bug#42382: 26.3; url-http handling of Location redirection headers containing whitespace

[Daniele Nicolodi]

On 16-07-2020 10:08, Robert Pluim wrote:
>>>>>> On Wed, 15 Jul 2020 14:40:36 -0600, Daniele Nicolodi 
>>>>>> <daniele@grinta.net> said:
> 
>     Daniele> In RFC 7231 the Location header is defined to carry a 
> URI-reference.
>     Daniele> According to RFC 3986 it should be percent-encoded and thus 
> should not
>     Daniele> contain spaces. However, there are HTTP server implementation 
> (notably
>     Daniele> nginx) that do not do that. While this is a bug in those HTTP 
> server
>     Daniele> implementations, I think Emacs should follow what most other 
> HTTP client
>     Daniele> implementatios (all the ones I tested) and use the content of the
>     Daniele> Location header unmodified. Stripping of angle bracket quotes is
>     Daniele> unnecessary as they are not valid according to the RFCs.
> 
> Nor is embedded whitespace in the URI :-)

I don't understand this remark. Truncating at the first whitespace
character (current behavior) is a valid arbitrary decision for an
RFC-invalid URI-reference value. However, it is different from what all
other HTTP clients implement and it results in practical problems.

> Are you sure this won't break anything? ie are you sure there are 0
> server implementations out there that send angle brackets?

I don't see any reason why there should be angle brackets around the
value of a Location header and the current code or changelog or commit
messages does not provide a justification or a case where these have
been encountered. No other HTTP client I looked at does something like
this. I think there are many HTTP client implementations out there that
are more widely used and tested for interoperability than url-http.

> Iʼd be conservative, and just replace the truncation on whitespace
> with percent-encoding of said whitespace.

Why is percent-encoding better? The URI-reference value should not be
interpreted in any way, simply passed along. Again, all other HTTP
clients I looked at do not do this, or other manipulation of the header.

>     Daniele> Also, accordingly to the RFCs, the location header may contain a
>     Daniele> relative location. Thus the comment that suggest that such a 
> response is
>     Daniele> a bug in the server should be reworded.
> 
>     Daniele> The attached patches implement the proposed changes.
> 
> The second patch is small enough that I think you can combine the two.

They are divided to provide justification for the changes in the commit
messages.

Thank you.

Daniele