Skip to content

Latest commit

 

History

History
236 lines (169 loc) · 9.94 KB

CONTRIBUTING.md

File metadata and controls

236 lines (169 loc) · 9.94 KB
Raw

Contribution Guidelines

Reporting Issues

If you encounter a problem when using infer or if you have any questions, please open a GitHub issue.

Hacking on the Code

We welcome contributions via pull requests on GitHub.

Development Dependencies

You'll want to install a few more dependencies to comfortably hack on the infer codebase. Simply run:

make devsetup

Building Infer for Development

  • Build the code faster: make -j BUILD_MODE=default. By default make builds infer with flambda enabled, which makes it very slow (but makes infer significantly faster).

  • Faster edit/build cycle when working on OCaml code inside infer/src/: build inside infer/src/ (skips building the models after infer has been built), and build bytecode instead of native: make -j -C infer/src byte. You need to have run make -j (with or without BUILD_MODE=default) at some point before.

  • In general, make commands from the root of the repository make sure that dependencies are in a consistent and up-to-date state (e.g., they rebuild infer and the models before running steps that use infer), while running make commands from within subdirectories generally assumes that dependencies are already up-to-date.

    For instance, running make direct_java_infer_test will rebuild infer and the models if necessary before running the test, but running make -C infer/tests/codetoanalyze/java/infer test will just execute the test.

  • To switch the default build mode to flambda disabled, you can export BUILD_MODE=default in your shell.

Debugging OCaml Code

  • Printf-debug using Logging.debug_dev. It comes with a warning so that you don't accidentally push code with calls to debug_dev to the repo.

  • Browse the documentation of OCaml modules in your browser with make doc

  • When using ocamldebug, and in particular when setting break points with break @ <module> <line> don't forget that an infer module M is in reality called InferModules__M, or InferBase__M, or ... See the html documentation of the OCaml modules from make doc if you're unsure of a module name.

$ ledit ocamldebug infer/bin/infer.bc
(ocd) break @ InferModules__InferAnalyze 100
Breakpoint 1 at 9409684: file backend/InferAnalyze.ml, line 99, characters 18-78

  • To test the infer OCaml code you can use the OCaml toplevel. To build the OCaml toplevel with the infer modules pre-loaded, run make toplevel and follow the instructions.

    To pass infer options to the toplevel, use INFER_ARGS, for instance: INFER_ARGS=--debug^-o^infer-out-foo.

    Many operations require the results directory and database to be initialized with ResultsDir.assert_results_dir "".

Hacking on the Code in facebook-clang-plugins

Infer uses ASTExporter from the facebook-clang-plugins repository. To change that part of the code:

  1. Create a pull request in facebook-clang-plugins with the changes.
  2. Create a pull request in this repository updating the version of the git submodule.

Contributor License Agreement

We require contributors to sign our Contributor License Agreement. In order for us to review and merge your code, please sign up at https://code.facebook.com/cla. If you have any questions, please drop us a line at [email protected].

You are also expected to follow the Code of Conduct, so please read that if you are a new contributor.

Thanks!

Coding Style

All Languages

  • Indent with spaces, not tabs.

  • Line width limit is 100 characters (except for Python for which the limit is 80).

  • In general, follow the style of surrounding code.

OCaml

  • The module IStd (infer/src/istd/IStd.ml) is automatically opened in every file. Beware that this can cause weird errors such as:
$ pwd
/somewhere/infer/infer/src
$ cat base/toto.ml
let b = List.mem true [true; false]
$ make
[...]
File "base/toto.ml", line 1, characters 17-21:
Error: This variant expression is expected to have type 'a list
       The constructor true does not belong to type list

  • All modules open IStd using open! IStd. This is to make that fact more explicit (there's also the compilation flag mentioned above), and also it helps merlin find the right types. In particular this also opens Core.Std.

  • Do not add anything to IStd unless you have a compelling reason to do so, for instance if you find some utility function is missing and is not provided by Core.

  • Polymorphic equality is disabled; use type-specific equality instead, even for primitive types (e.g., Int.equal). However, if your module uses a lot of polymorphic variants with no arguments you may safely open PolyVariantEqual.

