demosh

tl;dr: demosh can run shell scripts or Markdown files in a very interactive way, showing commentary in between running commands, and waiting for you to hit RETURN to proceed before running commands. It was created as a tool for doing live demos of relatively complex things. See testing.md and testing.sh for examples.

To exit demosh, hit Q (capital Q). demosh deliberately ignores signals and won't respond to control-C or control-D, so that the commands it runs don't get confused.

To install demosh: just run pip install demosh! or see INSTALLING.md if you want to install from source.

To use demosh to demo itself: see demo/DEMO.md.

To work on demosh itself: see DEVELOPING.md.

---

demosh is a Demo SHell: it reads shell scripts or Markdown files and executes shell commands from them. However, it can also output commentary from the script, show commands before running them, and pause before (or after) running each command. Pausing and what to show can be controlled by inline comments in the script itself. See testing.md and testing.sh for examples.

Running

Run demosh path-to-script [arg [arg [...]]] as if demosh were itself the shell.

demosh starts out not showing commentary and not being interactive, so that the script can do initialization quietly. Use the @SHOW directive to switch into the fully-interactive mode (and use @HIDE to go back).

Init Scripts and Builtins

On startup, demosh will search for one of four startup files to read:

• .demoshrc in the current directory
• .demoshrc.md in the current directory
• $HOME/.demoshrc
• $HOME/.demoshrc.md

These files are checked in the order above; if any are present, the first one found will be loaded, with .demoshrc files treated as shell scripts and .demoshrc.md files treated as Markdown. The --no-init flag prevents this behavior.

The startup file is treated exactly the same as files supplied on the command line. In particular, @SHOW directives will cause them to go interactive (as discussed below), and anything they define will be available for files on the command line to use.

demosh will also load a set of builtin definitions on startup, unless the --no-builtins flag is present on the command line. Builtins are discussed below with other directives.

Executing and Waiting

When demosh has a command to execute in interactive mode, it will:

• Optionally type the command slowly
• Optionally wait for a RETURN
• Run the command
• Optionally wait for a RETURN

By default, typing the command and waiting for RETURN before execution are enabled, and waiting for RETURN after executing the command is disabled.

While "waiting for RETURN", there are several things you can actually type:

• Hitting space while the command is being typed will type the rest of the command immediately.

• Hitting RETURN while the command is being typed will type the rest of the command immediately and then execute it immediately.

• Hitting Q will quit demosh. This must be a capital Q; lowercase q will do nothing.

• Hitting - will repeat the previous command (note that this currently doesn't work well when executing a macro).

• Hitting + will skip to the next command without executing this one (note that this currently doesn't work well when executing a macro).

When demosh has a command to execute in noninteractive mode, it just executes it.

Directives

When reading from shell scripts, demosh directives look like comments:

#@SHOW
#@HIDE

etc. There must not be any spaces between # and @ -- only #@ can start a directive.

When reading Markdown, directives can still be in comments in bash blocks, but most of the time it works better to put them in Markdown comments:

<!-- @SHOW -->
<!-- @HIDE -->

etc. Note that there is no leading # in this case.

Valid directives:

• @SHOW: start showing commentary, displaying commands before running them, waiting for the user to hit RETURN before running things, etc. This is demosh's fully-interactive-during-a-demo mode.

• @HIDE: the inverse of @SHOW. Just run commands without showing comments or waiting. demosh starts in this mode.

• @SKIP: don't do anything until an @SHOW directive -- don't show commentary and don't run commands. This is mostly a debugging aid: when working out what a demo should contain, it's very handy to have an easy way to skip sections entirely and focus in on the parts that aren't working so smoothly.

• @wait: wait for RETURN before proceeding. This can be useful for e.g. pausing while showing longer blocks of commentary.

• @waitafter: wait after running the next command, as well as before.

• @nowaitbefore: do not wait before running the next command.

• @noshow: do not display the next command.

• @notypeout: don't slowly type the next command, just spit it out all at once.

• @immed or @immediate: don't display the next command and don't wait before or after it. This is a way to run a command inline without showing it to the viewers.

• @print arg [arg [...]]: like the shell's echo command, but it colorizes the output as appropriate. For example:

  #@print # This will be colorized like a comment
  

  will output text that's red like an inline comment. Also, since @print has to be a directive, it will never produce output when running the script using the normal shell.

• @import: see "Imports" below.

• @macro: see "Macros" below.

• @hook: see "Hooks" below.

• @ifhook: see "Hooks" below.

Some "directives" are actually macros defined as builtins:

• @wait_clear waits for RETURN, then clears the screen.

• @browser_then_terminal tries to use hooks named show_browser and show_terminal to wait, show a browser, then return to the terminal. Both must be defined for the macro to have any effect.

• @start_livecast, likewise, tries to use hooks named show_slides and show_terminal to show a slide deck, wait, clear the terminal screen, and show the terminal. Again, both hooks must be defined for the macro to have any effect.

