A repository isn't a payload
Four extensions had to become downloadable packages, and the packer refused one of them, which turned out to be the right call.
Only runtime files
Each extension lives in its own repository alongside a README, licence, store artwork and tests, and since none of that belongs in what a browser loads, a shared packer stages only the runtime files.
The refusal
Chrome rejects any top-level name starting with an underscore except _locales, and one tree held a _metadata folder. That's something Chrome writes into copies it installed itself, and this one had been carried back from an installed profile long ago, staying invisible to a clean git status.
If it had shipped, it would have downloaded, verified and unpacked successfully, then failed to load on a customer's machine minutes later, with nothing pointing at the cause. It's now dropped automatically, while any other reserved name still stops the pack with the source path in the message.
Loading before publishing
Packing correctly isn't the same as loading, though. Since the Hub already loads extensions over the DevTools pipe, every package goes through Extensions.loadUnpacked before anything is published:
load antiblur-1.0.0 ok load argon-0.1.0 ok load sodium-3.4 ok load typeswap-1.0.0 ok
The packer also stages into a real directory before zipping, rather than passing exclusions to zip, because a staging directory can be inspected, whereas an exclusion pattern that silently fails to match can't be.
This is a shorter version of a post on the Pioneering blog, so if you want the full story, read the original on pioneering.app.
Next postCapturing a window that is completely covered
