Branch: master
Find file Copy path
316 lines (258 sloc) 13.8 KB
*ale-development.txt* For Vim version 8.0.
ALE Development Documentation
CONTENTS *ale-development-contents*
1. Introduction.........................|ale-development-introduction|
2. Design Goals.........................|ale-design-goals|
3. Coding Standards.....................|ale-coding-standards|
4. Testing ALE..........................|ale-development-tests|
4.1. Writing Linter Tests.............|ale-development-linter-tests|
1. Introduction *ale-development-introduction*
This document contains helpful information for ALE developers, including
design goals, information on how to run the tests, coding standards, and so
on. You should read this document if you want to get involved with ALE
2. Design Goals *ale-design-goals*
This section lists design goals for ALE, in no particular order. They are as
ALE code should be almost 100% VimL. This makes the plugin as portable as
ALE should run without needing any other plugins to be installed, to make
installation simple. ALE can integrate with other plugins for more advanced
functionality, non-essential functionality, or improving on basic first party
ALE should check files with as many tools as possible by default, except where
they cause security issues or make excessive use of resources on modern
ALE should be free of breaking changes to the public API, which is comprised of
documented functions and options, until a major version is planned. Breaking
changes should be preceded by a deprecation phase complete with warnings.
Changes required for security may be an exception.
ALE supports Vim 8 and above, and NeoVim 0.2.0 or newer. These are the
earliest versions of Vim and NeoVim which support |job|, |timer|, |closure|,
and |lambda| features. All ALE code should be written so it is compatible with
these versions of Vim, or with version checks so particular features can
degrade or fail gracefully.
Just about everything should be documented and covered with tests.
By and large, people shouldn't pay for the functionality they don't use. Care
should be taken when adding new features, so supporting new features doesn't
degrade the general performance of anything ALE does.
LSP support will become more important as time goes on. ALE should provide
better support for LSP features as time goes on.
When merging pull requests, you should respond with `Cheers! :beers:`, purely
for comedy value.
3. Coding Standards *ale-coding-standards*
The following general coding standards should be adhered to for Vim code.
* Check your Vim code with `Vint` and do everything it says. ALE will check
your Vim code with Vint automatically. See:
Read ALE's `Dockerfile` to see which version of `Vint` it uses.
* Try to write descriptive and concise names for variables and functions.
Names shouldn't be too short or too long. Think about others reading your
code later on.
* Use `snake_case` names for variables and arguments, and `PascalCase` names
for functions. Prefix every variable name with its scope. (`l:`, `g:`, etc.)
* Try to keep lines no longer than 80 characters, but this isn't an absolute
* Use 4 spaces for every level of indentation in Vim code.
* Add a blank line before every `function`, `if`, `for`, `while`, or `return`,
which doesn't start a new level of indentation. This makes the logic in
your code easier to follow.
* End every file with a trailing newline character, but not with extra blank
lines. Remove trailing whitespace from the ends of lines.
* Write the full names of commands instead of abbreviations. For example, write
`function` instead of `func`, and `endif` instead of `end`.
* Write functions with `!`, so files can be reloaded. Use the |abort| keyword
for all functions, so functions exit on the first error.
* Make sure to credit yourself in files you have authored with `Author:`
and `Description:` comments.
In addition to the above general guidelines for the style of your code, you
should also follow some additional rules designed to prevent mistakes. Some of
these are reported with ALE's `custom-linting-rules` script. See
* Don't leave stray `:echo` lines in code. Use `execute 'echo' ...` if you must
echo something.
* For strings use |is#| instead of |==#|, `is?` instead of `==?`, `isnot#`
instead of `!=#`, and `isnot?` instead of `!=?`. This is because `'x' ==# 0`
returns 1, while `'x' is# 0` returns 0, so you will experience fewer issues
when numbers are compared with strings. `is` and `isnot` also do not throw
errors when other objects like List or Dictionaries are compared with
* Don't use the `getcwd()` function in the ALE codebase. Most of ALE's code
runs from asynchronous callback functions, and these functions can execute
from essentially random buffers. Therefore, the `getcwd()` output is
useless. Use `expand('#' . a:buffer . ':p:h')` instead. Don't use
`expand('%...')` for the same reason.
* Don't use the `simplify()` function. It doesn't simplify paths enough. Use
`ale#path#Simplify()` instead.
* Don't use the `shellescape()` function. It doesn't escape arguments properly
on Windows. Use `ale#Escape()` instead, which will avoid escaping where it
isn't needed, and generally escape arguments better on Windows.
* Don't use the `tempname()` function. It doesn't work when `$TMPDIR` isn't
set. Use `ale#util#Tempname()` instead, which temporarily sets `$TMPDIR`
appropriately where needed.
* Use `snake_case` names for linter names, so they can be used as part of
variable names. You can define `aliases` for linters, for other names people
might try to configure linters with.
* Use |v:t_TYPE| variables instead of `type()`, which are more readable.
Apply the following guidelines when writing Vader test files.
* Use 2 spaces for Vader test files, instead of the 4 spaces for Vim files.
* If you write `Before` and `After` blocks, you should typically write them at
the top of the file, so they run for all tests. There may be some tests
where it make sense to modify the `Before` and `After` code part of the way
through the file.
* If you modify any settings or global variables, reset them in `After`
blocks. The Vader `Save` and `Restore` commands can be useful for this
* If you load or define linters in tests, write `call ale#linter#Reset()` in
an `After` block.
* Just write `Execute` blocks for Vader tests, and don't bother writing `Then`
blocks. `Then` blocks execute after `After` blocks in older versions, and
that can be confusing.
Apply the following rules when writing Bash scripts.
