alpha-build-core 0.1.0

AlphaBuild's core

AlphaBuild

AlphaBuild is a simple monorepo build tool based on Make with out-of-the-box support for numerous tools in the Python, Bash, Jupyter Notebooks, Markdown, YAML ecosystems and with a strong focus on extensibility.

The way AlphaBuild works draws inspiration heavily from monorepo build tools such as Pants, Bazel, Buck. It can run at once multiple linters, formatters, type checkers, hermetic packers, testing frameworks, virtual environment managers etc.

1. Platforms
2. Usage
   1. Goals - what tools we run
   2. Targets - what files we run the tools on
3. AlphaBuild structure
4. IDE Integration
5. Example monorepo running AlphaBuild
6. Common admin actions
   1. Installation
   2. CI/CD Setup
   3. Upgrade
   4. Change goal definitions
   5. Add goals
   6. Update PYTHONPATH
   7. See/Change tools config
   8. Third party environments
   9. Nested Makefiles
   10. Generate requirements.txt for each sub-project
   11. Generate setup.py for each sub-project
   12. Markdown badge
7. Out-of-the-box tools by language
8. High-level comparison with Pants, Bazel, Pre-commit, Makefiles
9. Limitations
10. Detailed comparison: Make vs Pre-commit vs Tox/Nox vs Bazel vs Pants

Platforms

AlphaBuild works on Linux distributions, MacOS, WSL and Windows with Git Bash. Make and Bash are AlphaBuild's only pre-requisites (Note: Bash with GNU utilities, not BSD utilities).

Linux, WSL

Since Linux is awesome and WSL follows its footsteps, AlphaBuild should just work there.

MacOS

AlphaBuild relies heavily on the GNU version of find and egrep, so, if you are running an OS which, by default, uses BSD rather than GNU (pointing fingers to MacOS here, you may need to brew install findutils and grep, see: https://xenodium.com/gnu-find-on-macos/)

Also, AlphaBuild may not work if you have ancient versions of Make / Bash, so try upgrading them if some commands don't seem to work. Macs typically come equipped with ancient versions of Bash.

Windows

On Windows, Make and Bash are not supported out-of-the-box, so it is recommended to use AlphaBuild on Windows within Git Bash. Note that Git Bash does not come with Make pre-installed. Once you downloaded and installed Git Bash run build-support/git-bash-integration/install_make.sh running Git Bash as administrator. Alternatively, you may get Make from conda.

Usage

Usually to format, lint, type-check, test, package, ... the code, one needs to run a bunch of commands in the terminal, setting the right flags and the right parameters. This Make-based build system helps with running these commands with a very simple interface:

make <goal> <optional-targets>

where

• goal = what tools we run
• targets = over which files we run these tools.

Goals - what tools we run

Goals mean what command line tools to run. This build system can run one more tools at once as follows:

• Single individual tool
  • e.g. make mypy, make flake8, make isort
• Multiple tools for a specific language
  • e.g. make fmt-py runs all Python formatters e.g. isort, black, docformatter, flynt, autoflake
  • e.g. make fmt-sh runs shfmt
  • e.g. make lint-py runs Python all linters and all formatters in "verification" mode, that is flake8 + pylint + check whether the code is already formatted with isort, black, docformatter, flynt, autoflake
  • make fmt-md, make lint-yml, test-sh, test-py ... work similarly
• Multiple tools for multiple languages
  • e.g. make fmt runs all formatters for all supported languages (Python, Bash, Markdown, YAML, ...)
  • e.g. make lint runs all linters for all supported languages
  • e.g. make test runs all test suites for all supported languages

It is possible to run multiple goals at once like make lint test. In addition, it is very easy to change the meaning of goals that run more than one command since they are very simply defined in Make based on other goals. For example, one can remove the shfmt from bash linting simply by doing the below:

# Before
lint-sh: shellcheck shfmt-check  # where shellcheck and shfmt-check run the respective commands
# After
lint-sh: shellcheck

Per-tool config files (e.g. mypy.ini, pyproject.toml) are found in build-support/<language>/tools-config/.

