Closed vv-monsalve closed 2 months ago
Please remember that you need to have the fixed version of harfbuzz installed on your computer to avoid the InDesign problem. If you are testing on InDesign 2023, I would use the CI builds.
Right. Going back now to a previous chat conversation, you said
To reproduce the GHA build you would need to build and install a new version of Harfbuzz on your computer, which is quite tricky. I would recommend just using the GHA builds since they will be the final build targets anyway.
However, we must provide building instructions for all the projects so any user can build the fonts. From the above, should we point the build instructions to rely solely on the CI run for all the users from now on?
Well, then we just make a requirement (in the documentation) that the user must use Harfbuzz >=8.4.0 on their system (Remember, this is a system package requirement, not a Python package) to build the fonts correctly, or they use the GHA versions.
Actually I may have come up with a cleaner solution, using Python subsetter for the build instead of hb-subset
We discussed making the CI fonts reliable ones for all users. Ensuring this requires:
This doesn't mean the local build can't be done, but the centralized, reliable fonts should be the CI ones. This would also impact the final Build instructions, which will need to be defined in the Readme file.
cc @casasin
@simoncozens, after testing the current fonts at commit
7dc6a43
in InDesign, and comparing the result with the latest control fonts from thelang-build
branch, the shaping differences are back for some families.~This is a regression from previous fonts on the
builder2
branch at commit4cf78b1
, where these shaping issues were absent.~Edit: The different results happen for the fonts produced locally after running the
make build
command, not those from the GHA. The CI-run fonts do not show these differences. What is needed to ensure 1:1 results between the local run and the CI build?Playwrite CO