perl Shell 0.004 release date),
Tim says timtosh ``is focused quite differently than psh is, but is currently still waiting for a rewrite of the command line
parsing. (It has been for almost a year now)''.
Tom Christiansen and Nathan Torkington's book Perl Cookbook, published by O'Reilly in 1998 (ISBN 1-56592-243-3) has ``Example 15-4. vbsh'' on page 531 for section 15.11 (Editing Input). It stands for Very Bad SHell.
As an aid to comparing/contrasting these different shells, here is a brief table indicating whether or not each has certain features.
Key to features:
PE : Perl evaluation of Perl expressions
SHE: shell-like evaluation of shell-like expressions, including
'exec'ing executables searched for in PATH
CLE: command-line editing
JC : job control
Key to symbols:
* : feature present
- : feature absent
? : don't know
The shells:
Shell Name PE SHE CLE JC psh (this one) * * * * Larry Wall shell * - - - Perl debugger shell * - * - perlsh * - * - Krynicky PSH.pm * * ? - SoftList * ? * ? timtosh - * ? * vbsh ? ? ? ?
psh - The Perl Shell executable script.
.pshrc - The user's Perl Shell `profile'. May be in $HOME or the current directory; if both are present, both will be read in the
order mentioned.
Gregor N. Purdy, <gregor@focusresearch.com>
The following people have contributed to the development of psh:
Markus Peter <warp@spin.de> added job and signal handling, completion code, Win32 port, i18n code, bash compatibility builtins and environment variables and some more minor updates.
Glen Whitney <gwhitney@post.harvard.edu> added evaluation strategies, improved interrupt/job handling, &Psh::evl, $Psh::echo, more extensive documentation, and other more minor features.
Code examples showing how to apply the Term::ReadLine package were
contributed by Billy Naylor <billy.naylor@eu.net> (in his pash.pl program, which is his own Perl shell).
Billy Naylor <billy.naylor@eu.net> also had an example of a symbol table printing function that was used as
the starting point for the psh
function psh::symbols(). The psh version adds the ability to specify a package name, and it also filters out
some special variables. The implementation technique is also different from
Billy's.
Matthew D. Allen <s2mdalle@titan.vcu.edu> contributed an enhanced prompt string handling routine that emulates the bash prompt variables. This was expanded into the form now present.
Allan Kelly <akelly@holyrood.ed.ac.uk> found some problems with the generated documentation.