Targets - what files we run the tools on

We have seen that Make gives us the power to run multiple terminal commands effortlessly. Using a Makefile like described above is standard practice in many projects, typically running the different tools over all their files. However, as projects grow, the need to run these tools at different granularities (e.g. in a specific directory, over a given file, on the diff between two branches, since we last committed etc). This is where targets come into play.

With default targets

The default targets per-language are defined at the top of the Makefile in language specific variables e.g. ONPY=py-project1/ py-project2/ script.py and ONSH=scripts/.

• make lint runs:
  • all Python linters on all directories (in the $ONPY) that contain Python/stub files.
  • all notebook linters on all directories (in $ONNB) that contain .ipynb files.
  • all Bash linters (shellcheck) on all directories (in $ONSH) that contain Bash files.
  • a Haskell linter (hlint) on all directories (in $ONHS) that contain Haskell files.
  • a YAML linter (yamllint) on all directories (in $ONYML) that contain YAML files.
• make lint, make fmt -j1, make type-check work similarly
• remember that the $(ONPY), $(ONSH), ... variables are defined at the top of the Makefile and represent the default locations where AlphaBuild searches for files certain languages.

With specific targets

To specify manually the files/directories you want to run your tools on, AlphaBuild leverages the "on" syntax:

• file: make lint on=app_iqor/server.py runs all Python linters on the file, same as make lint-py on=app_iqor/server.py
• directory: make lint on=lib_py_utils runs a bunch of linters on the directory, in this case, same as make lint-py on=lib_py_utils/.
• files/directories: make lint on="lib_py_utils app_iqor/server.py" runs a bunch of linters on both targets.
• globs: make lint on=lib_*
• aliases: make fmt on=iqor where at the top of the Makefile iqor=app_iqor/iqor app_iqor/test_iqor, this is the same as make fmt on=app_iqor/iqor app_iqor/test_iqor because iqor is an alias for app_iqor/iqor app_iqor/test_iqor. Even though this example is simplistic, it is useful to alias combinations of multiple files/directories. It is recommended to set aliases as constants in the Makefile even though environment variables would also work.
• same for make fmt, make test, make type-check.

With git revision targets

• make fmt -j1 since=master runs all formatters on the diff between the current branch and master.
• make fmt -j1 since=HEAD~1 runs all formatters on all files that changed since "2 commits ago".
• make lint since=--cached runs all linters on all files that are "git added".
• all goals that support the "on" syntax also support the "since" syntax

Mixed "on" and "since"

One can use the "on" and "since" syntaxes at the same time. For example:

• make lint on=my_dir/ since=HEAD~2 will run all linters on all files in my_dir/ that changed since "3 commits ago".

Constraints

Different languages may have different goals, for example Python can be packaged hermetically with Shiv, while Bash obviously can't.

The following goals must support the "on" and "since" syntax and ensure that they are only run if there are any targets for the language they target:

• format
• lint
• type-check
• test

If you want to learn more about the API of a specific goal, check the source code.

AlphaBuild structure

• Makefile: AlphaBuild's entry point, this is where all components come together.
• 3rdparty/: Files required to build environments of 3rd party dependencies (e.g. requirements.txt files, package.json or lock files)
• build-support/: Makefile library inspired by Pants/Bazel to run linters, formatters, test frameworks, type checkers, packers etc. on a variety of languages (Python, Jupyter Notebooks, Bash, Haskell, YAML, Markdown)
  • The flags used per-tool (e.g. setting the paths to config files) can be found in build-support/alpha-build/config/<lang>.mk
  • The core part of AlphaBuild lives in build-support/alpha-build/core/, this comprises the build-system backbone resolver.mk and recipes to run lots of readily-available tools. This should be the same for all monorepos that use AlphaBuild.
  • By convention, repo-specific custom goals go in build-support/alpha-build/extensions/ following the examples in core.
  • build-support/<other-programming-lang-than-make>/ contain things like config files for each tool and other files required for your custom AlphaBuild goals.

