When using a compilation database (compile_commands.json) in very large
projects, significant delays would occur when changing files --
particularly those that happened to be far down the db. Rather than
iterating over the whole list every time, we now build up a lookup table
based on the tail of the filename (and tail of the directory for
widening searches) and iterate over the much smaller list of compile
commands for files with the given name.
Test metrics (from compile_database_perf/test.sh) show a 90% performance
improvement -- from 25 seconds to 2.5 seconds per run.
* Add support for https://github.com/saibing/bingo
* Add docs for ale-go-bingo
* Use go.mod when found
* Add test for bingo FindProjectRoot
* Simplify ale_linters#go#bingo#GetCommand
With earlier elm versions, a separate package file is maintained for
tests, which when properly configured enabled the compiler to find what
it needed to compile the tests. Under elm 0.19, test dependencies are
managed in the top-level package file, so `elm make` will fail on the
tests. `elm-test make` is required in this case.
See https://github.com/elm-explorations/test/issues/64
- added a cd into the direcotry containing the file in the buffer
in order to properly check for a config file
- added command_callback tests for graphql
See: https://github.com/testdouble/standard
StandardRB is to RuboCop what StandardJS is to ESLint. This commit
naively copies the RuboCop linter and fixer to point at the standardrb
executable. Any other adjustments are very minor (the only I can think
of is that standardrb takes a `--fix` option instead of
`--auto-correct`).
This raises a confusing point to me as both developer and a user: since
ale enables all linters by default, won't this run both RuboCop and
StandardRB (the results of which will almost always be in conflict with
one another)? How does ale already solve for this for the similar case
of StandardJS and ESLint?
It's common to add SwiftLint as a CocoaPod dependency, instead of as a global
binary. In this case we should use that version of SwiftLint before looking
for any others. Note that I'm also adding support for SwiftLint in ReactNative
projects here as well, where the Pods directory would be nested inside an ios
directory.
The linter can correctly parse pydocstyle output with any of the following
command-line options enabled: --explain, --source, --debug, and/or
--verbose
The command used to invoke the LSP process was being escaped wrong.
Also added a new option to set a different java executable and fixed the
documentation.
There is currently a check that tries to prevent c-flags that contain
'-' in them from being unintentionally split and included in the list of
commands. For example, we wouldn't want "-fno-exceptions " to appear as
"-fno" and "-exceptions ". The way this check was done was by making sure
the last character of the split string was a space.
This meant that the very last option to appear in the compile command
was ignored (as it doesn't end with a space). This fix explicitly skips
the ends-with-space check on the last option in the command-line.
This isn't the best fix. Really we should be using the same
argument-processing rules as a shell would rather than just splitting on
'-'. That's a much larger and more complicated change though.
The output format used by older checkstyle versions differs from the one
of new versions. This commit adds a second parsing iteration on the
output lines with a suitable pattern to support both versions in
parallel. Due to the differences in the order of matching groups this is
hard to achieve in a single pass through the output lines.
An appropriate test case is added.
Removed ale_virualtext_prefix from debugging since it's not requried for
the functionality to work.
Sorted debugging info to make the list easier to navigate/diff.
Problem: ocamlformat is configured to format files in-place and thus go
via creating a temporary file for that. Because temporary file resides
in a different directory ocamlformat can't find `.ocamlformat`
configuration files in an original location of source files.
Solution: ocamlformat since version 0.8 can read sources on stdin and
spur result on stdout. We reconfigure ocamlformat to use a simpler
interface.
Previously, elixir-ls would treat each sub-project within an umbrella as
standalone, which isn't desirable from a language server perspective.
Added ale#handlers#elixir#FindMixUmbrellaRoot, which locates the current
project's root and then continues searching upwards for a potential
umbrella project root. This literally looks just two levels up to keep
things simple while keeping in line with Elixir project conventions.
Use this new function to determine elixir-ls's LSP project root.
* Allow configuration of hamllint executable
The hamllint executable was hard-coded, preventing it from being
overridden. Fix the executable to be dynamic to allow custom executable
paths.
This adds generic configuration dictionary support to the elixir-ls
linter. This is useful for disabling its built-in Dialyzer support, for
example, which can improve startup time.
