/ngs

Next Generation Shell (NGS)

Primary LanguageCGNU General Public License v3.0GPL-3.0

NGS logo - cloud and UNIX shell icon

Discord Commits in dev since last release Build Status Docker Build Status Homebrew snapcraft license

Next Generation Shell is a modern programming language for DevOps.

The Problem

bash or Python? That's a square pegs for a round hole situation. Both are inadequate for Ops tasks. NGS aims to reduce your frustration and let you be more productive. There is no reason that you should be in this situation. (Python here represents also other general purpose programming languages).

bash does not meet any modern expectations for syntax, error handling nor has ability to work with structured data (beyond arrays and associative arrays which can not be nested). Let it go. You are not usually coding in assembly, FORTRAN, C, or C++, do you? They just don’t match the typical Ops tasks. Don’t make your life harder than it should be. Let it go. (Let’s not make it a blanket statement. Use your own judgement when to make an exception).

Python along with many other languages are general purpose programming languages which were not intended to solve specifically Ops problems. The consequence is longer and less readable scripts when dealing with files or running external programs, which are both pretty common for Ops.

See bash or Python? The Square Pegs and a Round Hole Situation blog post.

Shell user interface, the atrocity. Current shells as well as proposed alternatives treat UI as if nothing happened since the 70-s: mostly typing commands and getting some text back. Watch Unix shell - We can do better now 15 minutes video about what's wrong with the user interface and how NGS is fixing (WIP) that.

Suggested Solution - NGS

I like helping people. Reducing frustration and making people more productive just feels good. A scalable way to help is thorugh tools. That is why I created NGS. It is a programming language with typical Ops tasks in mind. See the use cases below. The next big planned part is the interactive shell.

NGS Use Cases

You as the Ops person, probably wondering whether NGS would be a good fit for your tasks. NGS aims to be best fit for the use cases below.

  • Data manipulation: Got some JSON from an API call / external program / stdin and want to do something with it? Extract, modify, shove it into next API call / external program / stdout.
  • Typical Ops Scripting
    • Run external programs
      • NGS has syntax for pipes and i/o redirection
      • Use built-in Argv facility for convenient building of command line arguments
      • Let NGS built-in mechanism to throw exceptions when exit codes indicate error
      • Let NGS built-in mechanism to parse commands' output
      • Easily pipe data in and out of external program
    • Easily output JSON or a human-readable nicely formatted table
    • Use built-in debug() function which prints debug information, conditionally, depending on environment variable per-section setting
    • Use built-in status() reporting function (previous status is overridden with new status on screen)
    • Use built-in log() function which prints timestamped output
    • Use built-in retry() and stop re-implementing it hundreds of times
  • Testing: test CLI programs and API endpoints using built-in testing mini-framework
  • WIP/PoC: AWS library
    • Easy integration with existing infra
    • Create new or modify existing infrastructure
    • Easier scripting as opposed to declarative tools (CloudFormation, Terraform) allows modeling of processes and not just final state
  • Future: interactive shell with UI which is not from the 70s.

See Use Cases wiki page for more information about the use cases and examples.

Project Status

The language is very useful. See the the bin folder for examples. NGS is used in Beame.io for miscellaneous scripting such as testing CLI tools, performance tests orchestration, cloud manipulation, etc.

The work on the shell has just started. It is not usable yet. The shell is being implemented in NGS. See the design document.

From George Nachman, creator of iTerm2:

Neat! This is totally the project I would do if I had unlimited free time :) I've wished for something like this for a long time. I think there's a lot of room for improvement over a basic TTY, and you've nailed some of the important things

Installing

Using Script

On Linux: make sure curl and sudo are installed

On MacOS: make sure you have brew installed.

curl https://ngs-lang.org/install.sh | bash
ngs -pi 'sum(0..10)'

Using Homebrew

brew install ngs

Using Snap

sudo snap install ngs

Using Docker

Docker repository is at https://hub.docker.com/r/ngslang/ngs/

docker run -it --rm ngslang/ngs

Using GitHub Action

Add to your Github Action the following (make sure to release the version as required)

steps:
	- uses: ngs-lang/ngs@v0.2.13

After that, ngs can be used by simply calling ngs in any run step

steps:
	- uses: ngs-lang/ngs@v0.2.13
	- run: ngs -pi 'sum(10..100)'

Running NGS as AWS Lambda Custom Runtime

To run NGS as AWS Lambda Function see https://github.com/ngs-lang/ngs-aws-lambda

Using iPython or Jupyter Notebook

Please refer to extension located in https://github.com/ngs-lang/ngs-ipython-extension

