If a function is annotated with the log-macro the function's name gets logged on entry and exit:
- Log output before the actual function body executes
> my_function_name
- Log output just before the function returns
< my_function_name
This is inspired by how logging can be realized with concepts from Aspect Oriented Programming (AOP). Although this example is extremely simplified, it shows how additional logging can be added, merely through applying an attribute. Very similar to how annotation-based logging works, like for example in Java.
The "annotated"[1] function is parsed into an ItemFn, using the syn crate.
It reads out the function signature (especially name and return type) and the body.
Than the code is reconstructed with quote.
While an entry- and an exit-log-statement gets "injected" in the user-code.
Little interesting aside. It is important that the macro checks, whether the function returns anything else but the Unit () type.
Because if that is the case, the very last statment of the function has to be quoted before the exit-log-statment.
The parent crate is use-log-macro. And it, like the name implies, uses the #[log] macro on some example functions, in it's main.rs.
The log-macro is an attribute-macro and has to live in it's own crate.
Here it is nested as a library inside of use-log-macro, which references log by relative path declaration in it's Cargo.toml.
I think this pattern shows how it might be possible to inject all sorts of runtime measurement logic, simply by adding attributes to functions. For example benchmark calculations. Or reporting logic could be added and reduced in this very simple way. Maybe even by writing to a database. All without cluttering the primary logic of the functions.
[1] I know it is not called 'annotation' in Rust. But what would you say then: "The attributeapplied function...?"