The configuration dictionary is a little verbose. I considered reducing
the user configuration to only the nested settings dictionary (and
having the linter implementation wrap it in the top-level `elixirLS`
dictionary), but leaving it fully configurable simplifies the code and
removes any assumptions about current or future ElixirLS behavior.
Each LSP connection now stores its configuration dictionary. It is
initially empty (`{}`) and is updated each time the LSP connection is
started. When a change is detected, the workspace/didChangeConfiguration
message is sent to the LSP servers with the updated configuration.
This is the callback-based variant of the existing `lsp_config` linter
option. It serves the same purpose but can be used when more complicated
processing is needed.
`lsp_config` and `lsp_config_callback` are mutually exclusive options;
if both an given, a linter preprocessing error will be raised.
The runtime logic has been wrapped in `ale#lsp_linter#GetConfig` for
convenience, similar to `ale#lsp_linter#GetOptions`.
This also adds documentation and an `AssertLSPConfig` test function for
completeness.
* add prolog/swipl linter
* use load_files/2 instead of read_term/2
Because it also checks some semantic warnings / errors
not only syntactic warnings / errors.
e.g.:
* singleton warning
* discontiguous warning
* ...
cf. http://www.swi-prolog.org/pldoc/doc_for?object=style_check/1
* support error messages with no line number
:- module(module_name, [pred/0]).
causes
ERROR: Exported procedure module_name:pred/0 is not defined
* add test for prolog/swipl handler
* cosmetic fixes
* detect timeout using SIGALRM
* rename g:prolog_swipl_goals to g:prolog_swipl_load
* write doc for prolog/swipl linter
* update toc and README
* fix ignore patterns
* Only run stack if a stack.yaml config is found
It is necessary to check for a stack.yaml file to distinguish between
cabal-only projects or stack projects (which are also cabal projects
since stack is built on top of cabal).
* Test that stack is called if stack.yaml exists
ElixirLS (https://github.com/JakeBecker/elixir-ls) is an LSP server for
Elixir. It's distributed as a release package that can be downloaded
from https://github.com/JakeBecker/elixir-ls/releases or built locally.
The easiest way to start it is via Unix- and Win32-specific helper
scripts, so that's the basis of this command integration. Alternatively,
we could implement the contents of those platform-specific scripts in
the linter's command callback in a language-neutral way, but there isn't
any benefit to doing that aside from eliminating the platform check, and
that could prove to be too tight of a coupling going forward.
* FIX: use mix from the project root directory
* Move find root project function to autoloaded handlers
* add tests for #ale#handlers#elixr#FindMixProjectRoot
PMD is currently not working properly for Java classes that use [unnamed
packages](https://docs.oracle.com/javase/specs/jls/se11/html/jls-7.html#jls-7.4.2).
Consider the following Java class that does not contain a `package`
declaration:
```java
public class App {
String getGreeting() {
return "Hello world.";
}
static void main(String... args) {
System.out.println(new App().getGreeting());
}
}
```
Running PMD in the command line agaist the Java class above produces an
output with empty string `""` in the `"Package"` column:
```sh
$ pmd -R category/java/bestpractices.xml -f csv -d './src/main/java/App.java'
Oct 02, 2018 9:10:39 PM net.sourceforge.pmd.PMD processFiles
WARNING: This analysis could be faster, please consider using Incremental Analysis: https://pmd.github.io/pmd-6.7.0/pmd_userdocs_incremental_analysis.html
"Problem","Package","File","Priority","Line","Description","Rule set","Rule"
"1","","/Users/diego/Projects/github.com/dlresende/kata-fizz-buzz/src/main/java/App.java","2","7","System.out.println is used","Best Practices","SystemPrintln"
```
But the pmd.vim handler's current pattern refuses everything coming
from a Java class that does not have a package name (2nd column):
```vim
let l:pattern = '"\(\d\+\)",".\+","\(.\+\)","\(\d\+\)","\(\d\+\)","\(.\+\)","\(.\+\)","\(.\+\)"$'
```
The solution I am proposing is to also accept empty strings as package names.
These test vars were covering up a bug in the hlint linter
implementation. Without these vars we can see the behavior that is
exhibited in `vim` proper.
* Add better support for Haskell stack compiler tools
This commit adds support for `stack` as the executable of a tool. This
follows a pattern that has been implemented for `bundler`'s tool chain.