Finally, any other directive will be interpreted as an immediate command, so (in shell mode):

#@foobar

is exactly like

#@immed
foobar

meaning that it will immediately run foobar as a shell command: no command display, no waiting. The main difference is that the one-line version looks like a comment if you're executing the script without demosh, so if you're running your script with bash it will not run foobar at all. The two-line version leaves the foobar command looking like a command even when running without demosh.

Imports

The @import <pathname> directive reads the given pathname and inserts its contents into the input stream. This is mostly a way of getting annoying setup code out of the main script, to make the main script easier for others to read.

You can import Markdown into shell scripts and vice-versa: @import will read its argument as Markdown if the pathname ends in .md, or as shell otherwise.

Macros

Macros allow giving groups of commands simpler names. They are very simple: running a macro is the moral equivalent of just inserting the macro's contents into the script in place of the macro. They don't take arguments and they don't have any dynamic behavior.

The syntax is as follows:

#@macro macro-name
  #@command1
  #@command2
  ...
#@end

Then, later:

#@macro-name

will run command1, command2, etc.

It's not technically necessary to have every command in a macro start with #@; likewise, it's not strictly necessary to invoke the macro with a leading #@. It is usually a good idea, though, since it means that all macros will be no-ops when running the script without demosh.

Hooks

Hooks are a special kind of macro-ish thing that create functions based on environment variables whose names start with DEMO_HOOK_.

#@hook do_the_thing THING

will cause do_the_thing to execute the contents of DEMO_HOOK_THING if that variable is present in the environment, or to be a no-op if it's not set. This provides a simple mechanism to e.g. control a livestreaming setup by setting environment variables, and still have a working script if they're not set.

Note that hooks can appear in macros, and that, again, it's a good idea to only invoke hooks with the #@ prefix so that the calls are ignored if you run the script without demosh.

There is also a special form, @ifhook, for hooks:

#@ifhook hookname
  #@command1
  #@command2
  ...
#@endif

in which the commands enclosed between @ifhook and @endif will be run only if the named hook is not a no-op. This permits, for example, skipping a sequence with multiple @waits if it's not going to have any useful effect, so that the user doesn't have to press RETURN multiple times to continue.

Limitations

demosh has some serious limitations, mostly stemming from two things:

1. Shell syntax is terrifyingly complex. demosh doesn't even try to be a full shell parser.

2. demosh executes commands one at a time, rather than messing about with trying to have a persistent shell process.

Parsing

demosh doesn't even try to fully parse the insanity of shell syntax. Instead, it does things more simply:

When reading shell scripts

• Lines starting with # are considered commentary, and will be displayed in interactive mode. Blank lines are considered commentary, too, but multiple blanks are folded into one.

• Once demosh sees a line that doesn't look like a comment, it reads lines until an unescaped newline that's not inside curly braces is found. This forms a single command.

• "Unescaped" means that the newline is not preceded by a backslash. The curly-brace thing is for shell functions. We do NOT parse quoted strings at present; I don't see the benefit for demo scripts.

When reading Markdown

• Anything outside a code block is considered commentary. Multiple blank lines will be folded into one.

• Within a code block marked as either bash or sh, lines get parsed as if demo were reading a shell script.

Processing

Since demosh executes commands in independent subshells, it has to do some complex stuff to make things appear to be a single shell session.

1. demosh keeps track of environment variables on its own. Immediately after reading a command, anything of the form ${VARNAME} is interpolated with the value from demosh's environment.

   NOTE WELL: only the curly-brace form is interpolated. The unbraced form $FOO and complex forms like ${FOO:-default} or the like are not handled, because that way lies madness without a lot more work in the parser.

   Also note that positional variables ($1 etc) will be taken from demosh's command line itself: see below for more.

2. If the command looks like an environment variable assignment, we update demosh's environment internally. This is the environement passed to each command being executed.

   "Looks like an assignment" means that it has optional whitespace, then an identifier followed immediately by an equals sign, then optional content.

3. If the command looks like a shell function definition, we save the line and prepend it to all subsequent shell commands. Again, separate subshells. demosh does not prepend functions to the cd command or variable assignments.

   Only the identifier () { and function identifer () { forms of function definitions are supported.

4. If the first word of the command is cd, we handle that in the demosh itself.

Assignments and cd are both handled by running the command in a new shell and then having the shell echo the result back to us, so that the shell can manage more complex variable expansions, etc.

Why do things like this? Because it's pretty simple and it generally works for demos. An alternative would be to use a single shell and drive it using a pty and multiple threads, but signal handling is much harder in that world.

Environment variables

demosh maintains the environment passed to each command. When demosh starts, it saves its own command line in this enviroment as the positional parameters:

• $0 is the script passed to demosh;
• $1 etc. are command-line parameters after the script; and
• $SHELL is demosh itself (as a fully-qualified path).

Signal Handling

When executing a command, you can use INTR (usually control-C) as usual to interrupt the command. demosh ignores SIGINT and SIGTERM so that you can't accidentally interrupt demosh itself.

License and Copyright

demosh is copyright 2022 Buoyant, Inc., and is licensed under the Apache License Version 2.0. For more information see the LICENSE file.