Cmds

Cmds tries to make it easier to read, write and remember using shell commands in Ruby.

It treats generating shell the in a similar fashion to generating SQL or HTML.

Best read at

http://www.rubydoc.info/gems/cmds/

where the API doc links should work and you got a table and contents.

Ya know, before you get too excited...

It's kinda starting to work. I'll be using it for stuff and seeing how it goes, but no promises until 1.0 of course.

MIT

Or, "what's it look like?"...

• Instead of

  `psql \
    --username=#{ (db_config['username'] || ENV['USER']).shellescape } \
    #{ db_config['database'].shellescape } \
    < #{ filepath.shellescape }`

  write

  Cmds 'psql %{opts} %{db} < %{dump}',
    db: db_config['database'],
    dump: filepath,
    opts: {
      username: db_config['username'] || ENV['USER']
    }

  to run a command like

  psql --username=nrser that_db < ./some/file/path

  Cmds takes care of shell escaping for you.

• Instead of

  `PGPASSWORD=#{ config[:password].shellescape } \
    pg_dump \
      --username=#{ config[:username].shellescape } \
      --host=#{ config[:host].shellescape } \
      --port=#{ config[:port].shellescape } \
      #{ config[:database].shellescape } \
    > #{ filepath.shellescape }`

  which can be really hard to pick out what's going on from a quick glance, write

  Cmds.new(
    'pg_dump %{opts} %{database}',
    kwds: {
      opts: {
        username: config[:username],
        host: config[:host],
        port: config[:port],
      },
      database: config[:database],
    },
    env: {
      PGPASSWORD: config[:password],
    },
  ).stream! { |io| io.out = filename }

  I find it much easier to see what's going on their quickly.

  Again, with some additional comments and examples:

  # We're going to instantiate a new {Cmds} object this time, because we're
  # not just providing values for the string template, we're specifying an
  # environment variable for the child process too.
  # 
  cmd = Cmds.new(
    # The string template to use.
    'pg_dump %{opts} %{database}',
    kwds: {
      # Hashes will automatically be expanded to CLI options. By default,
      # we use `--name=VALUE` format for long ones and `-n VALUE` for short,
      # but it's configurable.
      opts: {
        username: config[:username],
        host: config[:host],
        port: config[:port],
      },
      # As mentioned above, everything is shell escaped automatically
      database: config[:database],
    },
    # Pass environment as it's own Hash. There are options for how it is
    # provided to the child process as well.
    env: {
      # Symbol keys are fine, we'll take care of that for you
      PGPASSWORD: config[:password],
    },
  )
  
  # Take a look!
  cmd.prepare
  # => "PGPASSWORD=shhh\\! pg_dump --host=localhost --port=5432 --username=nrser blah"
  
  # Now stream it. the `!` means raise if the exit code is not 0
  exit_code = cmd.stream! { |io|
    # We use the block to configure I/O. Here we send the standard output to
    # a file, which can be a String, Pathname or IO object
    io.out = filename
  }

Add this line to your application's Gemfile:

gem 'cmds'

And then execute:

bundle install

Or install it globally with:

gem install cmds

Cmds is based around a central {Cmds} class that takes a template for the command and a few options and operates by either wrapping the results in a {Cmds::Result} instance or streaming the results to IO objects or handler blocks.

Templates are processed with "Embedded Ruby" (eRuby/ERB) using the Erubis gem.

For how it works check out

1. {Cmds::ERBContext}
2. {Cmds::ShellEruby}
3. {Cmds#render}

1. Use the args array made available in the templates with entry indexes.

   Example when constructing:

   Cmds.new(
     'cp <%= args[0] %> <%= args[1] %>',
     args: [
       'source.txt',
       'dest.txt',
     ],
   ).prepare
   # => "cp source.txt dest.txt"

   This will raise an error if it's called after using the last positional argument, but will not complain if all positional arguments are not used.

2. Use the arg method made available in the templates to get the next positional arg.

   Example when using "sugar" methods that take args as the single-splat (*args):

   Cmds.prepare  'cp <%= arg %> <%= arg %>',
                 'source.txt',
                 'dest.txt'
   # => "cp source.txt dest.txt"

Just use the key as the method name.