Using AWS Lambda

Please refer to extension located in https://github.com/ngs-lang/ngs-aws-lambda

Manually Compiling and Running

Clone from Git

git clone https://github.com/ngs-lang/ngs.git

# or if using git over ssh
git clone git@github.com:ngs-lang/ngs.git

cd ngs

Install with Dependencies - Debian-based Linux and MacOS

./install.sh

Install without Dependencies

sudo make install

Run Tests

make tests

Compile

make build

Build is currently tested with the GitHub action in .github/workflows/build.yml and we currently test the following 64-bit architectures:

system gcc clang
ubuntu 20.04 9, 10, 11 9, 10, 11
debian 11.1 9, 10 9, 11
macOS 10.15 9, 10, 11 10
centos 7 9, 10, 11 --
amazonlinux 2 default (7.3) default (11)
fedora 34 default (11) default (12)
fedora 35 default (11) default (13)
archlinux default (11.1) default (13)

Notes:

  • 32-bit architectures are currently not supported.
  • Debian 11 with gcc 10 is also tested with the following architectures: aarch64, s390x, ppc64le

Run

# If NGS is not installed (from the root of ngs project):
NGS_PATH=lib ./build/ngs SCRIPT_NAME.ngs

# If NGS is installed:
ngs SCRIPT_NAME.ngs

Tested as follows (some time ago):

  • Pandoc 2.2
  • Debian Stretch: gcc 4.8.5 + 4.9.3 + 5, clang 3.6
  • Debian Jessie: gcc 4.8.4 + 4.9.2, clang 3.5
  • Ubunty Trusty: gcc 4.8.4, clang 3.4

If you have troubles compiling, please try to compile the commit tagged tested.

Uninstall

cd build
for i in $(<install_manifest.txt);do rm "$i";done

Debug - Mac

# Debug when stuck (note to self mostly)
killall -SIGSEGV ngs
lldb --core /cores/core.XXXXX

Debug - Homebrew builds

  • Checkout the relevant Homebrew repository (probably a fork, example: https://github.com/alebcay/homebrew-core.git).
  • Copy the original formula from the original repository (at Formula/ngs.rb) somewhere else.
  • In the ngs.rb copy, modify:
    • url "https://github.com/ngs-lang/ngs/archive/v0.2.8.tar.gz" -> url "file:///PATH/TO/ngs.tgz"
    • Add version "0.0.1-dev" (or alike)
  • Development cycle:
    • In a folder just above ngs, run tar czf ngs.tgz --exclude=build ngs
    • brew install -d -s ./ngs.rb (wherever the ngs.rb is)

Build and run docker

# Build the docker
docker build -t ngs .
# Run the container
docker run -it --rm ngs
# Use NGS inside the container
ngs -pi 'sum(0..10)'

Generate Documentation

On Linux

# --- Linux ---
cd doc
rm -r out
mkdir out
./make.ngs out

On Mac, follow the instructions to create case sensitive volume.

# --- Mac --- One time setup ---
cd doc/
ln -s /Volumes/YOUR-VOLUME-NAME out

# --- Mac --- Each time ---
cd doc/
rm -r out/*;
./make.ngs out

Code Examples

Arrays

a = [1, 2, 3]
arr = a.map(X*2) # arr is now [2, 4, 6]
for i in arr {
	echo(i)
}

Hashes

h = {"a": 1, "b1": 2, "b2": 3}
echo(h.filterk(/^b/).mapv(X+10))  # {b1=12, b2=13}

Functions (multimethods) and multi-dispatch

F my_func(x:Int) x*10 # Single expression does not require { ... } syntax

doc This method is documented!
F my_func(s:Str) {
	t = s * 2
	"[" + t + "]" # Last value returned as the result
}

echo(my_func(1))      # 10
echo(my_func("xyz"))  # [xyzxyz]
echo(my_func)         # <MultiMethod with 2 method(s)>

More information about the language and syntax in particular is in ngslang.1

Basic Cloud

NGS has AWS library based on concept of Declarative Primitives

TLDR:

  • top-to-bottom execution
  • resource-level idempotence
  • just a more convenient scripting
    • no dependency graph
    • no state file

This is how an instance can be created using NGS (real working code).

