Google limits the number of item in the UnreadMails array to 30, but report the total number of unread e-mail correctly. This seems reasonable way to reduce the network overhead by such notification.
The result is that the size of UnreadMails maybe smaller or equal to UnreadMailCount. This behavior should be documented in the spec as being normal. Nothing in the spec currently deny this.
I think at some stage I made the spec wording explicitly say the opposite, but then reverted it... good thing I did, really :-)
Could you provide a patch please? I think the behaviour you've implemented in Gabble (as proposed here) will confuse people if we don't draw attention to it.
Oops, title was upside-down.
Updated my patch base on Simon comments on IRC:
<smcv> stormer: the tp:rationale there doesn't make sense to me... it's of the form "thing (rationale: rationale for an unrelated thing)"
<smcv> stormer: adding "This list MAY be shorter than <tp:member-ref>UnreadMailCount</> would indicate" would fix that
<smcv> stormer: (i.e. just before the rationale)
<smcv> stormer: thanks for catching up on this
<smcv> stormer: I'd prefer it if UnreadMailCount's docstring had proper HTML paragraphs, now that it's more than one paragraph
Will be in 0.19.4.