IDE Integration

• PyCharm / IntelliJ
  • Windows: On Windows, it is advised to set Git Bash as the default "Terminal". In "settings" search for "terminal", go to "Tools → Terminal" and set "Shell Path" to something like C:\Program Files\Git\bin\bash.exe.
  • PYTHONPATH: If you are writing Python, please mark the directories for each project as "Sources Roots", such that PyCharm discovers your imports.
  • AlphaBuild hotkeys: It is easy to use AlphaBuild with hotkeys. For example, mapping Alt+F to make fmt -j1 on=<current-file> and Alt+L to make lint on=<current-file>. To set this up, follow the example in build-support/ide-integration/windows/ to set up an external tool. Unix systems would be similar. Next, just map the new external tools to your hotkeys.

Example monorepo running AlphaBuild

To see AlphaBuild at work in a real-world example check https://github.com/cristianmatache/workspace out. Workspace extends AlphaBuild with support for Prometheus, Alertmanager and Grafana.

Common admin actions

Installation

To add this build system to an existing repo, one needs to simply copy over build-support/, 3rdparty/, Makefile.

Create a conda environment, activate it and make env-default-replicate, as a one-off, to set up the python, markdown and bash environments (mostly pip/npm install-s).

CI/CD setup

Since all CI/CD pipelines essentially rely on running some scripts in a certain order, AlphaBuild can be called directly from any CI/CD pipeline regardless of CI/CD technology provider. AlphaBuild helps with ensuring that both the CI pipelines and developers run the exact same commands. Since, one can easily select targets within the repo, setting pipelines on a per sub-project basis, is effortless with AlphaBuild.

Upgrade

To upgrade an existing installation if new tools are added or changes are made to the target resolution infrastructure, one would simply need to replace the lib-support/alpha-build/core directory. To do that please run:

pip install alpha-build-core --target tmp/
tar -xvf tmp/alpha_build_core.tar.gz
rm -rf tmp/

Change goal definitions

Let's say, for example, you don't want to run pylint as part of your python linting. You would simply go to the Makefile and change the definition of the lint-py goal to not include pylint.

Add goals

The goals that are available out of the box are found in build-support/alpha-build/core/<language>/. You can extend/replace the core goals for new languages and/or tools by writing .mk code in build-support/alpha-build/extensions/<language>/ following the examples in build-support/alpha-build/core/. For example, https://github.com/cristianmatache/workspace extends AlphaBuild with Prometheus and Alertmanager goals.

Update PYTHONPATH

The PYTHONPATH is set at the top of the Makefile. For example, to add new directories to the PYTHONPATH (i.e. to mark them as sources roots) set PY_SOURCES_ROOTS at the top of the Makefile.

See/Change tools config

Let's say you want to change the way mypy is configured to exclude some directory from checking. Then head to build-support/alpha-build/config/python.mk check what is the path to the mypy config file, go there and update it. All other tools work similarly.

Third party environments

• Exact reproduction of the default environment: The recipes to fully replicate the default environment (mostly using pip, conda and npm) are found in build-support/alpha-build/core/<langugage>/setup.mk, where they use dependency files and lock files that can be found in 3rdparty/. In practice, run make env-default-replicate inside a conda environment. Also make sure you also have npm installed because markdownlint and bats bash testing framework come from npm (if you don't need them no need to worry about npm just exclude the markdown environment rule from the pre-requisites of env-default-replicate)
• Create/Upgrade/Edit default environment: If you want to edit the default environment, for example to add, remove, constrain packages edit the requirements.txt not the constraints.txt file (in 3rdparty/). The constraints.txt is only used for reproducibility. If you just want to upgrade your third party dependencies there is no need to temper with the requirements.txt files. Then run make env-default-upgrade and check the lock files back into git.
• Add a new environment: To add a new environment, first add the dependency files (e.g. requirements.txt) in 3rdparty/<new-env-name>, add a new goal in build-support/alpha-build/extensions. For environment management over time, we strongly encourage maintaining the approach split between creation/upgrade/edit and exact reproduction of environments.