{
	NGS_BUILD_CIDR = '192.168.120.0/24'
	NGS_BUILD_TAGS = {'Name': 'ngs-build'}

	vpc    = AWS::Vpc(NGS_BUILD_TAGS).converge(CidrBlock=NGS_BUILD_CIDR, Tags=NGS_BUILD_TAGS)
	gw     = AWS::Igw(Attachments=[{'VpcId': vpc}]).converge(Tags=NGS_BUILD_TAGS)
	rtb    = AWS::RouteTable(VpcId=vpc).converge(Routes=Present({"DestinationCidrBlock": "0.0.0.0/0", "GatewayId": gw}))
	subnet = AWS::Subnet(VpcId=vpc, CidrBlock=NGS_BUILD_CIDR).converge()

	sg = AWS::SecGroup("ngs-build-sg", vpc).converge(
		Description = "ngs-build-sg"
		Tags = NGS_BUILD_TAGS
		IpPermissions = [ AWS::util::world_open_port(22) ]
	)

	ami = AWS::Image(OwnerId=AWS::AMI_OWNER_DEBIAN, Name=Pfx('debian-jessie-amd64-hvm'), RootDeviceType='ebs', VolumeType='gp2').latest()

	instance = AWS::Instance(
		ImageId = ami
		State = null
		KeyName = ENV.get('AWS_NGS_BUILD_KEY', 'ngs-build')
		SecurityGroups = sg
		SubnetId = subnet
		PublicIpAddress = true
		Tags = NGS_BUILD_TAGS
	).converge(
		State = 'running'
	)

	# Get SSH fingerprint from machine's console
	AWS::add_to_known_hosts(instance, 'PublicIpAddress')
}

Sample Scripts

  • describe ec2 instances. The script has nicely aligned output for humans. It uses stdlib's Table to do output layout and columns configuration. Table handles columns presence and order and it can be configured via environment variable.
  • build chunk of hosts file for a management machine. Hosts named env-role or env-role-N, depending on whether you have one or more machines of specific role in the environment.
  • Race condition and locks demo.

Contributing

  • Fork on GitHub
  • Work on whatever you like, preferably something from the GitHub issues of this project
  • Remember to add entry to CHANGELOG.md
  • Make a pull request (to the dev branch).

If the change is big or involves modifying the syntax, it's better to coordinate with Ilya before you start.

Planned Features

( This section is moving to Wiki and Issues )

See "feature" issues and wiki: https://github.com/ngs-lang/ngs/issues?q=is%3Aissue+is%3Aopen+label%3Afeature

You are welcome to open new issues with feature-request label if there is something you would like to see in NGS.

Cross-system

  • Toaster/script prepare mode/assist
    • After a command is run (successfully?) a key shortcut key would append it (the last command) to a buffer / file.

Development

  • Code completion
  • Variables values shown when editing the commands / code (think ls $a, when the cursor is on $a)
  • Running scripts will once in a while update current line/column in the job info
  • Ability to start tracing already running scripts

Later / Unformed / Unfinished Thoughts

  • BIG: Arguments / etc. - description language. Think Javadoc.
    • Python (and other high level languages) is half-way there with argparse. If a special comment is present meaning the script is "argparse safe", it can be run with the shell replacement for argparse to inspect the arguments.
    • Auto discovery of file arguments: run a command with a unique value for an argument and see if it tries to open such file. Tricky, Dangerous.
    • Auto discovery for jars?
    • Anyway, there must be a way to specify externally argument types and objects (file/url/pid... see bash doc about completion for more types)
    • Learn from interaction. Example curl URL -> curl has argument of type URL.
      • Provide easy access to modify/delete/blacklist learned commands' arguments.
      • Shortcut key to define object type. Example curl [SHORTCUT_KEY_PRESSES] -> menu with object types -> remember the selection.
      • Use same format for learned and pre-defined arguments, allowing easy adding to the shell package and interchange between people.
        • The format (future feature, low priority) will include version detection and which switches are supported in which versions.
        • The format will include how to do completion for specific arguments. Think ec2kill < ec2din ....

How to Run the POC

Following instructions should work (tested on Debian)

cd small-poc
mkdir ssl
cd ssl
openssl req -x509 -nodes -days 365 -newkey rsa:2048 -keyout mysitename.key -out mysitename.crt
cd ..
npm install
nodejs server.js
  • Go to https://127.0.0.1:8443/main.html
    • Commands to try:
      • ls
      • pr - a long process with progress bar
      • sleep - a process that sleeps for 5 seconds
      • fail - a process that fails

Have you heard of project X? How it compares to NGS?

