null-ls provides helpers to streamline the process of transforming command line output into LSP diagnostics, code actions, or formatting.
The plugin exports available helpers under null_ls.helpers
:
local helpers = require("null-ls.helpers")
Please see the built-in files for examples of how to use helpers to create generators.
generator_factory
is a general-purpose helper that returns a generator which
spawns a command with the given options, optionally transforms its output, then
calls an on_output
callback with the command's output. It accepts one
argument, opts
, which is a table with the following structure.
All options are required unless specified otherwise.
local helpers = require("null-ls.helpers")
helpers.generator_factory({
args, -- table (optional)
check_exit_code, -- function or table of numbers (optional)
command, -- string or function
env, -- table (optional)
cwd, -- function (optional)
dynamic_command, -- function (optional)
format, -- "raw", "line", "json", or "json_raw" (optional)
from_stderr, -- boolean (optional)
from_temp_file, -- boolean (optional)
ignore_stderr, -- boolean (optional)
multiple_files, -- boolean (optional)
on_output, -- function
runtime_condition, -- function (optional)
timeout, -- number (optional)
to_stdin, -- boolean (optional)
to_temp_file, -- boolean (optional)
use_cache, -- boolean (optional)
})
null-ls validates each option using vim.validate
when the generator runs for
the first time.
A table containing the arguments passed when spawning the command. Defaults to {}
.
null-ls will transform the following special arguments before spawning:
-
$FILENAME
: replaced with the current buffer's full path -
$DIRNAME
: replaced with the current buffer's parent directory -
$TEXT
: replaced with the current buffer's content -
$FILEEXT
: replaced with the current buffer's file extension (e.g.my-file.lua
produces"lua"
) -
$ROOT
: replaced with the LSP workspace root path
Can either be a table of valid exit codes (numbers) or a callback that receives two arguments:
code
: contains the exit code from the spawned command as a number
stderr
: error output from the job as a string
The callback should return a boolean value indicating whether the code indicates success.
If not specified, null-ls will assume that a non-zero exit code indicates failure.
A string containing the command that the generator will spawn or a function that
takes one argument, params
(an object containing information about the current
editor status) and returns a command string.
If command
is a function, it will run once when the generator first runs and
keep the same return value as long as the same Neovim instance is running,
making it suitable for resolving executables based on the current project.
Optional callback to set the working directory for the spawned process. Takes a
single argument, params
, which is a table containing information about the
current editor state (described in MAIN). If the callback returns
nil
, the working directory defaults to the project's root.
Optional callback to set command
dynamically. Takes one arguments, a params
object containing information about the current buffer's state. The generator's
original command (if set) is available as params.command
. The callback should
return a string containing the command to run or nil
, meaning that no command
should run.
dynamic_command
runs every time its parent generator runs and can affect
performance, so it's best to cache its output when possible.
Note that setting dynamic_command
will disable command
validation.
Specifies the format used to transform output before passing it to on_output
.
Supports the following options:
-
"raw"
: passes command output directly asparams.output
(string) and error output asparams.err
(string).This format will call
on_output(params, done)
, wheredone()
is a callback thaton_output
must call with its results (see Generators in MAIN for details). -
nil
: same asraw
, but does not receive error output. Instead, any output tostderr
will cause the generator to throw an error, unlessignore_stderr
is also enabled (see below). -
"line"
: splits generator output into lines and callson_output(line, params)
once for each line, whereline
is a string. -
"json"
: decodes generator output into JSON, setsparams.output
to the resulting JSON object, and callson_output(params)
. The wrapper will automatically calldone
onceon_output
returns. -
"json_raw"
: same asjson
, but will not throw on errors, either fromstderr
or fromjson.decode
. Instead, it'll pass errors toon_output
viaparams.err
.
To sum up:
-
If you want to handle each line of a source's output, use
format = "line"
. -
If you are handling JSON output, use
format = "json"
if you don't intend on handling errors andformat = "json_raw"
if you do. -
If you are processing a source's entire output, use
format = nil
if you don't intend on handling errors andformat = "raw"
if you do.
Captures a command's stderr
output and assigns it to params.output
. Note
that setting from_stderr = true
will discard any stdin
output.
Not compatible with ignore_stderr
.
Reads the contents of the temp file created by to_temp_file
after running
command
and assigns it to params.content
. Useful for formatters that don't
output to stdin
(see formatter_factory
).
This option depends on to_temp_file
.
For non-raw
output formats, any output to stderr
causes a command to fail
(unless from_stderr
is true
, as described above).
This option tells the runner to ignore the command's stderr
output. This is
like redirecting a command's output with 2>/dev/null
, but any error output is
still logged when debug
mode is on.
If set, signals that the generator will return results that apply to more than
one file. The null-ls diagnostics handler allows applying results to more than
one file if this option is true
and each diagnostic specifies a bufnr
or
filename
specifying the file to which the diagnostic applies.