When constructing:

Cmds.new(
  'cp <%= src %> <%= dest %>',
  kwds: {
    src: 'source.txt',
    dest: 'dest.txt',
  },
).prepare
# => "cp source.txt dest.txt"

When using "sugar" methods that take kwds as the double-splat (**kwds):

Cmds.prepare  'cp <%= src %> <%= dest %>',
              src: 'source.txt',
              dest: 'dest.txt'
# => "cp source.txt dest.txt"

If possible, avoid naming your keys:

• arg
• args
• initialize
• get_binding
• method_missing

If you must name them those things, don't expect to be able to access them as shown above; use <%= @kwds[key] %>.

Normally, if you try to interpolate a key that doesn't exist you will get a KeyError:

Cmds.prepare "blah <%= maybe %> <%= arg %>", "value"
# KeyError: couldn't find keys :maybe or "maybe" in keywords {}

I like a lot this better than just silently omitting the value, but sometimes you know that they key might not be set and want to receive nil if it's not.

In this case, append ? to the key name (which is a method call in this case) and you will get nil if it's not set:

Cmds.prepare "blah <%= maybe? %> <%= arg %>", "value"
# => "blah value"

Cmds.prepare "blah <%= maybe? %> <%= arg %>", "value", maybe: "yes"
# => "blah yes value"

Cmds automatically shell-escapes values it interpolates into templates by passing them through the Ruby standard libray's Shellwords.escape.

Cmds.prepare "cp <%= src %> <%= dest %>",
  src: "source.txt",
  dest: "path with spaces.txt"
=> "cp source.txt path\\ with\\ spaces.txt"

It doesn't always do the prettiest job, but it's part of the standard library and seems to work pretty well... shell escaping is a messy and complicated topic (escaping for which shell?!), so going with the built-in solution seems reasonable for the moment, though I do hate all those backslashes... they're a pain to read.

You can render a raw string with <%== %>.

To see the difference with regard to the previous example (which would break the cp command in question):

Cmds.prepare "cp <%= src %> <%== dest %>",
  src: "source.txt",
  dest: "path with spaces.txt"
=> "cp source.txt path with spaces.txt"

And a way it make a little more sense:

Cmds.prepare "<%== bin %> <%= *args %>",
  'blah',
  'boo!',
  bin: '/usr/bin/env echo'
=> "/usr/bin/env echo blah boo\\!"

Render multiple shell tokens (individual strings the shell picks up - basically, each one is an entry in ARGV for the child process) in one expression tag by prefixing the value with *:

Cmds.prepare  '<%= *exe %> <%= cmd %> <%= opts %> <%= *args %>',
              'x', 'y', # <= these are the `args`
              exe: ['/usr/bin/env', 'blah'],
              cmd: 'do-stuff',
              opts: {
                really: true,
                'some-setting': 'dat-value',
              }
# => "/usr/bin/env blah do-stuff --really --some-setting=dat-value x y"

ARGV tokenization by the shell would look like:

[
  '/usr/bin/env',
  'blah',
  'do-stuff',
  '--really',
  '--some-setting=dat-value',
  'x',
  'y',
]

• Compare to without splats:

  Cmds.prepare  '<%= exe %> <%= cmd %> <%= opts %> <%= args %>',
                'x', 'y', # <= these are the `args`
                exe: ['/usr/bin/env', 'blah'],
                cmd: 'do-stuff',
                opts: {
                  really: true,
                  'some-setting': 'dat-value',
                }
  # => "/usr/bin/env,blah do-stuff --really --some-setting=dat-value x,y"

  Which is probably not what you were going for... it would produce an ARGV something like:

  [
    '/usr/bin/env,blah',
    'do-stuff',
    '--really',
    '--some-setting=dat-value',
    'x,y',
  ]

You can of course use "splatting" together with slicing or mapping or whatever.

