Expose as a reflection

[autaut03]

Hey for the third time :)

I'm the author of https://github.com/good-php/reflection - a small reflection library that's suitable for runtime usage and supports parsing PHPDocs. It works, but there are two major downsides:

• with cold cache, it's relatively slow (milliseconds to reflect a simple class structure, while native reflection is in nanoseconds range)
• it's hard to keep up with PHP features, PHPStan features as well as just with PHP quirks

This would be the case with every other reflection library written in PHP. It's either an insane maintenance burden and complexity, or "slow" speed.

So having Mago which correctly handles reflection of internal classes, PHPDoc types, inheritance and other things - I thought it could be useful to provide some of Mago's internals as a fast extension, written with https://ext-php.rs/getting-started/hello_world.html and using Mago's crates.

The questions are:

• is this feasible?
• is it possible to extract reflection-like data using one of Mago's crates, without analyzing the whole codebase?
• how stable is the API of those crates?

[azjezz]

is this feasible?

Yea, all mago crates are published on crates.io, so you can use them.

is it possible to extract reflection-like data using one of Mago's crates, without analyzing the whole codebase?

Yes, we actually have mago_prelude crate that pre-builds CodebaseMetadata containing all built-in symbols, you can then get code from user, parse it, scan it for CodebaseMetadata, merge it with the existing one which contains built-in symbols, and then populating the final codebase to link things up.

how stable is the API of those crates?

We are currently breaking API whenever a new version is released, if you stick to one version ( e.g 1.0.0-beta.12, you will be fine, then you can upgrade whenever a new version of mago is out and see if you need to change anything.

The parser ( mago_syntax ) and reflection ( mago_codex ) are unlikely to change much, we did have major BC-breaks before, but i don't see something like that happening soon.

[autaut03]

Yes, we actually have mago_prelude crate that pre-builds CodebaseMetadata containing all built-in symbols, you can then get code from user, parse it, scan it for CodebaseMetadata, merge it with the existing one which contains built-in symbols, and then populating the final codebase to link things up.

Would that be more performant than the couple milliseconds I achieve with the PHP implementation, from your estimates? Or is it hard to tell?