Make it possible to add a colrv1 json test without altering a match

[rsheeter]

#817 broke the ability to add a test, run with an environment variable set to generate the master data, and then have your test work. Instead it will initially fail until a match statement is updated.

For context, cross-crate path building breaks in google3. A build.rs based approach, such as #816, should work but to date I haven't gotten that to function in google3.

[drott]

Not super happy #817 got merged as is, was there a rush or time pressure? #817 requires redundantly defining the same list that's already defined here: https://github.com/googlefonts/fontations/blob/main/skrifa/src/color/traversal_tests/mod.rs#L338
Can the file list in the change be generated from that list by extending the macro?

[rsheeter]

For context the pressure is that it's blocking import into google3 which is something we're actively trying to get online so we can start migrating C++ code => safe code. Perhaps I'm wrong but I don't think stalling that to avoid a papercut when adding a test of a specific type is a good tradeoff.

[drott]

I certainly see the value of that, but it seems like this is not the only obstacle or step to be completed to achieve that. We also usually expect incoming changes not to break existing stuff, and it surprised me that we were willing to sidestep that here.

Thanks for looking into it. One approach might be to move the test definition data, such as the variable axis coordinate configurations etc., and the glyph sequences into a static array in font_test data. Then that test raw data can be used to generate the file list, as well as then be referenced in traversal tests to execute and generate the test binaries.

[rsheeter]

it seems like this is not the only obstacle or step to be completed to achieve that

In testing locally it appears that after #817 Skrifa can build and test on google3.

One approach might be to move the test definition data, such as the variable axis coordinate configurations etc., and the glyph sequences into a static array in font_test data

Ty, that's worth considering.

[rsheeter]

Note to self: per f2f, try cargo package, perhaps the requisite files are simply not being pushed to crates. That might explain why the build.rs approach didn't work in google3.

[rsheeter]

When I package locally I do seem to see the expected files:

$ cargo package
$ tar -tvf target/package/font-test-data-0.1.2.crate | grep colrv1
lists all the test files