bug#12827: [2.0.6] web client: fails to parse 404 header

[Ludovic Courtès]

Hi,

Daniel Hartwig <address@hidden> skribis:

> On 10 November 2012 04:52, Ludovic Courtès <address@hidden> wrote:
>> Anyway, I think it’s fine if the documentation makes it clear whether
>> functions expect absolute or relative URIs.  WDYT?
>
> Yes.  With the new predicates, it should be clear enough to use the
> (pseudo-)type names in the usual scheme-doc way:
>
>   -- Scheme Procedure: uri-resolve base-uri uri-reference
>
> and not need to repeat too much in the prose.  Of course, doing so
> when appropriate.  I'll try to draft something sensible.

Yes.

>>> The build-uri validation works on the values before the <uri> object
>>> is constructed, so I was just thinking of a separate build method with
>>> different, less strict validation.
>>>
>>> We just have to think of <uri> and uri? as guile implementation
>>> details, not RFC.  Another option, is to rename <uri> to
>>> <uri-reference>.  Then uri? can mean the same as absolute-uri? (as per
>>> the RFC).
>>
>> Out current URI objects are actually absolute URI references, right?  In
>> that case, we’ll indeed have to make ‘uri?’ synonymous with
>> ‘absolute-uri?’, for backward compatibility.
>
> More-or-less, the only exception being when validation is disabled:
>
> scheme@(guile-user)> (uri? (build-uri #f #:path "foo" #:validate? #f))
> $1 = #t
>
> that object has no scheme, and is not an absolute-uri.  This is a bit
> of an edge case.

Yes, but when the user sets #:validate? to #f, then they take the
responsibility for anything that will happen.  IOW, #:validate? #f
allows users to forge broken URI objects, but that’s part of the
contract anyway.

> The current documentation only defines a URI as an absolute-uri and
> does not talk about anything else.  Most functions (uri->string, etc.)
> will not work when passed something without a scheme.  So I think your
> suggestion is ok as any users of the current API will most certainly
> be using only absolute-uri's.

Good.  So that means that URI refs can be added without introducing any
incompatibility.

Thanks,
Ludo’.