When to check entity WFness according to 4.3.2

[Jeff Rafter <lists@jeffrafter.com>]

Continuing along the same lines I have been going for the past week I 
have another question about entity details. Section 4.3.2 of the XML rec 
says:

"An internal general parsed entity is well-formed if its replacement 
text matches the production labeled content."

My question is: when should such a WFness check be performed.

(a) Immediately after the literal is parsed
(b) Immediately after the DTD has been parsed
(c) Only when it is referenced

I don't think (c) is the answer-- especially because of some of the 
points that were made in this thread [1]. So that leaves us with (a) and 
(b). The problem with (a) is that element declarations important  to the 
WF check of the literal value might not have occurred. Consider:

<!DOCTYPE doc [
<!ELEMENT doc (foo)>
<!ENTITY e "<foo id='This is not an id!'/>">
<!ELEMENT foo EMPTY>
<!ATTLIST foo id ID #IMPLIED>
]>
<doc>&e;</doc>

versus:

<!DOCTYPE doc [
<!ELEMENT doc (foo)>
<!ELEMENT foo EMPTY>
<!ATTLIST foo id ID #IMPLIED>
<!ENTITY e "<foo id='This is not an id!'/>">
]>
<doc>&e;</doc>

This leaves us with option (b)-- perform the WFness check of the literal 
value once the DTD has been parsed. The only hitch with this is test 
case valid-sa-86 from the xml test suite.

<!DOCTYPE doc [
<!ELEMENT doc (#PCDATA)>
<!ENTITY e "">
<!ENTITY e "<foo>">
]>
<doc>&e;</doc>

This test is supposed to be wellformed which would only be the case if 
we accepted option (c), or if overridden entity literals are discarded 
and we go with option (b). This makes sense when considering the 
distinction between the literal entity text and the replacement text. 
Section 4.3.2 refers only to checking the WFness the replacement text 
not the actual literal.

Is this the correct interpretation?

Thanks,
Jeff Rafter

[1] http://lists.xml.org/archives/xml-dev/200405/msg00292.html