Re: [dmarc-ietf] Versioning and XML namespaces in aggregate reports (#33, #70)

[Matthäus Wander <[e-mail]>]

Alessandro Vesely wrote on 2021-05-10 18:29:
> On Mon 10/May/2021 17:28:20 +0200 Dave Crocker wrote:
>> If an new spec merely /adds/ to a previous spec, then the presence of
>> the new constructs is self-declaring.  The only requirement is to have
>> the base specification declare that unrecognized constructs are to be
>> ignored.  So, versioning adds the illusion of utility, but really only
>> adds unnecessary complexity.
> 
> 
> I think the format we'll end up with will be pretty compatible with the
> existing practice, meaning that existing report consumers that use a
> proper XML parser and ignore unknown tags can work unchanged.  I don't
> think any consumer parses reports "by hands".

Alright, introducing incompabilities is off the table and backward
compability is a must. This brings #51 into question, which may affect
backward compability.
https://trac.ietf.org/trac/dmarc/ticket/51


Regarding the existing top-level <version> below <feedback>:
Even if parsers don't require the version to function, it remains useful
for measuring the adoption of the different DMARC specifications (as
requested in #70). In fact, one implementation I looked at (parsedmarc)
uses it for only this purpose. A missing <version> is logged as "draft"
schema version.


Regarding the XML namespace declaration:
The XML schema serves not only as specification for developers, but can
be also used for automatic syntax checks of reports -- provided that the
namespace declaration is fixed. XSD validation is an immensely useful
tool for testing the output of report generators. It helped me to
discover two nasty bugs in an implementation, which appeared in 2 out of
~10k reports and would have gone unnoticed otherwise.
A version number within the schema is not necessary for this use case.

A different matter is whether automatic XSD validation on the report
consumer side is a supported use case. There is some value in it: two
lines of code suffice to perform input validation. However, the
validation is strict and does not allow for being liberal in what you
accept (might be handy for protocol police, though). Achieving upward
compatibility is not trivial, because there is no general "ignore all
unknown elements" statement in XSD. It is possible to define a <xs:any>
placeholder in the schema, but this element must be inserted explicitly
into each place where extensibility is desired. This would require
careful foresight in the schema design.

Regards,
Matt