Nested Makefiles

Supposing you want to use a different config file for black for a project in your monorepo. You would have 2 options: change the config file globally from build-support/alpha-build/config/python.mk (this would affect other projects) or create another Makefile in your specific project if you just want different settings for your little project (whether this is a good or a bad idea is more of a philosophical debate, I would argue that globally consistent config files are preferable, but I acknowledge that this may be needed sometimes). So, to have nested Makefiles that work with different config files:

# root/Makefile
black.%:  # Add this goal to be able to delegate to inner Makefile-s
    $(MAKE) -C $(subst .,/,$*) custom-black
fmt-py: black black.my-proj  # Add your custom "black" goals here

# root/my-proj/Makefile
include build-support/alpha-build/core/python/format.mk
custom-black:
    $(eval targets := $(onpy))
    $(MAKE) black targets="$(on)" BLACK_FLAGS="-S --config my-proj/pyproject.toml"

This way make fmt-py at the root would call the regular black rule but will also delegate to the inner Makefile custom-black rule to run the same tool differently. Don't forget to exclude my-proj/ from black in the outer Makefile, otherwise black would be run twice (once from the outside Makefile and again from the inner one).

Generate requirements.txt for each sub-project

Run make reqs-py.

Generate setup.py for each sub-project

Run build-support/python/packaging/generate_pip_install_files.py

Markdown badge

If you like AlphaBuild, wear the Markdown badge on your repo:

[![powered_by_alpha_build](https://img.shields.io/badge/Powered%20by%20-AlphaBuild-lightblue?style=flat&logo=CMake&logoColor=lightblue)
](https://github.com/cristianmatache/alpha-build)

Out-of-the-box tools by language

AlphaBuild has recipes to run the following tools. However, if you don't use some of them you don't need to have them installed in your enviornments. For example, let's say you don't use bandit then you don't need to have bandit installed in your environment provided that you are not using the bandit goal per-se or as part of a composite goal like lint-py or lint.

• Python:
  • Setup: pip / conda
  • Type-check: mypy
  • Test: pytest (along with doctest and other plugins)
  • Format + Lint: black, docformatter, isort, autoflake, flynt, pre-commit, pyupgrade
  • Lint only: flake8, pylint, bandit, pydocstyle (along with plugins like darglint, flake8-bugbear)
  • Package: pipreqs, shiv
• Jupyter:
  • Setup: pip
  • Format + Lint: jupyterblack, nbstripout
  • Lint only: flake8-nb
• Bash:
  • Setup: npm and conda
  • Test: bats (bash testing: bats-core, bats-assert, bats-support)
  • Format + Lint: shfmt
  • Lint only: shellcheck
• Haskell:
  • Lint: hlint
• YAML:
  • Setup: pip
  • Lint: yamllint, prettier
• Prometheus and Alertmanager YAML:
  • Lint: promtool check, amtool check-config
• Markdown:
  • Setup: npm
  • Format + Lint: markdownlint, prettier
• HTML + CSS:
  • Setup: npm
  • Format + Lint: prettier
• JavaScript:
  • Setup: npm
  • Format + Lint: prettier
• TypeScript:
  • Setup: npm
  • Format + Lint: prettier
• reStructuredText Text (.rst)
  • Lint: rstcheck
• Swift:
  • Format: SwiftLint, swift-format
  • Lint: SwiftLint, swift-format
• Kotlin:
  • Format: ktlint
  • Lint: ktlint It is very easy to extend this list with another tool, just following the existing examples.

High-level comparison with Pants, Bazel, Pre-commit and traditional Makefiles

Modern build tools like Pants or Bazel work similarly to AlphaBuild in terms of goals and targets, but they also add a caching layer on previous results of running the goals. While they come equipped with heavy machinery to support enormous scale projects, they also come with some restrictions and specialized maintenance and contribution requirements.

