* Add fourmolu fixer
Fourmolu is aversion of Ormolu that supports configuration. This fixer
was modeled after the Ormolu one, but using the "stack executable"
approach of the Brittany and Stylish Haskell fixers.
* Sort supported-tools.md
Support replacing ALE's display of problems with sending problems to the Neovim diagnostics API.
:help g:ale_use_neovim_diagnostics_api
Co-authored-by: David Balatero <dbalatero@users.noreply.github.com>
Co-authored-by: Georgi Angelchev <angelchev@live.co.uk>
Co-authored-by: w0rp <devw0rp@gmail.com>
Default virtual-text to the Comment highlight group and prefix
virtual-text messages with comment text for each language by default.
Messages can now be formatted with `%type%` to print the error type.
The Vim 9.0 version has been updated in the Docker image to add test
coverage for virtual-text.
Consider a file like
```
#lang racket
(require racket/gui)
```
Type `Go(eventspace-`.
Pressing <C-x><C-o> to trigger omnicomplete should suggest
```
eventspace-handler-thread
eventspace-shutdown?
eventspace-event-evt
```
It does not (instead producing "top-level" completions, as if
`(eventspace-` wasn't even there).
Debugging, place the cursor on a space _after_. Now
`ale#completion#OmniFunc(1, '')` correctly returns `1`, but when given
`(0, 'eventspace-')` it returns either the empty list or generic
completion results as described above. I'm not entirely sure of the
mechanism, but it seems that `b:ale_completion_info.prefix` is the key,
and that this is set by `ale#completion#GetPrefix`. Calling
`ale#completion#GetPrefix('racket', line('.'), col('.'))` returned `''`!
Now, it returns `eventspace-` and the completions work correctly again.
Ref #4293, #4186, #3870
Gcc does not support `x c*-header` when using `-` as input filename,
which is what ALE does.
Rework the feature to only use `-x c*-header` flag when using Clang and
not GCC.
The feature is now also controlled with the variable
`g:ale_c_cc_use_header_lang_flag` and
`g:ale_cpp_cc_use_header_lang_flag`.
Add configuration files for pyright (JSON and TOML) to list of files
which identify a project root directory. Update documentation
accordingly.
Co-authored-by: Andreas Doll <andreas.doll@posteo.de>
When linting an header file in C or C++, `-x c-header` or
`-x c++-header` should be used instead of `-x c` or `-x c++`.
Using `-x c` or `-x c++` for headers files can lead to unused variables
and functions marked as static inlined as seen in #4096.
Using `-x c-header` or `-x c++-header` solve these issues.
The list of file extensions that are considered as header files can be
configured with the variables `g:ale_c_cc_header_exts` and
`g:ale_cpp_cc_header_exts`.
Deno LSP automatically detects config files named deno.json or
deno.jsonc since version 1.18.
For Deno 1.18+ this means that ALE no longer needs to resolve the
project root. However, removing the project root logic from ALE means
breaking changes for people that are still using an older version.
Adding deno.json to the list of looking files to look for will keep the
behavior consistent and compatible with the Deno config file naming
convention.
See also:
https://deno.com/blog/v1.18#auto-discovery-of-the-config-file
* Remove virtual text via types-filter
This is more robust and has the additional sideeffect that it will make
it easier to implement showing virtual text for all warnings
simultaneously.
We definitely do not want to do a call to prop_remove() for every
virtual text as that will cause noticeable lag when many warnings are
present, thus we can use this to remove all virtual text lines with one
call in the future.
Fixes#4294
refs: https://github.com/vim/vim/pull/10945
* Allow virtual text to appear for all warnings of the buffer
This can be enabled with:
let g:ale_virtualtext_cursor = 2
It is implemented both for neovim and vim 9.0.0297.
Note that sometimes it may appear like some warnings are displayed
multiple times. This is not a bug in the virtual text implementation,
but a sideeffect of multiple linters returning similar results.
For example for Rust, the 'cargo' and 'rls' linters appear to be
activated at the same time, but they sometimes return identical errors.
This causes the virtual text to show the same warning twice.
In the future we can mitigate this problem by removing duplicate errors
from our internal location list.
However users can also achieve cleaner warnings simply by activating
only one linter for each language (or multiple unambiguous linters).
For example for Rust, the problem could be solved with:
let g:ale_linters = {'rust': ['analyzer']}
Fixes#2962Fixes#3666
Regression was introduced in d93bc2baf7
The problem was that we did not handle the edge case where there is no
last popup to close, which caused old vim versions to enter code by
accident that was only supposed to be run by vim 9.
We fix this by guarding the if condition for vim 9.
Fixes#4290
* Add extra config options for virtualtext
* Undo virtualtext changes and move to floating preview
* revert changes to pass hightlight group to floating preview
* rename var
* Document changes
* Add updates based on feedback
* Check for string type and attempt to call the function
* Fix lint errors
Co-authored-by: Shaun Duncan <shaun@speedscale.com>
Our current virtual text implementation for vim emulates it by abusing
the textprop and popupwin feature from vim 8.2 (for more details see
commit 708e810414).
This implementation sometimes is janky, for example the popups may leak
into other vim windows next to the current window.
Luckily, vim just got native virtual-text support as a proper subtype to
the prop_add() function. By using the 'text' option, the text property
automatically becomes virtual text that is appended to the current line
if col is zero.
Note that the prop_add() method now returns negative IDs for virtual
text properties.
This feature was added in vim 9.0.0067, but it got a lot of bugfixes
which is why we only use this new API if vim has at least version
9.0.0214.
However, there are still some minor bugs with vim's native virtual text,
so we might have to bump the version check again in the future.
Also see #3906.
Now with proper virtual text support for both vim and neovim available,
we can tackle #2962 in the future by simply tracking multiple virt-texts
instead of just the last one.
In the future we might also want to disable our virtual text emulation
support for vim, as it is a total hack, but for now we should keep it
for backwards compatibility.
When 'close_cb' is set for job_start(), but out_cb or err_cb isn't, vim
buffers data instead of dropping it (in case someone wanted to read and
process it in close_cb), and additionally polls for new data every 10
milliseconds, causing excessive wakeups and CPU usage. Since we don't
read the data anywhere outside of out_cb/err_cb, any LSP that prints an
error to stderr triggers this and vim keeps spinning until :ALEStopAllLSPs.
Fix this by always setting both callbacks, thus dropping any data we're
not interested in.
See https://github.com/vim/vim/issues/10758 for an upstream report of
the excessive polling. It's possible this is intentional, I dunno.
Fixes: b42153eb17 ("Fix #4098 - Clear LSP data when servers crash")
