Handling 404s for DELETEs

[nnuss]

It seems sane to me to accept HTTP 404 as success for delete()s as the result is what we want.

I toyed with returning "0E0" instead but decided it probably benefitted few people.

Cheers,
-Nate

[xant]

one would argue that 404 is 'NOT FOUND' so no 'delete' actually took place ... are you sure this is the desired behaviour?

[nnuss]

I did consider that ; I was weighing returning 0E0 in similar fashion to DBI. But ReplicatedRest reduces the result down to a 0/1 anyway based on max_failures. I'm still open to that idea if you think direct SKV::S::Rest users would benefit from being able to detect this case vs. surprising them with handling the "true" zero pattern.

My thinking is that when calling delete() I care foremost about the behavior:
I expect to have the key/object removed/absent upon success - or - receive an error and have an expectation that the key may [read: probably does] persist.

Since delete is idempotent (multiple deletes achieve the same state a single delete) I think the client(s) benefit from not having to defend against 404-as-failure. Particularly since this type of "failure" is not easily distinguishable to the caller doing delete().