For example, Pants which, in my opinion, is the most suitable modern build tool for Python doesn't allow building environments with arbitrary package managers (e.g. conda, mamba), does not work on Windows, prohibits inconsistent environments (which is good but sometimes simply impossible in practice), does not yet support multiple environments. Bazel, requires maintaining the dependencies between Python files twice, once as "imports" in the Python files (the normal thing to do) and twice in some specific BUILD files that must be placed in each directory (by contrast Pants features autodiscovery). Maintaining the same dependencies in two places is quite draining. Of course, these tools come with benefits like (remote) caching, incrementality and out-of-the-box support for hermetic packaging (e.g. PEXes), remote execution etc. Moreover, playing with some new command line tools, or new programming languages / types of files (e.g. Jupyter Notebooks, Markdown, YAML) may be challenging with these frameworks. The Pants community is very welcoming and supportive towards incorporating new tools, so it would be good to give Pants a try first. However, if any of the mentioned shortcomings is a hard requirement, Make seems like a good and robust alternative in the meanwhile which withstood the test of time in so many settings. AlphaBuild's strengths are its flexibility, simplicity, transparency and tooling richness. One can quickly hack/add a new tool, see the commands that run under the hood and does not need to worry about BUILD files or the config language.

Since AlphaBuild is essentially a script manager (Python, Bash, Perl, anything) enhanced with advanced target/file/directory selection, AlphaBuild would allow an incremental adoption of large-scale build tools like Pants. For example, in the main Makefile, one could do:

# Makefile

lint-with-pants:
   $(eval on := ::)  # Default value if the user does not specify "on" in the terminal command like: make goal on=.
   ./pants lint $(on)

lint: lint-md lint-nb lint-yml lint-with-pants

such that running a command like the below would delegate most of the work to Pants while using AlphaBuild's core or custom capabilities not yet available in Pants (e.g. linting notebooks, markdown or YAML files).

make lint on=my-dir/

Pre-commit and typical usages of Make work exceptionally well on small projects but they don't really scale well to multi-projects monorepos. The build system proposed here, already incorporates pre-commit and is obviously compatible with any existing Makefiles. This approach simply takes the idea of advanced target selection and ports it over to classical techniques like pre-commit and Make.

Limitations

Since AlphaBuild is essentially a small-repo tool (Python Makefile) adapted to work on larger codebases (through target selection), there is a point from where it will no longer be able to scale up. Fortunately, that point is quite far away from medium-sized repos/teams.

In addition, AlphaBuild requires that the commands it builds are shorter than getconf ARG_MAX characters.

Detailed comparison: AlphaBuild vs Make vs Pre-commit vs Tox/Nox vs Bazel vs Pants

Most small/medium popular Python open source projects use Make, Pre-commit and/or Tox/Nox with a crushing majority for formatting, linting and/or testing and/or publishing. For the same purposes, fewer projects simply use a bunch of Bash scripts or monorepo-style build tools like Bazel (or Pants, Buck, Please). Make, pre-commit, nox/tox work pretty well together in the same repo and are often used so.

Note 1: Every time we talk about Pants we talk about Pants V2, which is a fundamentally different product from Pants V1. Even Twitter gave up on Pants V1 and is moving to Bazel (see https://twitter.com/jin_/status/1255133781876330497).

Note 2: IMHO Pants (V2) is the best (but not yet perfect) build tool for large monorepos out there and has great potential.

Pros and Cons

Click to expand and see the pros and cons of each tool!

