:hammer: Migrate codegen to {boilerplate.rs}

[serkor1]

📚 What?

TA-Lib, and therfore {talib}, is a huge library. It supports +150 (depending on how you count it) different indicators in separate files. The R wrappers were handwritten in the early development days, and as more more indicators were wrapped, even small additions/corrections triggered a full copy+paste, find-and-replace and other small makeshift BASH commands.
And then I finally opted for a full-scale BASH-templating approach first manually, and then with the assistance of AI (Thank you OpenAI and Anthropic).
While this approach did ease the development by alot, it also introduced a whole new branch of bugs, which shifted the focus from properly developing the TA-Lib wrapper, and a solution that were not portable to other projects; so each project were basically starting from scratch on the boilerplate generation.

Either way, there is now a tool called {boilerplate.rs} (which I have developed, with the help of AI) that abstracts the boilerplate generation, and is generally portable to other projects. {boilerplate.rs} will be implemented in the repository to reduce the size of this repository, reduce the exposure of bugs and generally test out and improve {boilerplate.rs}.

📋 TODO

The implementation will be done in two phases (c-wrappers, then R-wrappers (unit-tests included)), and the TODO list applies to both.

• Map gen_code/indicators to config.toml
• Recode splices to {boilerplate.rs} protected regions (Found in templates/)
• Recode key-value pairs
• Stage + commit
• Check diffs (Rinse and repeat untill the diff is limited to key-value pairs)

📖 Notes and gotcha's

• With the new release of TA-Lib the function signatures can be found in ta_codegen/input/<indicator>/<indicator>.c which are 1:1 mapable to R - which also implies that splices in .Call() are now redundant altogehter.

Note

These can be mined manually or through a simple BASH-script

• Each ta_codegen/input/<indicator>/<indicator>.md contains a formula section which can be mapped to .Rd with some modifcation so formulas are rendered correctly.

💣 Plan of Attack

• Make AI scan input/<input>/ twice to extract relevant parts of the C and R signatures by giving it a list of indicators, examples and formats. Remember to add formulas and indicator specific documentation.

• Construct relevant key-value pairs manually (Do not forget multicolumn output, and shifted pointers)

[serkor1]

Other than faster generation of files, there is no "real" benefit of going down that path.

An alternative is X-Macros built in TA-Lib Organization.

Sources:

• https://en.ittrip.xyz/c-language/advanced-c-x-macros

[serkor1]

X-Macros can be handled downstream - A small example is created here: https://github.com/serkor1/X-macros and the macros supports variadic signatures (Its unclear which version of C that supports it, and how it affects in the minimum R verison). See here: https://stackoverflow.com/questions/679979/how-to-make-a-variadic-macro-variable-number-of-arguments

Either way:

1. Stick to BASH for extracting arguments and signatures
2. Use X-macros to construct the C-functions

It seems that the R side is the only side that would benefit from a more efficient boilerplate generation.