* Move hlint command to linter file
* Add vader test for stack exec handling
* Update ghc-mod to support stack execution
`ghc-mod` was previously broken into 2 linters.
1. ghc_mod
2. stack_ghc_mod
This additional linter is not necessary with proper support for
executable variables and `stack exec` handling.
* Support stack exec in hfmt
* Support stack in hdevtools
* Don't add newlines when not a control statement for Python
* Add test for accidental newline fix
* Add docstring detection to avoid adding unnecessarily newlines
* Add tests for docstring detection
In a lint context, it's useful to assume that included files sit next to
the current file by default. Users can still further customize this
configuration variable to add more include paths.
When set to true, and the buffer is currently inside a pipenv,
GetExecutable will return "pipenv", which will trigger the existing
functionality to append the correct pipenv arguments to run each linter.
Defaults to false.
I was going to implement ale#python#PipenvPresent by invoking
`pipenv --venv` or `pipenv --where`, but it seemed to be abominably
slow, even to the point where the test suite wasn't even finishing
("Tried to run tests 3 times"). The diff is:
diff --git a/autoload/ale/python.vim b/autoload/ale/python.vim
index 7baae079..8c100d41 100644
--- a/autoload/ale/python.vim
+++ b/autoload/ale/python.vim
@@ -106,5 +106,9 @@ endfunction
" Detects whether a pipenv environment is present.
function! ale#python#PipenvPresent(buffer) abort
- return findfile('Pipfile.lock', expand('#' . a:buffer . ':p:h') . ';') isnot# ''
+ let l:cd_string = ale#path#BufferCdString(a:buffer)
+ let l:output = systemlist(l:cd_string . 'pipenv --where')[0]
+ " `pipenv --where` returns the path to the dir containing the Pipfile
+ " if in a pipenv, or some error text otherwise.
+ return strpart(l:output, 0, 18) !=# "No Pipfile present"
endfunction
Using vim's `findfile` is much faster, behaves correctly in the majority
of situations, and also works reliably when the `pipenv` command doesn't
exist.
Solargraph allows to set configuration options by creating a
.solargraph.yml file at the root of the project using it. Therfore this
file is a good canditate for finding ruby projects root paths.
Initial discussion:
https://github.com/w0rp/ale/issues/1874#issuecomment-418316168
* The project style linter now runs while you type.
* Now the scripts for checking the project require blank lines.
* Many style issues have been found and fixed.
* Add stylish-haskell as a fixer
`stylish-haskell` is a common formatting tool for the haskell toolchain.
It is not as advanced as `brittany` or `hindent`, but it is commonly
used for formatting of imports and data declarations. This adds it as a
fixer in ALE.
I see no reason to do this? It is just setting the environment to what
it already is?
It was originally added in #297, but that entire PR is not a great idea
in the first place; that PR (together with #270) tried to make the Go do
non-standard and non-supported stuff like compiling packages outside of
GOPATH.
That's not something that works well (I tried), so was eventually
removed in #465, but these "go env" calls remained, for no reason in
particular, as far as I can think of.
This will improve on #1834; you will now no longer get a confusing error
(but still won't get a meaningful error; need to think how to do that).
* Adding support for haskell-ide-engine
* Work with the current directory if no stack.yaml file is found
* Added Cabal file detection, updated documentation and added tests
* Updated help
fixes#1738
- Replace previous `hh_client` usage with LSP client
- Add `HHAST` linter
- Split Hack from PHP: Hack is increasingly diverging from PHP:
- Hack tools do not understand PHP
- Most PHP tools do not handle Hack code well (including vim's syntax
highightling files)
- http://github.com/hhvm/vim-hack now sets filetype to `hack`
* Add kotlin languageserver linter definition
* Added kotlin languageserver references in docs, fix missing !! on other linters
* Added Vader tests for root path detection in Kotlin Language Server
* Rust Cargo linter: Improve workspace support
When using Cargo workspaces [1], there is a 'Cargo.toml' directory in a
top level directory, listing all the crates in the project. If we are
currently editing one of the crates, 'cargo build' should execute in
that directory for that crate's separate `Cargo.toml`, otherwise Cargo
may spend more time possibly rebuilding the entire workspace, and maybe
failing on one of the other crates, instead of succeeding on the current.
