"SHOULD" vs "should"

[jernst]

This is a bit nitpicky, but "SHOULD" is part of the controlled vocabulary for the spec, and thus spelled in uppercase.

There are many places in the spec where the lower-case "should" is used instead.

Probably a distinction without an intended difference, but confusing.

Example from section 3 where the two are used right next to each other.

Servers SHOULD validate the content they receive to avoid content spoofing attacks. (A server should do something at least as robust as checking that the object appears as received at its origin, but mechanisms such as checking signatures would be better if available).

A similar comment can be made about "shall" and perhaps other terms like it.

[nightpool]

When used in uppercase, SHOULD creates a new conformance requirement for servers to follow. When used in lowercase, it's just a normal english word and should be interpreted as such. For example, in the section you quote, the parenthetical is simply elaborating on the "SHOULD validate" requirement, not creating a brand new requirement on its own)

[TallTed]

Widespread confusion persists about how to interpret the lowercase "should", even with the RFC 2119 and RFC 8174 citations early in the spec.

Contrary to what is found in those RFCs, best practice today is to avoid use of the lowercase "should", etc., and to use different wording to get the intended meaning across, e.g., to change A server should do something... to A server is expected to do something....

[evanp]

All right. Thanks for bringing this up. I think the next step is to identify "should" lowercase and determine if they are intended to be normative and ought to be "SHOULD" or if they are non-normative and ought to be converted to another phrased ("are expected to", etc.).

For each of these, I'll create an ERRATA entry.

I'm counting about 38 uses of "should". I think it's OK to exempt the ones in non-normative sections.

I see about 5 uses of "must", and about 40 uses of "may".

This seems like a big task. At worst, we're looking at about 83 ERRATA.md entries for text.

[TallTed]

I think it's OK to exempt the [uses of "should"] in non-normative sections.

It's OK, but I recommend not. Again, it is common for readers to read the lowercase "should" as the uppercase "SHOULD", even when encountered in non-normative sections. Often enough, it leads to readers thinking those sections (or at least, sub-sections) were erroneously labeled as non-normative.

This seems like a big task.

Yeah, it can be painful to make this transition. Fortunately, it's not so painful to maintain once existing docs are converted.

[nightpool]

I could see the benefit of addressing this in the editor's draft if we're working towards a new TR update, but I don't think it's worth errata, they would just be too numerous to be helpful.

[evanp]

Adding 83 errata for this is a lot. I am wiling to come back to this and find the most confusing, but I wonder if the best next step is to address this in a future version. I'm going to leave this ticket open, but I probably won't generate all the errata, at least for a while.