Nope!

In Rule 59 the S? is defined by:

White Space
  [3] S ::=  (#x20 | #x9 | #xD | #xA)+
[download]

So (  toto  |  tata  ) is just as valid as (  toto  | tata  ). The problem is that expat and XML::Parser report _everything_ in the document, including non-significant spaces. Now Expat is required by the spec to do so. XML::Parser could probably choose to normalize those whitespaces but does not, and XML::DOM definitely should normalize them but does not. The SAX-2 extension for declaration for example normalizes declarations by removing all spaces around tokens.

The bottom-line is that this is a feature of XML::DOM that could easily pass for a bug (and should be fixed as soon as somebody provides a patch, I certainly don't want to get involved in XML::DOM but line 1970 of the DOM.pm in libxml-enno-1.04pre3 seems like the place to insert the normalization for attribute declarations, then on to element content models before line 2343).

As a side note, the best way to read the XML specification is probably to go to The Annotated XML Specification for the spec and Tim Bray's comments on xml.com

Update: Crap! My mistake (as usual :--(, XML::DOM (or maybe XML::Parser) does normalize attribute declarations. Actually that's why the test fails. It is fixed in libxml-enno-1.04pre though. The only remaining problem is that 1.04 does not pass the test on 5.6.0