[1] https://doc.rust-lang.org/book/second-edition/ch14-03-cargo-workspaces.html
* Set `--parser` option in Prettier's fixer
* Add expected `--parser` option to tests
* Disable Prettier `--parser` detection if file extension exists
* Manually default Prettier `--parser` to "babylon"
* Add `--parser` test for TypeScript
* Add tests for Prettier `--parser`
* Add JSON5 to the suggested fixer for Prettier
* Guard the ballooneval settings
* Mark main objectives to do to get nice Hover
* Make tweaks to make the tooltip work - See " XXX: comments
* Guard balloon_show call
* Use return instead of finish for functions
* ale#hover#show : Add optional arguments to specify arbtirary position
This change is requested to be able to call the function with mouse
position to enable hover information in vim's balloon
* ale#ballon#Disable : Remove feature guards
* ale#balloon : Show 'ALEHover' output on balloon if no diagnostic found
* ale#hover#HandleLSPResponse : remove the check for cursor position
This check prevented the 'ALEHover in balloon' feature, since mouse
position is almost never cursor position.
* ale#balloon#MessageForPos : Change the return of balloonexpr
balloonexpr evaluation now works even without balloon_show for basic
diagnostics, leaving the balloon_show call to ale#hover#Show, which can
then feature guard the call to avoid errors
* ale#hover#Response : Feature guard balloon_show calls
* ale#hover : always display 'Hover' information in messages
Also add a small comment to warn readers the different outputs the
ale#hover#Show will write to
* {LSP,TS}Response : use only variables from the Response
It is clearer that we only rely on l:options to get the relevant data to
build the LSP Response string
* hover#ShowDetails : fix an issue where not having focus broke balloons
The issue was caused by not using a buffer-specific version of getline()
to cap the value of the column sent in the message to LSP. Therefore a
cursor on column 10 in an inactive window could send a message with
column=0, if the active window had a buffer with too few lines
* {LSP,TS}Response : Remove redundant checks for balloon_show call
With the upcoming change in ale_set_balloons default value (see Pull
Request w0rp/ale#1565), this check will be useless
* balloonexpr? : Add a flag to separate hover#Show() calls
The goal of this flag is to make `:ALEHover` calls not pop a balloon
under the cursor, since the user has probably no interest in their
cursor while typing the command
The flag is a default argument which is overridden only in ballonexpr
call of ale#hover#Show, and stays set in the hover_map until the
callback for the LSP handles it.
There are no automated tests for this feature right now, and the nature
of the addition (one optional argument in the API) should make it
transparent to existing tests.
Since the differentiation is now possible, the check for moved cursor
has been put back in ale#hover#HandleLSPResponse
* ale#hover#hover_map : Protect accesses to hover_map
Using get() is safer than trying to access directly with ., as the tests
show.
* Raise timeout to try to get Appveyor happy
* Review : Fix comments
* Review : pass the optional argument 'called_from_balloonexpr' in a Dict
This optional dictionary has documentation just before the function
using it, ale#hover#Show, and allows easier extension in the future.
Adding a couple of tests to demonstrate how IsCheckingBuffer behaves
during specific autocommand hooks:
* At ALELintPre, no linters have actually executed yet, hence
IsCheckingBuffer should be returning false.
* ALEJobStarted, fires as early as reasonably possible after a job has
successfully started, and hence hooking into IsCheckingBuffer here
should return true.
This distinction is important when using these two events during things
like statusline refreshes, namely for "linter running" indicators.
For now, it only detects undefined steps. The nearest `features` dir
above the buffer file is loaded, so step definitions should be found
correctly.
Tested only with Cucumber for Ruby, but it should work for any cucumber
that follows a substantially similar directory structure.
This removes the argument if the specified toolchain is empty.
As far as I can tell there is no +nighly (or similar) option [1] leading to
the termination of the server. But since people needed this option and
have yet to complain about it it stays the default for now.
