A modern shell
Project description
What's New
This release revisits the commands and syntax for storing and retrieving streams, in files and variables.
The commands for operating on files are read
and write
. (out
, which was present in earlier
releases, has been renamed to write
). Both of these commands provide support for reading text,
parsing CSV files (with command or tab delimiters), and for pickled data.
The commands for operating on variables are load
and store
. These commands store streams which
cannot be accessed except via pipelines (sequences of commands). For this reason, the internal
format is not specified, and no formatting options are supported.
Marcel has had redirection syntax all along. The symbols >
and >>
were provided to operate
on variables only. To operate on files, the read
and write
(formerly out
) commands had to
be used directly. This was confusing, as the use of >
and >>
for files is so ingrained for
the users of many shells.
(Bash also uses the redirection symbol <
to provide a process's stdin. So in bash you might
write foo < a > b
. The marcel equivalent would be a > foo > b
.)
In this release, the symbols >
and >>
are provided to operate on files, consistent with bash usage.
To operate on variables, the equivalent symbols are >$
and >>$
.
Examples:
- Generate the stream
[0, 1, 2]
and store it in variablestream012
:
gen 3 >$ stream012
The stream in stream012
can now be fed into a pipeline, e.g.
stream012 >$ map (x: (x, x ** (1/2)))
- Generate the stream
[0, 1, 2]
and store it in file/tmp/file012
:
gen 3 > /tmp/file012
There is now a file /tmp/file012
containing
0
1
2
- Generate a sequence of (x, sqrt(x)), for x = 1 through 20, and store it in the variable
f
.
gen 20 1 | (x: (x, x ** (1/2))) >$ f
- Append more square roots, for x = 21 through 30.
gen 10 21 | (x: (x, x ** (1/2))) >>$ f
- Format and print the contents of
f
f >$ write --format 'sqrt({}) = {}'
Marcel
Marcel is a shell. The main idea is to rely on piping as the primary means of composition, as with any Unix or Linux shell. However, instead of passing strings from one command to the next, marcel passes Python values: builtin types such as lists, tuples, strings, and numbers; but also objects representing files and processes.
Linux has extremely powerful commands such as awk
and find
. Most
people know how to do a few simple operations using these commands.
But it is not easy to exploit the full power of these commands
due to their reliance on extensive "sublanguages" which do:
- Filtering: What data is of interest?
- Processing: What should be done with the data?
- Formatting: How should results be presented?
By contrast, marcel has no sublanguages. You use marcel operators combined with Python code to filter data, process it, and control command output.
The commands and syntax supported by a shell constitute a language which can be used to create scripts. Of course, in creating a script, you rely on language features that you typically do not use interactively: control structures, data types, and abstraction mechanisms (e.g. functions), for example. Viewed as a programming language, shell scripting languages are notoriously bad. I didn't think it was wise to bring another one into the world. So marcel takes a different approach, using Python as a scripting language, (see below for more on scripting).
Pipelines
Marcel provides commands, called operators, which do the basic work of a shell.
An operator takes a stream of data as input, and generates another stream as output.
Operators can be combined by pipes, causing one operator's output to be the next operator's input.
For example, this command uses the ls
and map
operators to list the
names and sizes of files in the /home/jao
directory:
ls /home/jao | map (f: (f, f.size))
- The
ls
operator produces a stream ofFile
objects, representing the contents of the/home/jao
directory. |
is the symbol denoting a pipe, as in any Linux shell.- The pipe connects the output stream from
ls
to the input stream of the next operator,map
. - The
map
operator applies a given function to each element of the input stream, and writes the output from the function to the output stream. The function is enclosed in parentheses. It is an ordinary Python function, except that the keywordlambda
is optional. In this case, an incomingFile
is mapped to a tuple containing the file and the file's size.
A pipeline
is a sequence of operators connected by pipes. They can be used directly
on the command line, as above. They also have various other uses in marcel. For example,
a pipeline can be assigned to a variable, essentially defining a new operator.
For example, here is a pipeline, assigned to the variable recent
, which selects
File
s modified within the past day:
recent = [select (file: now() - file.mtime < hours(24))]
- The pipeline being defined is bracketed by
[...]