  • All of the shells below

    • ... have no interaction with objects on the screen: if you run a command to describe EC2 instances for example, there is no way to interact with the shown list. Such interaction is a planned feature in NGS.
    • (edit: outdated, this part is on hold) ... have no built-in interaction with a cloud. In NGS follows the principles described in "declarative primitives for the cloud". Declarative primitives are somewhat similar to Chef or Puppet resources. The main difference between declarative primitives and a configuration management resource is that I'm proposing just a library function which can be called when you need it and not a control-grabbing framework. So for example AwsElb(...) function call will make sure the load balancer exists and is configured as described. NGS has work-in-progress AWS library. It's already usable for a subset of AWS resource types.
    • ... have no CLI (planned NGS feature) that is written in the shell language itself.
  • oil shell is a very promising project with motivation similar to that behind NGS. It's too early to tell the differences.

  • fish shell has very nice features and improvements but is still more bash-like. For example it doesn't have nested data structures nor a full-featured programming language.

  • Plumbum makes it easier to call shell commands from python. Too verbose to be used as a shell or shell script. It helps when you need to use python and call external programs. Primary target audience seems to be Python developers, not system engineers.

  • Xonsh Python with bash-like additions. Python is not a domain specific language, making bash-like additions can not bring it to be optimal for system tasks.

  • rc shell is much closer to Bash than to NGS.

  • Es: A shell with higher-order functions . ES and NGS share quite a bit of common ideas. NGS goes further with making a shell a real programming language. ES vs NGS would probably be a matter of personal preference. ES is simpler and been here for a while.

  • elvish features nestable data structures, lambdas and namespacing and is pretty close to general purpose programming language when it comes to the semantics. When compared to NGS, it prefers to extend traditional mechanisms like output capture and pipelines as more expressive programming constructs, and looks a bit more like traditional shells (it is non-POSIX though). It is currently terminal-oriented.

    • Elvish has neat interactive features.
    • The programming language:
      • Simple and consistent (vs NGS' more richer language). Built around the pipes paradigm (vs NGS' two syntaxes, one minimalistic bash-like and the other more suitable for programming).
      • Single space for built-in functions and external programs faces the same problem that bash has with while ...;do ... done, hence joins and splits functions which avoid name clashing.
      • In Elvish nested data structures can flow in pipes. Pipes in NGS currently carry bytes and the need to carry data structures is not felt because in "code" mode this happens naturaly in other ways such as mydata.map(myfunc).
      • New data types can not be defined in Elvish (as opposed to NGS' user data types and mutlti-dispatch methods).
      • External programs that Elvish runs that exit with code other than 0 are converted to exceptions. This is simpler and more consistent approach than NGS takes. NGS has customizable decision system with few sane defaults that knows which exit codes for which programs are exceptions. false for example must return 1 and it's not an exception in NGS. test -f ... that returns 0 or 1 is fine, 2 is syntax error which is converted to exception. See blog post about NGS exit code handling.
  • Windows PowerShell is probably the best thing that ever happened to Windows. I'm not familiar with it enough but here are my points

    • PowerShell is built on top of .NET while NGS is a standalone language (as of writing, NGS will be a shell). In my opinion, PowerShell is an adaptation of .NET for scripting while NGS is built from ground up for scripting. I wrote some bootstrapping script in PowerShell and it felt very inconvenient and weird compared to bash or NGS.
    • Syntax
      • PowerShell has also two syntaxes. They are called parsing modes. These roughly correspond to commands and expression mode of NGS. Compared to NGS, the rules of switching between the two parsing modes are numerous and complex.
      • PowerShell is too verbose by default.
      • NGS syntax is much better in my opinion than syntax of PowerShell.
    • Extending PowerShell is either inconvenient because you have to write in PowerShell which is inconvenient by itself or you have to know C# (or other .NET language?).
    • PowerShell got some things right compared to other shells: structured data and consistent $ in front of variables come to mind.
    • Despite some similarities, writing a script in PowerShell and NGS is a completely different experience. You should try both and pick NGS without any doubt :)
  • Shill - Scripting with Least Privilege . Security focused (capability-based), runs on FreeBSD only (looks like Shill kernel module is required), examples mostly show security features, written in Racket. Not much development since initial commit at 2014. Real world usability is unclear. At this point I assume NGS as a programming language is much more usable.

