From Frank's 1.0.31 announcement:
Extensions to the test fixtures for previewCitationCluster(),
which revealed shortcomings in the name state restoration
code.
Properly restore name state in previewCitationCluster().
Roll back previewCitationClusterStatic() command that
was introduced in the last (unannounced) 1.0.30 release.
Discussion and more careful thought showed that it is not
safe and not generally a good idea to make blind assumptions
about document state.
From Frank's 1.0.33 announcement:
In preview mode, avoid tampering with the registry data,
of items in the citation to be previewed that already
exist in the registry. Fixes errors in subsequent
rendering of year suffixes.
Suppress redundant return value for current cite, when it is
tainted by an update by processCitationCluster().
From Frank's 1.0.34 announcement:
Include year suffix in clone of disambiguation
configuration. Failure to include this parameter
appears to have been behind some intermittent
glitches in preview handling of year suffixed
citations.
Avoid processor crash when unregistered citation
objects are encountered in preview mode. The crash
points occur during tainting, which is irrelevant
in preview mode, so no harm is done; but I'm not
really sure why this would ever happen. Possibly
specific to the Zotero version used for testing.
Add a couple of paranoia operations to processCitationCluster():
Issue a bogus warning citation in the event of a processor
crash. This should come out for production.
Slice citationsPre and citationsPost when they enter
the processor. I've been exploring some instability
with large numbers of same-author/same-year cites in
Zotero, and thought that mallability of list input
might be part of the problem. No idea whether these
lists can or do change length or content during processing,
but just in case.
6d4d953859