A callback function that receives a params
object, which contains information
about the current buffer and editor state (see Generators in
MAIN for details).
Generators created by generator_factory
have access to an extra parameter,
params.output
, which contains the output from the spawned command. The
structure of params.output
depends on format
, described below.
Optional callback called when generating a list of sources to run for a given
method. Takes a single argument, params
, which is a table containing
information about the current editor state (described in MAIN). If
the callback's return value is falsy, the source does not run.
Be aware that the callback runs every time a source can run and thus should avoid doing anything overly expensive.
The amount of time, in milliseconds, until null-ls aborts the command and
returns an empty response. If not set, will default to the user's
default_timeout
setting.
Sends the current buffer's content to the spawned command via stdin
.
Writes the current buffer's content to a temporary file and replaces the special
argument $FILENAME
with the path to the temporary file. Useful for formatters
and linters that don't accept input via stdin
.
Setting to_temp_file = true
will also assign the path to the temp file to
params.temp_path
.
Caches command output on run. When available, the generator will use cached output instead of spawning the command, which can increase performance for slow commands.
The plugin resets cached output when Neovim's LSP client notifies it of a buffer
change, meaning that cache validity depends on a user's debounce
setting.
Sources that rely on up-to-date buffer content should avoid using this option.
Note that this option effectively does nothing for diagnostics, since the handler will always invalidate the buffer's cache before running generators.
formatter_factory
is a wrapper around generator_factory
meant to streamline
the process of capturing a formatter's output and replacing a buffer's entire
content with that output. It supports the same options as generator_factory
with the following changes:
-
ignore_stderr
: set totrue
by default. -
on_output
: will always return an edit that will replace the current buffer's content with formatter output. As a result, other options that depend onon_output
, such asformat
, will not have an effect.
make_builtin
creates built-in sources. It optimizes the source to reduce
start-up time and allow the built-in library to continue expanding without
affecting users.
make_builtin
is specifically intended for built-ins included in this plugin.
Generally, integrations should opt to create sources with one of the factory
methods described above, since they are opt-in by nature.
The method accepts a single argument, opts
, which contains the following
options. All options are required unless specified otherwise.
local helpers = require("null-ls.helpers")
helpers.make_builtin({
factory, -- function (optional)
filetypes, -- table
generator, -- function (optional, but required if factory is not set)
generator_opts, -- table
method, -- internal null-ls method (string)
meta, -- table
})
A function called when the user registers the source. Intended for use with the
helper factory
functions described above, but any function that returns a
valid generator will work.
A list of filetypes for the source, as described in MAIN.
A simple generator function. Either factory
or generator
must be a valid
function, and setting factory
will override generator
.
A table of arguments passed into factory
when the user registers the source,
which should conform to the opts
object described above in
generator_factory
.
Defines the source's null-ls method, as described in MAIN.
Adds metadata to enrich the source's documentation. null-ls will use the following fields:
meta.url
: the path to the source's official website / repositorymeta.description
: a description of the source and its capabilitiesmeta.notes
: an array of notes converted into a Markdown list
Converts the visually-selected range into character offsets / rows and adds the converted range to the spawn arguments. Used for sources that provide both formatting and range formatting.
null_ls.helpers.range_formatting_args_factory(base_args, start_arg, end_rag, opts)
-
base_args
(string[]): the base arguments required to get formatted output. Formatting requests will usebase_args
as-is and range formatting requests will append range arguments. -
start_arg
(string): the name of the argument that indicates the start of the range. -
end_arg
(string?): the name of the argument that indicates the end of the range. If not specified, the helper will insert the start and end of the range afterstart_arg
. -
opts
(table?): a table containing the following options:opts.use_rows
(boolean?): specifies whether to use rows over character offsets.opts.use_length
(boolean?): used to specify the length of the range inend_arg
instead of end the position.opts.row_offset
(number?): offset applied to row numbers.opts.col_offset
(number?): offset applied to column numbers.opts.delimiter
(string?): used to join range start and end into a single argument.
Helpers used to convert CLI output into diagnostics. See the source for details and the built-in diagnostics sources for examples.
Helpers used to cache output from callbacks and help improve performance.
Creates a function that caches the result of callback
, indexed by bufnr
. On
the first run of the created function, null-ls will call callback
with a
params
object containing information about the buffer's state and cache its
output using bufnr
as a key. On the next run, it will directly return the
cached value without calling callback
again.
This is useful when the return value of callback
is not expected to change
over the lifetime of the buffer, which works well for cwd
and
runtime_condition
callbacks. Users can use it as a simple shortcut to improve
performance, and built-in authors can use it to add logic that would otherwise
be too performance-intensive to include out-of-the-box.
Note that if callback
returns nil
, the helper will override the return value
and instead cache false
(so that it can determine that it already ran
callback
once and should not run it again).