hershel-theodore-layton/portable-hack-ast-linters

Hack source code linters authored with portable-hack-ast.


License
MIT

Documentation

portable-hack-ast-linters

Hack source code linters authored with portable-hack-ast

How to lint my code

If you've used HHAST before, you might expect that you'd be able to add a hhast-lint.json or alike to your repository to start linting right away. PhaLinters does not work this way. You must add a Hack source code file which does the linting and call the linters directly. The linters are in the HTL\PhaLinters namespace, so autocomplete away.

This library lints itself. You can check lint.hack and copy it verbatim. This file is explicitly licensed as MIT-0, not MIT like the rest of this library. This means you can do all the things permitted by the MIT license, without having to keep the license comment in the file intact. This choice of license does not change the license for the rest of this project.

Which linters are included?

For the full list, see bundled linters, almost all linters from HHAST are included, and some never before seen linters, only available in portable-hack-ast-linters.

Missing linters from HHAST

Some linters from HHAST are not included in this library.

  • HHClientLinter.hack
    • This linter does not use the HHAST framework. It is a wrapper around the hh_client --lint cli. The lints that hh_client suggest are very high quality and can/do utilize type information. The HHAST implementation has serious performance problems. A separate tool could be developed, since it doesn't need anything from HHAST (or Pha for that matter).
  • DataProviderTypesLinter.hack
    • This linter was a well intentioned attempt at making <<DataProvider(...)>> annotations typesafe. By its nature, it is a crude heuristic based approximation of a typechecker. It suggests you do things that are actively harmful like abusing the nothing type. The "real" solution is to the change the <<DataProvider>> mechanism to complement the Hack language.
  • MustUseOverrideAttributeLinter.hack
    • The output of this linter depends not just on the arguments, but on global information. A change in a parent class will require a file to be linted again. Caching lint results in a cross request manner would require extensive dependency tracking. hh_client --lint does the job of this linter better and doesn't depend on the \ReflectionClass api.
  • StrictModeOnlyLinter.hack
    • This linter was needed in the early days of Hack. Files used to have the .php or .hh extension and the mode was determined by a comment. <?hh would "kick your .php file into Hack partial mode". You'd explicitly enable "strict mode" by using <?hh // strict. Since the release of hhvm version 4.0 you can use .hack files.

Fixme directives

Lint errors can't (and shouldn't) always be fixed. Sometimes an await-in-a-loop isn't a sign of a false dependency for example in sgml-stream's ConcurrentReusableRenderer:

async function consider_the_following_code_async(): void {
  $awaitables = vec[];
  foreach ($snippets as $snippet) {
    $awaitables[] = $snippet->primeAsync($descendant_flow);
  }

  concurrent {
    // Race them all,...
    await AwaitAllWaitHandle::fromVec($awaitables);
    await async {
      foreach ($snippets as $snippet) {
        /* HHAST_IGNORE_ERROR[DontAwaitInALoop]
         * feedBytesToConsumer operates on the awaitables from the race.
         * There are no false dependencies here.
         * We just MUST collect bytes in order. */
        await $snippet->feedBytesToConsumerAsync($consumer, $successor_flow);
      }
    };
  }
}

The $snippet->feedBytesToConsumerAsync(...) method awaits an Awaitable that was already started in ->primeAsync(...). Internally, it also awaits on $consumer->consumeAsync(...) which must be called sequentially. This await-in-a-loop is required for the code to work correctly. To inform hhast, a HHAST_IGNORE_ERROR[DontAwaitInALoop] comment was added.

This comment-to-suppress-a-lint-error mechanism has always bothered me a little. I subscribe wholehartedly to the following quote from the CppCoreGuidelines:

Compilers don’t read comments ...
and neither do many programmers (consistently).

As a replacement for fixme comments, PhaLinters uses the pragma(...) directive and the <<Pragmas(...)>> annotation. The following pieces of code express the same intent.

/// hack
/* HHAST_FIXME_ERROR[DontAwaitInALoop] I'll get around to it. */
await $object->methodAsync();

// I'll get around to it.
pragma('PhaLinters', 'fixme:dont_await_in_a_loop');
await $object->methodAsync();