All of ERB is available to you. I've tried to put in features and options that make it largely unnecessary, but if you've got a weird or complicated case, or if you just like the HTML/Rails-esque templating style, it's there for you:

cmd = Cmds.new <<-END
  <% if use_bin_env %>
    /usr/bin/env
  <% end %>
  
  docker build .
    -t <%= tag %>
    
    <% if file %>
      --file <%= file %>
    <% end %>
    
    <% build_args.each do |key, value| %>
      --build-arg <%= key %>=<%= value %>
    <% end %>
    
    <% if yarn_cache %>
      --build-arg yarn_cache_file=<%= yarn_cache_file %>
    <% end %>
END

cmd.prepare(
  use_bin_env: true,
  tag: 'nrser/blah:latest',
  file: './prod.Dockerfile',
  build_args: {
    yarn_version: '1.3.2',
  },
  yarn_cache: true,
  yarn_cache_file: './yarn-cache.tgz',
)
# => "/usr/bin/env docker build . -t nrser/blah:latest
#       --file ./prod.Dockerfile --build-arg yarn_version=1.3.2
#       --build-arg yarn_cache_file=./yarn-cache.tgz"
# (Line-breaks added for readability; output is one line)

Cmds also supports a printf-style short-hand. Sort-of.

It's a clumsy hack from when I was first writing this library, and I've pretty moved to using the ERB-style, but there are still some examples that use it, and I guess it still works (to whatever extent it ever really did), so it's probably good to mention it.

It pretty much just replaces some special patterns with their ERB-equivalent via the {Cmds.replace_shortcuts} method before moving on to ERB processing:

And the escaping versions, where you can put anothe % in front to get the literal intead of the subsitution:

That's it. No printf formatting beyond besides s (string).

you can provide three types of arguments when executing a command:

1. positional arguments for substitution
2. keyword arguments for substitution
3. input to stdin

all Cmds instance execution methods have the same form for accepting these:

1. positional arguments are provided in an optional array that must be the first argument:

   Cmds "cp <%= arg %> <%= arg %>", [src_path, dest_path]

   note that the arguments need to be enclosed in square braces. Cmds does NOT use *splat for positional arguments because it would make a Hash final parameter ambiguous.

2. keyword arguments are provided as optional hash that must be the last argument:

   Cmds "cp <%= src %> <%= dest %>", src: src_path, dest: dest_path

   in this case, curly braces are not required since Ruby turns the trailing keywords into a Hash provided as the last argument (or second-to-last argument in the case of a block included in the method signature).

3. input and output is handled with blocks:

   `Cmds(“wc -l”){ “one\ntwo\nthree\n” }

   Cmds.stream './test/tick.rb <%= times %>', times: times do |io| io.on_out do |line| # do something with the output line end

   io.on_err do |line| # do something with the error line end end`

playbook = Cmds.new "ansible-playbook -i %{inventory} %{playbook}"
playbook.call inventory: "./hosts", playbook: "blah.yml"

currying

dev_playbook = playbook.curry inventory: "inventory/dev"
prod_playbook = playbook.curry inventory: "inventory/prod"

# run setup.yml on the development hosts
dev_playbook.call playbook: "setup.yml"

# run setup.yml on the production hosts
prod_playbook.call playbook: "setup.yml"

NEEDS TEST

can be accomplished with reuse and currying stuff

playbook = Cmds.new "ansible-playbook -i %{inventory} %{playbook}", inventory: "inventory/dev"

# run setup.yml on the development hosts
playbook.call playbook: "setup.yml"

# run setup.yml on the production hosts
prod_playbook.call playbook: "setup.yml", inventory: "inventory/prod"

c = Cmds.new("wc", input: "blah blah blah).call

want to be able to use to exec commands

kinda like sprintf formatters or string escape helpers in Rails, they would be exposed as functions in ERB and as format characters in the shorthand versions:

Cmds "blah <%= j obj %>", obj: {x: 1}
# => blah \{\"x\":1\}

Cmds "blah %j", [{x: 1}]
# => blah \{\"x\":1\}

Cmds "blah %<obj>j", obj: {x: 1}
# => blah \{\"x\":1\}

the s formatter would just format as an escaped string (no different from <%= %>).

other formatters could include

• j for JSON (as shown above)
• r for raw (unescaped)
• l or , for comma-separated list (which some commands like as input)
• y for YAML
• p for path, joining with File.join