• Tox/Nox:
  • Pros:
    • Good to run the same commands in multiple environments
    • Cross-platform
  • Cons:
    • Does not scale to large repos
    • Python-only
  • Notes: can be called from within Make, but can also call Make and pre-commit commands (see: https://tox.wiki/en/latest/config.html#conf-allowlist_externals)
• Pre-commit:
  • Pros:
    • Great to manage git hooks (as the name implies)
    • Can run over all files or over a specific set of files (slightly more advanced target selection)
    • Sort of incremental
  • Cons:
    • Does not work on Windows (docs say it doesn't but in fact it does partly)
    • Typically, very much geared towards format and lint.
  • Notes: can be called from within Make
• Make:
  • Pros:
    • Very flexible -> can essentially cover any tools/language (AlphaBuild is just Make with several built-in tools and target selection) and any way of fetching dependencies, multiple 3rd party environments etc.
    • Very transparent -> easy to support
    • Cross-platform
    • Supports nested Makefiles
  • Cons:
    • More scalable than pre-commit/tox/nox but not as scalable or hermetic as Bazel/Pants/Buck/Please (see https://github.com/thought-machine/please#why-please-and-not-make for details).
  • Notes: can run pre-commit, tox/nox but can also be run from tox/nox (not from pre-commit though)
• Bazel:
  • Pros:
    • Great for large scale projects (incremental, DAG, remote caching/execution)
    • Hermetic
  • Cons:
    • Need to maintain special BUILD files in every directory in which all dependencies are written again (once imported in the code, twice in the BUILD files)
    • Low tool coverage for Python/Jupyter ecosystem
    • Support for 3rd party Python environments was not great (not sure if it is still the case)
    • Has more sophisticated support needs (dedicated engineers and/or tuned CI infra)
  • Notes: can be called from within Make
• Pants V2:
  • Pros:
    • Great for large scale projects (incremental, DAG, remote caching/execution)
    • Dependencies are auto-discovered in BUILD files
    • Hermetic
  • Cons:
    • Environment support: no conda, no multiple environments, no arbitrary ways to create environments, no inconsistent envs
    • No support for Windows (only for WSL)
    • Tool/language support could be better (Pants's support for Python is better than Bazel's though)
    • Has more sophisticated support needs (e.g. dedicated engineers and/or tuned CI infra)
    • Does not readily support the equivalent of nested Makefiles
  • Notes:
    • Can be called from within Make
    • Still need to have BUILD files in every directory (much easier to work with than in Bazel)

Users by tool

Click to expand and see who uses each tool!

Nox, Tox, Pre-commit

These tools are simply extremely popular.

Bazel

• ray
• pytorch (Make for Python, Bash and CMake; in parallel with Bazel for other languages)
• selenium (with Rake i.e. ruby make) -> example BUILD file https://github.com/SeleniumHQ/selenium/blob/trunk/py/BUILD.bazel
• tensorflow
• keras
• protobuf
• jax (with pre-commit)

Pants

https://www.pantsbuild.org/page/who-uses-pants

Make

• Data
  • pandas (in parallel with pre-commit)
  • vaex
  • pydantic
  • pandera (in parallel with pre-commit, Make also runs nox)
• ML
  • pytorch (Make for Python, Bash and CMake; in parallel with Bazel for other languages)
  • xgboost
  • Determined AI (example of a more scalable Make-based infrastructure with nested Makefiles)
• Visualization
  • seaborn
• Distributed
  • celery (in parallel with tox and pre-commit)
• Stream processing
  • faust (Make also runs pre-commit)
• Web
  • requests
  • gunicorn (in parallel with tox)
  • sentry-python (Make also runs tox)
• AsyncIO (Aio-libs)
  • aiohttp (Make also runs pre-commit)
  • yarl
  • aiomysql
  • aioredis-py (Make also runs pre-commit)
  • aiopg
• DevOps
  • ansible (in parallel with tox)
  • pytest-testinfra (in parallel with tox)
• Documentation
  • sphinx: the home repo (in parallel with tox)
  • everything that uses sphinx for documentation
• Packaging
  • poetry (in parallel with tox and pre-commit)
• pyyaml (in parallel with tox)
• colorama (in parallel with tox)
• wrapt (Make also runs tox)
• pydata bottleneck
• facebook research
  • mephisto
  • demucs
  • diffq
  • dora
• apache
  • superset (in parallel with tox and pre-commit)
• IBM
  • lale (in parallel with pre-commit)
  • compliance-trestle (Make also runs pre-commit)
• reddit
  • baseplate.py
  • baseplate.py-upgrader
  • cqlmapper
• Other companies: AWS, Lyft, Microsoft, GoCardless, HewlettPackard