OCaml build tools & protocols Last updated Feb 11, 2025
Building software involves interfaces and protocols. The interfaces are determined by the command-line syntaxes of the tools. But constructing a good command line is not enough. To take the obvious example, to compile a module you construct a command line that contains all of its dependencies - that’s interface. But before you can execute that command, you must build the dependencies - that’s protocol.
Build protocols also involve discovery of such information as the
location of dependencies in the file system. To depend on a module,
you not only have to build it, you have to add the directory that
contains it to the compiler’s search path by adding -I path/to/dir
to the command line. But path/to/dir must be discovered by the build
system.
In Bazel, discovery of such information is an active process: the implementation of the rule that compiles modules is given a set of dependencies, which it must interrogate (using the Bazel API) in order to obtain paths and other information.
Build protocol also involves configuration of the build environment. For example, in the case of Bazel it is not enough to discover a dependency and its location, and then construct a command line. The dependency must also be explicitly registered as an input to the build action, which can be thought of as a kind of configuration action. To ensure replicability, Bazel ensures that only registered inputs will be accessible to build actions. A correctly-constructed command line whose inputs are not registered will fail because the build tool will not be able to find its inputs.
Each of the build tools in the OCaml distribution has its own command-line syntax and protocol.
The task for any build tool is to integrate OCaml’s build tools and protocols into its own interfaces and protocols. Bazel has a strict set of protocols that do not correspond exactly to OCaml’s build protocols, so a major design task for OBazl is to make them get along.