The priority of this has gone up for us substantially, as the idiom in DFDL v1.0 for a choice-by-dispatch to have a default case when the choiceDispatchKey doesn't match any choiceBranchKey is far too complex.
This appears to be what is needed:
<!-- just a shorthand so we can discriminate true on one DFDL line -->
<group name="discriminateTrue">
<sequence>
<annontation><appinfo ...>
<dfdl:discriminator>{ fn:true() }</dfdl:discriminator>
</appinfo></annotation>
</sequence>
</group>
....
<choice>
<choice dfdl:choiceDispatchKey='{ ... dispatch key expr... }'>
<sequence dfdl:choiceBranchKey="key1">
<group ref="p:discriminateTrue"/>
... parse the key1 branch ....
</sequence>
...
<sequence dfdl:choiceBranchKey="keyN">
<group ref="p:discriminateTrue"/>
... parse the keyN branch ....
</sequence>
</choice>
<!-- default case below if the dispatch key does not match anything -->
<element name="invalidKey" ....>
...
</element>
</choice>
.....
The need for a discriminator for the outer choice on every single branch of the inner choice is tedious at best. This is needed because if you don't have this discriminator, then the backtrack will choose the other choice alternative for any failure of
a branch, not just no branch key match, and one will get a diagnostic that the branch key was not found when it was found but the subsequent parse failed.
This structure is too complex for DFDL schema authors to get right. A way to specify an additional choice branch which is the default to be used on no-match to the other branch keys, is needed.
An explicit dfdlx:choiceBranchDefault="yes" is one way to do this.
A final choice branch with no dfdl:choiceBranchKey would also work but seems too error-prone for my taste. A default case should be explicitly marked as such, even if we require it to be the last branch.