svdtools: install rust version

[tones111]

This change removes a dependency on python/pip and is more consistent with the installation of the other supporting tools.

[tones111]

The CI needs the cache cleared but I'm not familiar with how to do that.

[Rahix]

You have to bump this cache key to have CI generate a new tools cache:

avr-device/.github/workflows/ci.yml

Line 37 in 07f32b8

key: rust-tools-20250104-001

[tones111]

The initial effort was a little too ambitious as there have been some breaking changes to the patch file processing. I'd like to compare against the current patched svd files, but I haven't managed to get the python version of svdtools to build. Is there somewhere I could download them from?

[Rahix]

You can download the CI artifacts from main branch, e.g. here: https://github.com/Rahix/avr-device/actions/runs/12615563082/artifacts/2386623070

[Rahix]

Do post the the diff of the generated svds in this pull request please :)

[Rahix]

@tones111, check the changes in this PR https://github.com/Rahix/avr-device/pull/157/files, I think a large part of the patch file porting has already been done there.

[tones111]

Took a bit to figure out the yaml syntax and why it was complaining but the changes to resolve ended up being pretty straightforward. I committed the generated svd files, so once we've looked at them I will remove it.

When comparing against the svd files @Rahix linked above there were some formatting improvements. Running the original patched files through xmllint --format really helped. Doing some spot checks it looks like the main difference is that the new svdtools removes the <usage>...</usage> from enumerated values (which makes sense).

[tones111]

avr-device.svd.diff.txt

Here's the diff against an xmllint version of 0.7.0. I see the following differences:

• <usage>...</usage> removed from enumerals
• improved description whitespace
• unicode improvements (ex: atmega128rfa1)
• Some ISC1 interrupt enumerals are removed (ex: attiny167, attiny2313, etc)
• Some ADC enumerals are reordered in attiny84

[Rahix]

Some ISC1 interrupt enumerals are removed

They are removed because the derived enumerals from ISC0 already include properly named values for them. This is still a breaking change, but one we do want in my eyes.

Some ADC enumerals are reordered in attiny84

Looks okay to me.

---

As a final check, you can diff the generated rust sources of each version. You can find the original ones in the crate archive contained in target/package/ dir of the artifacts archive. The .crate file is just a .tar.gz archive of the sources.

I would expect this diff to only show the whitespace changes and the removed ISC1 enumerated values.

[Rahix]

Changes look fine, after the aforementioned cross-check of the generated sources and removal of the comitted SVDs, this is ready for merge :)

BTW, comitting the SVDs wasn't necessary - CI again created an artifacts archive that includes them for your PR: https://github.com/Rahix/avr-device/actions/runs/12625939793?pr=174

[Rahix]

Ah, just noticed svdtools version 0.4.0 is hot off the stove, maybe we can update this PR to the new version immediately to reduce work later on? :)

[tones111]

Here's a new diff (against main svds) for the update to svdtools 0.4.0.
svdtools_040.diff.txt

A few more tweaks were required that resulted in some minor corrections

• updated atmega328pb TWAM1 description
• atmega8 UPE field is named PE (matches datasheet)
• attiny841 ACO{0,1} are read-only (matches datasheet)

However, it looks like the timer enum changes from #173 aren't working but I'm not sure why.

[tones111]

Maybe the atmega128rfa1 enum changes are there. The svd file has

<enumeratedValues derivedFrom="COM0B"/>

and the COM0B looks like it matches the patch file. So, I think the new version of svdtools knows how to deduplicate the enums.

Update: Ah, this is the same as the ISC1 enumerals above. As you mention, it knows that there are no changes from the derived enum.

[Rahix]

Thanks so much for taking care of this!