[tech] to postel or not to postel, that is the question

On Sat, Dec 26, 2020 at 6:25 PM Petite Abeille <petite.abeille at gmail.com>
wrote:

Looking at the various implementations of the URL/URI/IRI stack in
> mainstream libraries, one has to face Jon Postel's robustness ghost ?.
>

In fairness to Jon, the context makes it clear that he did not mean what he
is commonly taken to mean.  Being liberal in what you expect does not mean
"be willing to accept anything and make sense of it".

for example: i am perfectly willing to read poetry or even prose written in
lower case only, even though it is harder for me and most people to read.
but i would never (except in this one email) write anything that was not
conventionally capitalized.  postel's law applies: i am liberal in what i
accept (conventional capitals or no capitals) but conservative in what i
send (conventional caps only).

What Postel's law does _not_ mean is that
IAMWILLINGTOACCEPTSCRIPTIOCONTINUAINENGLISH
EVENTHOUGHROMANINSCRIPTIONSWEREMOSTLY
like this.  (Imagine pages and pages of this style.  No.)  That's because
Postel's law is broken: I will not accept something well outside the
protocol of English writing, and I will not normally send something
s??? s? ??n?????p s? to read either.


>
> Exhibit ? 1: Google's own net/url/url.go ?
>
> // See RFC 3986. This package generally follows RFC 3986, except where
> // it deviates for compatibility reasons.
>
> Indeed. A very pragmatic approach to reality, taking the robustness
> principle to heart: pretty much anything goes. Such is life.
>

They have two conflicting rules to deal with: Postel's law and "Don't break
the customer's code".  There's simply no good answer to that.


> Exhibit ? 2: Sean Conner's own Parsing Expression Grammar ? ? ?
> The anti-thesis of Postel's law, as strict as the MUST imperative.


The apotheosis of Postel's law: accept everything, and only, what the
protocol says.

So here's an example closer to Gemini: should %-encoding use all lower
case?  Most code does, so it's the conservative thing to send.  But the
protocol is clear: a conformant implementation has to accept both.  But
because people often copy broken implementations from other programmers'
broken implementations, sometimes they don't even know that upper case is
acceptable.  Again, being conservative in what you send makes sure you can
interoperate with such broken implementations.

What does Gemini want, as a protocol? Leniency as provided by Exhibit ? 1?
> Or strictness as demonstrated by Exhibit ? 2?
>

Strictness, of course, give the above.


> It is my personal opinion that Postel's Law is the root of all evils as we
> know them to this day:

It isn't _money_ that's the root of all evil.  It's the greed for money.



> Protocol MUST be strict. This is why we have them in the first place. To
> clearly define what's legal or not, protocol wise.
>

Amen.  (But note that there are multiple versions of most protocols.)



John Cowan          http://vrici.lojban.org/~cowan        cowan at ccil.org
weirdo:    When is R7RS coming out?
Riastradh: As soon as the top is a beautiful golden brown and if you
stick a toothpick in it, the toothpick comes out dry.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.orbitalfox.eu/archives/gemini/attachments/20201226/0914
b422/attachment.htm>

---

Previous in thread (3 of 4): 🗣️ Petite Abeille (petite.abeille (a) gmail.com)

View entire thread.