* Run `shellcheck`, and do everything it says.
* Try to write scripts so they will run on Linux, BSD, or Mac OSX.
4. Testing ALE *ale-development-tests* *ale-dev-tests* *ale-tests*
ALE is tested with a suite of tests executed in Travis CI and AppVeyor. ALE
runs tests with the following versions of Vim in the following environments.
1. Vim 8.0.0027 on Linux via Travis CI.
2. Vim 8.1.0204 on Linux via Travis CI.
3. NeoVim 0.2.0 on Linux via Travis CI.
4. NeoVim 0.3.0 on Linux via Travis CI.
5. Vim 8 (stable builds) on Windows via AppVeyor.
If you are developing ALE code on Linux, Mac OSX, or BSD, you can run ALEs
tests by installing Docker and running the `run-tests` script. Follow the
instructions on the Docker site for installing Docker.
NOTE: Don't forget to add your user to the `docker` group on Linux, or Docker
just won't work. See:
If you run simply `./run-tests` from the ALE repository root directory, the
latest Docker image for tests will be downloaded if needed, and the script
will run all of the tests in Vader, Vint checks, and several Bash scripts for
finding extra issues. Run `./run-tests --help` to see all of the options the
script supports. Note that the script supports selecting particular test files.
Generally write tests for any changes you make. The following types of tests
are recommended for the following types of code.
* New/edited error handler callbacks -> Write tests in `test/handler`
* New/edited command callbacks -> Write tests in `test/command_callback`
* New/edited fixer functions -> Write tests in `test/fixers`
Look at existing tests in the codebase for examples of how to write tests.
Refer to the Vader documentation for general information on how to write Vader
See |ale-development-linter-tests| for more information on how to write linter
When you add new linters or fixers, make sure to add them into the table in
the README, and also into the |ale-support| list in the main help file. If you
forget to keep them both in sync, you should see an error like the following
in Travis CI. >
diff and doc/ale.txt tables
Differences follow:
--- /tmp/readme.qLjNhJdB 2018-07-01 16:29:55.590331972 +0100
+++ /tmp/doc.dAi8zfVE 2018-07-01 16:29:55.582331877 +0100
@@ -1 +1 @@
- ASM: gcc, foobar
+ ASM: gcc
Make sure to list documentation entries for linters and fixers in individual
help files in the table of contents, and to align help tags to the right
margin. For example, if you add a heading for an `aardvark` tool to
`ale-python.txt` with a badly aligned doc tag, you will see errors like so. >
Look for badly aligned doc tags
Badly aligned tags follow:
doc/ale-python.txt:aardvark ...
Look for table of contents issues
Check for bad ToC sorting:
Check for mismatched ToC and headings:
--- /tmp/table-of-contents.mwCFOgSI 2018-07-01 16:33:25.068811878 +0100
+++ /tmp/headings.L4WU0hsO 2018-07-01 16:33:25.076811973 +0100
@@ -168,6 +168,7 @@
pyrex (cython), ale-pyrex-options
cython, ale-pyrex-cython
python, ale-python-options
+ aardvark, ale-python-aardvark
autopep8, ale-python-autopep8
black, ale-python-black
flake8, ale-python-flake8
Make sure to make the table of contents match the headings, and to keep the
doc tags on the right margin.
4.1 Writing Linter Tests *ale-development-linter-tests*
Tests for ALE linters take two forms.
1. Tests for handling the output of commands.
2. Tests for checking which commands are run, or connections are made.
Tests of the first form should go in the `test/handler` directory, and should
be written like so. >
" Load the file which defines the linter.
runtime ale_linters/filetype/linter_name_here.vim
" Unload all linters again.
call ale#linter#Reset()
Execute(The output should be correct):
" Test that the right loclist items are parsed from the handler.
\ [
\ {
\ 'lnum': 1,
\ 'type': 'E',
\ 'text': 'Something went wrong',
\ },
\ ],
\ ale_linters#filetype#linter_name#Handle(bufnr(''), [
\ '1:Something went wrong',
\ ]
Tests for what ALE runs should go in the `test/command_callback` directory,
and should be written like so. >
" Load the linter and set up a series of commands, reset linter variables,
" clear caches, etc.
" Vader's 'Save' command will be called here for linter variables.
call ale#assert#SetUpLinterTest('filetype', 'linter_name')
" Reset linters, variables, etc.
" Vader's 'Restore' command will be called here.
call ale#assert#TearDownLinterTest()
Execute(The default command should be correct):
" AssertLinter checks the executable and command.
" Pass expected_executable, expected_command
AssertLinter 'some-command', ale#Escape('some-command') . ' --foo'
Execute(Check chained commands):
" WithChainResults can be called with 1 or more list for passing output
" to chained commands. The output for each callback defaults to an empty
" list.
WithChainResults ['v2.1.2']
" Given a List of commands, check all of them.
" Given a String, only the last command in the chain will be checked.
AssertLinter 'some-command', [
\ ale#Escape('some-command') . ' --version',
\ ale#Escape('some-command') . ' --foo',
The full list of commands that will be temporarily defined for linter tests
given the above setup are as follows.
`WithChainResults [...]` - Define output for command chain functions.
`AssertLinter executable, command` - Check the executable and command.
`AssertLinterNotExecuted` - Check that linters will not be executed.
`AssertLSPLanguage language` - Check the language given to an LSP server.
`AssertLSPOptions options_dict` - Check the options given to an LSP server.
`AssertLSPConfig config_dict` - Check the config given to an LSP server.
`AssertLSPProject project_root` - Check the root given to an LSP server.
`AssertLSPAddress address` - Check the address to an LSP server.