[1] https://github.com/rust-lang-nursery/rls/blob/master/src/main.rs#L87
* Add first qmlfmt support
* Add GetCommand() function
- pass --error/-e option
* Add handle unittest
- fix pattern regex
- store col as integer
* Update docs
* Add command callback unit test
* Add fsc as a Scala linter
* Pull reused code into `autoload/ale/` directory
* Include fsc into the README
* Add unit test for testing the scala handler
* Add unit test for scala's fsc linter
* Rename scala unit tests for clarity
* Fix typo in README
* Fix typos in doc/ale.txt
* Fix author headline
* Put methods for fsc commands back into fsc.vim
* Move command_callback tests to correct location
* Rewrite handler test so it actually tests handler
* Clarify description of test in test_scala_handler
* handle multibyte string when linting text with redpen
* fix error when no string is provided, fix test's expect value
* remove ambiguious `==` operator
* Fixed (g)awk linter
* Made it secure, albeit less useful.
* Added gawk handler; the cpplint one was not working?
* Added gawk handler test.
* added warning to gawk handler.
* added gawk command callback test
* added comment about --source
* added back optional commandline option
* Handle flawfinder severity level
* Reverted code allowing Flawfinder to piggyback off of gcc's format handler
* Gave Flawfinder its own format handler and made requested changes.
Support for elm-format as a fixer has existed since Sept 2017, but it's not
easy to discover because the fixer was named `format`. This breaks the
convention of the other fixers that use the full name in the registry.
I've gone ahead and fixed this discrepancy, but I left the existing registry
entry in place. We should move people towards using `elm-format` as the fixer
name, but I'd hate to break existing setups.
* If a Perl script compiles, there are only warnings and no errors
* Let the first Perl error or warning win.
Take the following example:
***
sub foo {
my $thing;
***
This might have the following messages when we compile it:
Missing right curly or square bracket at warning.pl line 7, at end of
line
syntax error at warning.pl line 7, at EOF
warning.pl had compilation errors.
With the current behaviour, we just get a "syntax error" message, which
isn't all that helpful. With this patch we get "Missing right curly or
square bracket".
* Fix variable scope and pattern matching syntax
* Use named variable to enhance clarity when matching Perl output
* Add more tests for Perl linter
* Remove unnecessary parens
* Simplify check for pattern match
* Add configuration option to open lists vertically
* Add tests, clean up vertical list config
* Vertical list option cleanup
* Use is# for tests
* Order properties in documentation alphabetically
* Flawfinder support added for C and C++
A minor modification to gcc handler was made to support flawfinder's
single-line output format that does not have a space following the
colon denoting the warning level. gcc handler still passes its
Vader tests after this modification.
* Documentation fixes
* Revert documentation regression
* Added Flawfinder to table of contents
* Removed trailing whitespace
* Follow ALE conventions better
Added additional documentation and Vader tests
Erubi is yet another parser for eRuby. This is the default parser in
Rails as of version 5.1. It supports some additional syntax with similar
behavior to Rails' extensions to the language, though incompatible.
Rails currently still recommends their own syntax, so GetCommand still
has to do the translation introduced in
https://github.com/w0rp/ale/pull/1114 .
Erubi does not supply an executable—It is intended to be invoked only
from within a Ruby program. In this case, a one-liner on the command
line.
* When working on rust/cargo projects of varying sizes, it may be useful
to either build all possible features (i.e. lint all possible
conditionally compiled code), or even turn off other features for a
quicker edit-lint cycle (e.g. for large projects with large build times)
* Added a g:ale_rust_cargo_default_feature_behavior flag for instructing
cargo to not build any features at all (via `--no-default-features`),
building default features (via no extra flags), or building all possible
features (via `--all-features`)
* Also added a g:ale_rust_cargo_include_features flag for including
arbitrary features to be checked by cargo. When coupled with
g:ale_rust_cargo_default_feature_behavior this allows for full
customization of what features are checked and which ones are ignored
Typically proto files depend on and make use of proto definitions in
other files. When invoking protoc user can supply paths to inspect for
dependencies.
This patch makes it possible to configure flags passed to protoc. This
makes it e.g., possible to change include paths of the linter's protoc
invocation.
The test already handled arbitrary paths reasonably well, but setting
the directory interfered via leakage with others tests for some reason.
This patch removes the call to `SetDirectory` in the fixture setup and
the subsequent cleanup in the teardown as they are not required.
rustfmt normally acts on a file in place, and applies configuration
from rustfmt.toml files according to the path of the file.
Using a temporary file for rustfmt breaks this functionality, so
removing the '%t' from the rustfmt command.
Switches all vale instances to JSON output and provides an appropriate
handler for that. Without JSON, no end_col is provided and text
highlighting only catches the first character of every result.
