Incompatible differences in commands

[Wassasin]

For nxp-pac we are slowly migrating to a more stm32-metapac style of generating the code from extracted data, instead of directly generating the code from the SVD. During migration I had the following issue:

The commands generate, extract-all and extract-peripheral work slightly different, resulting in inconsistent output.

For example, the generate command uses svd2ir::convert_svd whilst extract_all and extract_peripheral use svd2ir::convert_peripheral. In order to generate correct code, svd2ir::convert_svd adds extra module namespaces in the form of 'regs' and 'vals' to names.

You can clearly note the PAC's generated directly and PAC's that use extraction and then manually generating code by the presence of these names in the namespaces. As an additional consequence, files with transforms for generation may not be directly usable when extracting.

Making it more consistent may require the commands to optionally fall back to their older behaviour.

[inferiorhumanorgans]

Agreed. This is something I've run into with ra-pac as I've had to write a fair amount of IR for the 4M1. Using namespaces would make the transition to generating code from the extracted data quite a bit easier.

The transforms are typically not portable but the biggest issue is that any manually written peripheral-specific IR gets added to every generated block.

[inferiorhumanorgans]

Re namespacing:

Currently with namespacing enabled you get YAML files with individually namespaced blocks. When trying to reassemble them with gen-block the result is a structure of namespace::namespace::*. What is the best way to proceed here?

The files can't easily be concatenated and passed to form as they include inner directives.

Does it make sense to change gen-block to handle multiple input files?

Should extract-peripheral strip the toplevel namespace before generation?