Closed ahelwer closed 10 months ago
I agree that copying specs should be the default. The reasons for introducing submodules for CCF, BlockingQueue, and AzureCosmosDB are:
CCF:
BlockingQueue:
Azure Cosmos DB
For posterity, we should consider setting up github mirroring to, e.g., Gitlab, our Inria machine, or to the Linux Foundation. IMO this should also include wiki pages and Github issues (I once looked into distributed issue trackers to avoid Github vendor lock-in, but they were too inconvenient to use).
AFAIK, the Github action/checkout can clone submodules. Would that help with the CI workflow validation?
You're right action/checkout can clone submodules. Involving submodules might complicate things if we can't massage the submodule module/model names & locations such that the manifest generation script will be able to recreate everything nicely. Or if people try to run manifest generation without having checked out submodules, although a step can be added to the instructions for that.
Was briefly discussed earlier in the year but IMO it would be better to just copy the specs into the repo. Reasons:
Thoughts? I will say that adding the specs as submodules is preferable to them only existing as a hyperlink though. One of the submodules added exposed a bug in tree-sitter-tlaplus pluscal parsing!