Effortless MATLAB Solutions: AI Code Generator Tool
https://zurl.co/RA7ha
#MATLAB #CodeGenerator #AITool #Efficiency #Technology #Programming #Automation #Productivity #Innovation #Solutions

Thanks to @made I recently learned that dynamic code generation doesn't play nice with certain Content Security Policies (CSP). This has a major impact on a few core packages in #ThingUmbrella, like the https://thi.ng/vectors package which contains ~900 vector functions, most of them code generated and optimized for different vector sizes/dimensions (incl. n-dimensional versions). This package (and some others using a similar approach) are key dependencies for dozens of other geometry/visualization related packages... However, I found code generation the only way to practically manage & maintain the sheer amount of functionality provided.

Because of this (CSP impact), I've been working on a new code generator, which converts the dynamically generated code into statically generated source code files. This will make the overall initial package size bigger, but this shouldn't be a major problem in practice, since there're also very positive effects, including:

- The new format allows for doc strings for _all_ generated vector ops (with the dynamic approach there was no way to properly attach those in TypeScript)
- The new file structure (single function per source file) massively helps with dead code elimination when using a bundler, resulting in smaller final file/bundle sizes. When NOT using a bundler, similar filesize savings can be had by using direct imports (to individual functions) rather than full package imports
- None of the unused versions need to be code generated at runtime anymore, so also improving startup time

The new codegen is already covering around more than a third of the 900 ops. If you want to keep an eye on progress & discussion, follow this issue:

https://github.com/thi-ng/umbrella/issues/497

Lapidary-render 0.12.0 is now released.

To the best of my knowledge it's the only #codeGenerator that properly handles #jsonSchema anyOf and allOf.
Also has support for oneOf, but not per the specs.

https://github.com/python-lapidary/lapidary-render/releases/tag/v0.12.0

How to structure a #python project with both generated and authored code?

Python tools don't handle well anything other than $root/$package or $root/src/$package

Code generator removes files that it finds, assuming they're outdated

editors can handle a separate subdirectory, but using linter or tests would required synchronizing that directory with the main source directory.

Each of these solutions sucks.

I'm thinking of designating special sub-package `extras` would do the work, it wouldn't clash with generated code, but perhaps others have better ideas.

I wanted to build a #python client for some #openSource project that offers a good #openApi 3.0 description - just to test my generator on real-live service, but also to support the community.

I looked at @forgejo but it turns out it uses swagger 2. I didn't realise #golang was old enough...

Any way, other ideas would be very welcome.

Effortless MATLAB Solutions: AI Code Generator Tool
Generate MATLAB code solutions effortlessly with our AI tool.
https://zurl.co/6y8eG
#MATLAB #CodeGenerator #AITool #Efficiency #Technology #Programming #Automation #Productivity #Innovation #Solutions

Generating #python code: #jinja or libcst?