To follow up on what Jacob Palme said ...
(a) Relative URL-s can be allowed if there is such a "Base:"
header.
Such a Base characteristic should not be required ;-).
I realize that there is language in RFC 1808 saying that relative
URLs are meaningless in the absense of a base. This is unfortunate.
There is a valid and useful interpretation for a set of HTML pages
conveyed by a MIME multipart where each part has a disposition
_relative_ URL and no base is declared anywhere in the message.
In this case, the receiving agent/client/transaction supplies a
disposition base for the root and, following 1808, the parts
inherit this and transform it to get absolute disposition URLs
for the parts.
You can look at this in either of two ways:
The parts inherit a base from the root part even
though the receiver and not the sender furnished
the base attribute for that root object.
The relative URLs identified for disposition are
valid for every recipient of the message, whereas
the base for their disposition is a variable
from recipient to recipient.
I suggest we adopt the second perspective -- that relative URLs
for a set of parts form a local coordinate frame, and that in
that local frame they are meaningful. The placement of this
local-coordinate patch in the global [absolute URL] coordinate
frame may be known or unknown, but the relative positions of the
set members are fixed by their respective relative URLs.
Al