Why is the flag called --regenerate and not --update as in lake exe shake --update?

I considered that --- note that I defined two flags, --update and --regenerate.
The former only ignores new errors: this is suitable when you e.g. don't want to split a file which your change pushed over the limit. The latter regenerates all exceptions; in practice, this will incur more churn since the current number of lines is encoded in the file... so, to me only making the latter change occasionally (but not as a side effect) to intentional.

With shake, --update corresponds to my --regenerate, so is a super-set of this.
In terms of consistency, I think this is fine: the default action should be to call --update; not --regenerate. (If I get around to fixing the nolints bot, the file will be regenerated once a week.)

I could rename --update to --ignore if this is clearer; perhaps even with an error on using --update to clarify the distinction.

So --regenerate is the same as "delete nostyle" + --update, right? Maybe it doesn't need to exist as a separate flag

Yup. But then, why did we have update-style-exceptions.py? Or, put differently, how are we going to remove entries from the list? Automatically updating seems useful enough (to expert users) to make it low-friction; it just doesn't need to be the first operation users are exposed to.

Well you are worried about unnecessarily updating existing entries, not about correctly removing redundant entries, right? So --update could

• Recompute the file
• Add the lines to be added
• Remove the lines to be removed
• Not touch the lines that are merely modified (so long as not touching them is not causing a linting error)

That will be more implementation work: but indeed, that is possible.