Modules (form)

[burrbull]

Closes #424 .

r? @therealprof

Speed tests are needed

[therealprof]

That's very nice! Need to have a closer look this evening.

[burrbull]

@therealprof you promised to look

[therealprof]

Didn't get to it, sorry. Maybe I can squeeze it in later today.

[therealprof]

I had a look. svd2rust runtime goes through the roof with the -foption, it takes 10x as long in my tests to dump the code. There's no change in build/check times of the generated code.

It would be great to see some noticeable benefit somewhere.

[therealprof]

The Module is a rather weird implementation causing quite a few clippy lints and warnings in case the superfluous #[allow(dead_code)] are removed.

We should also get rid of all the panics and instead implement proper error handling.

[duskmoon314]

Any plan on this?

Recently I tried the logic of xtask in esp-pacs on d1-pac, which first generates lib.rs and then transforms it into modules in src. Much simpler than this implementation in my mind.
esp-pacs
d1-pac

it takes 10x as long in my tests to dump the code.

I am not familiar with improving performance. Taking a simple look at the flame graph of executing xtask gen d1, I found that form::create_directory_structure is indeed costly. But I don't feel it annoying since I always use form after generating code with svd2rust.

The flame graph:

[rmsyn]

@burrbull you mentioned that you would like me to take a look at bringing this PR over the finish line.

Happy to do so. Would you mind rebasing on the latest HEAD to make finishing the work a bit easier?

[burrbull]

@burrbull you mentioned that you would like me to take a look at bringing this PR over the finish line.

Happy to do so. Would you mind rebasing on the latest HEAD to make finishing the work a bit easier?

I need to remember what I've tried to do and try to solve perfomance issues.