Empty S3 bucket during build and deploy action

[thomasheritage]

Currently when a new "build and deploy" takes place the new files are uploaded to the S3 bucket but any old files are not deleted from the S3 bucket.

See: https://github.com/SMPTE/html-pub/blob/main/scripts/build.mjs#L130

This has caused confusion when a file (e.g. a non-prose element of a Standard) is renamed because the old file is still available on the S3 bucket after the rename. If some/all links (e.g. in the html) have not yet been updated then it can still be retrieved and the "broken" links aren't noticed.

[palemieux]

Presumably this would only apply to artifacts associated with branches, e.g., main, and not individual commits, right?

In other words, the bucket subtree should be cleared only at:

const pubLinks = await s3Upload(buildPaths, branchName, generatedFiles);

The idea is that the branch upload represents the latest (and thus should not include old files) but individual commits and releases should be self-contained.

[thomasheritage]

When I said "empty S3 bucket" that was a bit imprecise.

My understanding is that s3Upload(buildPaths, TARGET, generatedFiles) uploads the files in buildPaths.pubDirPath to a "folder" in an S3 bucket where that "folder" has the path NAME/TARGET/ (where NAME is, at least sometimes, the repo name).

I think we want the "folder" NAME/TARGET/ to contain the same set of files as buildPaths.pubDirPath regardless of the value of TARGET? I think that would mean that all objects (if any) with a key beginning NAME/TARGET/ need to be deleted prior to uploading any files. (Or a more sophisticated sync algorithm could be used of course. I'm not sure how aws s3 sync . s3://amzn-s3-demo-bucket --delete works).