    If you try and use polymorphic equality = in your code you will get a compilation error, such as:

Error: This expression has type int but an expression was expected of type
         [ `no_polymorphic_compare ]

  • Alias and use module L = Logging for all your logging needs. Refer to its API in Logging.mli for documentation.

  • Check that your code compiles without warnings with make -j test_build (this also runs as part of make test).

  • Apart from IStd and PolyVariantEqual, refrain from globally opening modules. Using local open instead when it improves readability: let open MyModule in ....

  • Avoid the use of module aliases, except for the following commonly-aliased modules. Use module aliases consistently (e.g., do not alias L to a module other than Logging).

module CLOpt = CommandLineOption
module F = Format
module L = Logging
module MF = MarkupFormatter

  • Use [@@deriving compare] to write comparison functions whenever possible. Watch out for this issue when writing type nonrec t = t [@@deriving compare].

  • Use let equal_foo = [%compare.equal : foo] to write equality functions whenever possible.

  • Use named arguments whenever the purpose of the argument is not immediately obvious. In particular, use named arguments for boolean and integer parameters unless the name of the function mentions them explicitly. Also use named arguments to disambiguate between several arguments of the same type.

  • Use named arguments for functions taken as argument; it is common to name a function argument f. For instance: List.map : 'a list -> f:('a -> 'b) -> 'b list.

  • In modules defining a type t, functions that take an argument of that type should generally have that argument come first, except for for optional arguments: val f : ?optional:bool -> t -> ....

  • Use the _hum suffix to flag functions that output human-readable strings.

  • Format code with ocamlformat.

C/C++/Objective-C

Follow clang-format (see ".clang-format" at the root of the repository).

Testing your Changes

  • Make sure infer builds: make -j test_build. Refer to the installation document for details.

  • Run the tests: make -j 4 test (adjust 4 to the number of cores available of your machine). The tests (almost) all consist of the same three ingredients:

    1. Some source code to run infer on.
    2. An "issues.exp" file where each line represents one item of output of the test. For most tests, one line is one issue reported by infer.
    3. A Makefile that orchestrates the test, for instance running infer on the source code and comparing the results with issues.exp using diff.

  • If your changes modified some of the expected outputs and if the changes make sense, you can update the expected test results by running make test-replace.

  • If relevant, add a test for your change.

  • To add a test that infer finds (or does not find) a particular issue, add your test in "infer/tests/codetoanalyze/{language}/{analyzer}/". Look at the Makefile in that directory and make sure it runs your test. "{analyzer}" is often an infer analyzer (as in infer -a {analyzer}), with some special cases:

    • "errors" is "infer"
    • "frontend" is a mode where the expected output is the result of the translation of the program by infer's clang frontend into infer's intermediate representation.

    Name the procedures in your test following these conventions:

    • Test procedures where the analyzer should report an error should end with the suffix Bad.
    • Test procedures where the analyzer should not report an error should end with the suffix Ok.
    • Test procedures documenting current limitations of the analyzer should have the prefix FP_ (for "false positive") or FN_ (for "false negative") and a comment explaining why the analyzer gets the wrong answer.

  • To add a test that a certain build system integration or a command-line option works in a certain way, add a test in "infer/tests/build_systems/".

  • If you created a new Makefile for your test, add it to the root "Makefile", either to the DIRECT_TESTS (first case) or to the BUILD_SYSTEMS_TESTS variable (second case). Gate the test appropriately if it depends on Java or Clang or Xcode (see how other tests do it).

  • It can be useful to look at the debug HTML output of infer to see the detail of the symbolic execution. For instance:

$ infer --debug -- clang -c examples/hello.c
$ firefox infer-out/captured/hello.c.*.html

Updating opam and opam.locked

tl; dr: Run make opam.locked.

opam.locked records fixed versions of the opam dependencies known to work with infer and to respect the constraints in opam. This prevents unpredictable breakages of infer or its dependencies, especially for infer releases, for which it is more difficult to change their package constraints after the fact.

To add an opam package or update its version constraints, edit 'opam' then run make opam.locked.