  • the shok command shell

    • Similarity: opinion regarding current shells and status. I agree with most of the motivation page:
      • We can do much better.
      • Current attempts at solutions do not solve the problem.
      • New shell is needed
      • New programming language
    • Similarity: two-modes syntax. Apparently @nfomon also haven't figured out a way to have one syntax.
    • Difference: Shok shell is in C++. NGS' shell is not implemented yet but it will be in NGS.
    • Difference: Shok has modular design .
    • Status: as of 2016-12-30 the latest commit was over a year ago.
  • Ammonite

    • Similarity: opinion that shells can be much better - "Replacing Bash for the 21st Century", "You think that technology has improved in the last 38 years and a modern systems shell should be better than the shells of our forefathers"
    • Similarity: opinion that shell needs a full-featured programming language.
    • Difference: Scala as the shell language. NGS uses new, domain-specific language. I don't think any amount of tinkering with existing languages (maybe except a lot of it with Lisp) can make these languages as usable for shell as new language that was specifically designed to be a shell language.
    • Difference: Ammonite is JVM based. I think it would be really hard to convince anyone that manages systems to have JVM installed on the managed systems just to run a shell. NGS is written in C and compiles to native binary.
    • Difference: Ammonite's REPL looks very good. NGS does not have a REPL yet.
    • If you are OK with Scala, Ammonite is worth trying. I think Scala is too complicated, especially as a shell language. Looking at HTTP request: val resp = Http("https://api.github.com/repos/scala/scala").asString and val parsed = upickle.json.read(resp.body).asInstanceOf[upickle.Js.Obj]. In NGS that would be parsed=``curl -s "https://api.github.com/repos/scala/scala"``. On the other hand Ammonite-Ops and Ammonite-Shell aim to make common "operations" tasks convenient to handle.
  • sh module for Python: "sh is a full-fledged subprocess replacement for Python 2.6 - 3.5, PyPy and PyPy3 that allows you to call any program as if it were a function".

    • It's Python. If you have some code and you just need to run some commands, it's much better than builtin Python facilities.
    • As modules for other languages - you can't have the convenient syntax for common tasks.
      • sh.wc(sh.ls("-1"), "-l") is really not the same as ls -1 | wc -l. BTW, -1 is not needed because ls does that automatically when when stdout is not a tty (try ls | cat).
      • p = sh.find("-name", "sh.py", _bg=True) is not the same as p = $(find -name sh.py &) (NGS)
      • No redirections syntax so ls >/tmp/dir_contents (bash, NGS) becomes with open("/tmp/dir_contents", "w") as h: sh.ls(_out=h) in sh.
    • Exit codes handling
      • Similar option to NGS: give "ok" exit codes when running a program. What's not "ok" becomes an exception which you can catch (both in sh and NGS).
      • I did not see an option to customize the system so that you define once what's an exception for a specific command and then this logic is used every time when you run the specified command. NGS does have this capability.
  • Rash: The Reckless Racket Shell

    • Similar to NGS, Rash has special commands and expression syntaxes
    • NGS is a language ground-up built for Ops while Rash is built on top of Racket Lisp implementation, which is a general purpose language.

      Racket is a general-purpose programming language as well as the world’s first ecosystem for language-oriented programming. Make your dream language, or use one of the dozens already available, including these ...

  • Shell++

    • Similarity

      Shell++ is a programming language that aims bring features from modern languages, as facility to manipulate data structures, object oriented programming, functional programming and others, to shell script.

    • Differences
      • Shell++ is heavily influenced by Python while NGS is a new language specifically designed for Ops tasks.

        I wanted a language that runs shell commands like Bash, and manipulate data structure with the ease of Python

      • NGS sees "object oriented programming" very differently: types and methods and no classes.
      • Shell++ is not aiming for any UI
      • Shell++ is not intended for anyone else beyond author

        MAKE MY LIFE EASIER AND ONLY THAT

    • Overall impression: if you like Python and you need to do Ops tasks, you should try Shell++. If NGS was heavily influenced by Python, it would look at least somewhat similar to Shell++.
  • Nu Shell - compared on 2020-02-01, Nu Shell commit dcdfa2a866bbf1e5737d9f77a3ef5ba971b10083

    • Similarity
      • Structured data matters
      • Functional programming aspects are in
    • Differences
      • Nu Shell supports several serialization formats while NGS currently works only with JSON.
      • Programming language for the shell. It seems that a programming language in Nu Shell is not a priority. The guess is based on:

Discussion / requests / comments

  • If you are totally unsure - open GitHub issues.
  • Feel free to fork/edit/pull-request this document.