. (Without the brackets, marcel would attempt to evaluate the pipeline immediately, and then complain because the parameterfile
is not bound.) - The pipeline contains a single operator,
select
, which uses a function to define the items of interest. In this case,select
operates on aFile
, bound to the parameterfile
. now()
is a function defined by marcel which gives the current time in seconds since the epoch, (i.e., it is justtime.time()
).File
objects have anmtime
property, providing the time since the last content modification.hours()
is another function defined by marcel, which simply maps hours to seconds, i.e., it multiplies by 3600.
This pipeline can be used in conjunction with any pipeline yielding files. E.g., to locate
the recently changed files in ~/git/myproject
:
ls ~/git/myproject | recent
Functions
As shown above, a number of operators, like map
and select
, take Python functions as
command-line arguments. Functions can also be invoked to provide function arguments.
For example, to list the contents of your home directory, you could write:
ls /home/(USER)
This concatenates the string /home/
with the string resulting from the evaluation of
the expression lambda: USER
. USER
is a marcel environment variable identifying the
current user, so this command is equivalent to ls ~
.
If you simply want to evaluate a Python expression, you could use the map
operator, e.g.
map (5 + 6)
which prints 11
. In a situation like this, marcel also permits you to omit map
; it is
implied, so this also works:
(5 + 6)
Executables
In addition to using built-in operators, you can, of course, call any executable. Pipelines may contain a mixture of marcel operators and host executables. Piping between operators and executables is done via streams of strings.
For example, this command combines operators and executables.
It scans /etc/passwd
and lists the usernames of
users whose shell is /bin/bash
.
cat
, xargs
, and echo
are Linux executables. map
and select
are marcel operators.
The output is condensed into one line through
the use of xargs
and echo
.
cat /etc/passwd \
| map (line: line.split(':')) \
| select (*line: line[-1] == '/bin/bash') \
| map (*line: line[0]) \
| xargs echo
cat /etc/passwd
: Obtain the contents of the file. Lines are piped to subsequent commands.map (line: line.split(':'))
: Split the lines at the:
separators, yielding 7-tuples.select (*line: line[-1] == '/bin/bash')
: select those lines in which the last field is/bin/bash
.map (*line: line[0]) |
: Keep the username field of each input tuple.xargs echo
: Combine the incoming usernames into a single line, which is printed tostdout
.
Shell Features
Marcel provides:
- Command history: A
history
operator, rerunning and editing of previous commands, reverse search, etc. - Customizable prompts: Configured in Python, of course.
- Tab completion: For operators, flags, and filename arguments.
- Help: Extensive help facility, providing information on concepts, objects, and operators.
- Customizable color highlighting: The colors used to render output for builtin types such
as
File
andProcess
, andhelp
output can be customized too. - Dynamic reconfiguration: Changes to configuration and startup scripts are picked up without restarting.
Scripting
Marcel's syntax for constructing and running pipelines, and defining and using
variables and functions, was designed for interactive usage. Instead of extending
this syntax to a full-fledged scripting language, marcel provides a Python API,
allowing Python to be used as the scripting language.
While Python is
sometimes considered to already be a scripting language, it isn't really.
Executing shell commands from Python code is cumbersome. You've got to use
os.system
, or subprocess.Popen
, and write some additional code to
do the integration.
Marcel provides a Python module, marcel.api
,
which brings shell commands into Python in a much cleaner way. For
example, to list file names and sizes in /home/jao
:
from marcel.api import *
for file, size in ls('/home/jao') | map(lambda f: (f, f.size)):
print(f'{file.name}: {size}')
This code uses the ls
and map
functions, provided by marcel.api
. These correspond to
marcel operators that you can use on the command line. Output from the ls
is a stream
of File
s, which are piped to map
, which maps files to (file, file size) tuples.
ls ... | map ...
defines a pipeline (just as on the command line). The Python
class representing pipelines defines __iter__
, so that the pipeline's output can be
iterated over using the standard Python for
loop.
Installation
To install marcel:
python3 -m pip install marcel
This command installs marcel for the current user. To install for the entire system,
use sudo python3 -m pip install --prefix ...
instead. (The value of the --prefix
flag should
be something like /usr/local
.)
Marcel depends on dill. This package will be installed automatically if needed, when marcel is installed via pip.
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.