Re: [art] Against BCP 190

[Rob Sayre <[e-mail]>]

On Mon, Jul 22, 2019 at 10:54 PM Rob Sayre <[e-mail]> wrote:

> On Fri, Jul 12, 2019 at 9:46 AM Rob Stradling <[e-mail]> wrote:
>
>> ...since all modern
>> web servers can trivially rewrite paths to query components and back
>> again.
>>
>
> I don't think this assertion is correct. Query components don't reflect a
> hierarchy, they are unordered, and can be repeated. They can be rewritten
> to and from path segments, but not in a way that preserves their semantics.
>

Trying to be more constructive, I think the issue is that it's true that
you can write a seemingly-simple API that treats an entire host or path
below a certain level as opaque.

Enacting "simple" rules of that sort can cause problems as different HTTP
APIs coexist under one host, each with their own reserved suffixes, paths,
or query parameters. At the very least, you've prevented the host from
delegating its path space (say, for usernames). At worst, you've forced it
into using different hostnames, and taking on all of the associated DNS
latency and confidentiality issues.

It doesn't seem like writing a BCP190-compliant API is complex, but the
evidence does show that it's not the first avenue people think of.

thanks,
Rob