Program analysis for ReScript and OCaml projects targeting JS (ReScript) as well as native code (dune):

• Globally dead values, redundant optional arguments, dead modules, dead types (records and variants).
• Exception analysis.
• Termination.

Early release. While the core functionality is reasonably stable, the CLI and annotations are subject to change. However, this is a tiny surface at the moment.

For correct handling of ReasonReact components in the Dead Code Analysis, bucklescript version 7.3.2 is required (fixes the location in the JSX PPX).

The rest of this document describes the dead code analysis. For the Exception Analysis, build instructions are the same, and the command-line invocation is different.

Build and run on existing projects using the Build and Try instructions below. The analysis uses .cmt[i] files which are generated during compilation, so should be run after building your project. Remember to rebuild the project before running again.

# dead code analysis reanalyze.exe -dce # exception analysis reanalyze.exe -exception

The requirement is that bsconfig.json can be found by walking up the current directory.

# dead code analysis reanalyze.exe -dce-cmt root/containing/cmt/files # exception analysis reanalyze.exe -exception-cmt root/containing/cmt/files

Subdirectories are scanned recursively looking for .cmt[i] files.

The requirement is that the current directory is where file paths start from. So if the file path seen by the compiler is relative src/core/version.ml then the current directory should contain src as a subdirectory. The analysis only reports on existing files, so getting this wrong means no reporting.

The dead code analysis reports on globally dead values, redundant optional arguments, dead modules, dead types (records and variants).

A value x is dead if it is never used, or if it is used by a value which itself is dead (transitivity). At the top level, function calls such as Js.log(x), or other expressions that might cause side effects, keep value x live.

An optional argument ~argName to a function is redundant if all the calls to the function supply the argument, or if no call does.

A module is considered dead if all the elements defined it in are dead.

The type analysis repots on variant cases, and record labels.

• A variant case | A(int) is dead if a value such as A(3) is never constructed. But it can be deconstructed via pattern matching | A(n) => ... or checked for equality x == A(3) without making the case A live.

• A record label x in type r = {x:int, y:int} is dead if it is never read (by direct access r.x or pattern matching | {x:n, y:m} => ...). However, creating a value let r = {x:3, y:4} does not make x and y live. Note that reading a value r does not make r.x or r.y live.

While dead values can be removed automatically (see below), dead types require a bit more work. A dead variant case requires changing the type definition, and the various accesses to it. A dead record label requires changing the type definition, and removing the label from any expressions that create a value of that type.

The dead code analysis supports 2 annotations:

• @dead suppresses reporting on the value/type, but can also be used to force the analysis to consider a value as dead. Typically used to acknowledge cases of dead code you are not planning to address right now, but can be searched easily later.

• @live tells the analysis that the value should be considered live, even though it might appear to be dead. This is typically used in case of FFI where there are indirect ways to access values. In case of bucklescript projects using genType, export annotations immediately qualify values as live, because they are potentially reachable from JS.

The main difference between @dead and @live is the transitive behaviour: @dead values don't keep alive values they use, while @live values do.

Several examples can be found in examples/deadcode/src/DeadTest.res

Takes a comma-separated list of path-prefixes. Don't report on files whose path has a prefix in the list (but still use them for analysis).

reanalyze.exe -suppress one/path,another/path

Takes a comma-separated list of path-prefixes. Report on files whose path has a prefix in the list, overriding -suppress (no-op if -suppress is not specified).

reanalyze.exe -unsuppress one/path,another/path/File.res

Print debug information during the analysis

reanalyze.exe -debug ...

This overwrites your source files automatically with dead code annotations:

reanalyze.exe -write ...

There's a dead code ppx (values only, not types) in this repository. It can be used to automatically remove code annotated @dead, as if it had been commented out. Can be used after adding annotations automatically. The combination of automatic annotation and automatic elimination is a form of automatic dead code elimination. For projects that use a library, or that in general have code which is dead only temporarily. There's obviously a level of risk in doing this automatic elimination. The safety net you can rely on is that the code must still compile.

This automatically annotates @live all the items called foo or bar:

-live-names foo,bar

This automatically annotates @live all the items in file Hello.res:

-live-paths Hello.res

This automatically annotates @live all the items in the src/test and tmp folders:

-live-paths src/test,tmp

The CLI options -suppress and -unsuppress can also be set in bsconfig.json as follows:

{ "reanalyze": { "suppress": ["src/ToSuppress.res"], "unsuppress": [] } }

This is equivalent to adding --suppress and --unsuppress options at the beginning of the command-line. Note that the options are additive, so it's possible to suppress partially in the config and partially on the command line.

npm add --save-dev reanalyze 

opam switch 4.06.1 eval $(opam env) opam install dune dune build # _build/default/src/Reanalyze.exe

opam install dune dune build # _build/default/src/Reanalyze.exe

npm run build # or whatever command to build the project npm add --save-dev reanalyze npx reanalyze -dce

See for example project reanalyze-esy-example:

git clone https://github.com/cristianoc/reanalyze-esy-example cd reanalyze-esy-example esy esy dce esy check-exceptions

echo "let unused = 34" > test.ml ocamlc -c -bin-annot test.ml reanalyze.exe -dce-cmt ./test.cmt Warning Dead Value test.ml 1:1-15 unused is never used <-- line 1 let unused = 34 [@@dead "unused"]

mkdir test echo "let unused = 34 let used = 42" > test/test.ml echo "let _ = Test.used" > test/use.ml cd test ocamlc -c -bin-annot *.ml reanalyze.exe -dce-cmt . Warning Dead Value test.ml 1:1-15 unused is never used <-- line 1 let unused = 34 [@@dead "unused"] let used = 42 Warning Dead Value use.ml 1:1-17 _ has no side effects and can be removed <-- line 1 let _ = Test.used [@@dead "_"]

How to test on Infer :

• Make sure that dune builds both .cmt and .cmti files (see ocaml/dune#3182 as to why):

--- a/infer/src/Makefile +++ b/infer/src/Makefile -DUNE_BUILD = dune build --profile=$(BUILD_MODE) +DUNE_BUILD = dune build @check @all --profile=$(BUILD_MODE) 

• Build normally

make -j BUILD_MODE=dev 

• Go to the right directory from which file paths start:

% cd infer/infer 

• Run the analysis

% path/to/reanalyze.exe -dce-cmt _build/default/src/.InferModules.objs/byte/