Tighter check for ref type in resolver (this was broken but unnoticed) Skip table result classes in view part of resolver. No uninitialized warnings in resolver where it expects some misses (ugh!) Cleanup viewdeps test (and correct incorrect comparison) This should have not merged as is, but time is short to fix it properly. Hopefully next iter.
The capability system in its current form is insufficient - pending a redesign/rewrite. In the meantime remove mention of it from Changes and add an extra cap to disable the join-optimizer on request (prodded by timbunce :)
Some patchups to mitigate fallout from http://perl5.git.perl.org/perl.git/commitdiff/96d9b9cd Included several "time-bombs" set to go off when 5.13.8 is available (the "Contentious Changes freeze" release)
A regression slipped into 5.13.6 introduced by http://perl5.git.perl.org/perl.git/commit/304474c3 Fixed by: http://perl5.git.perl.org/perl.git/commit/cefd5c7c
Cleanup needed before merging extended rels (Storable chokes on the coderefs). Rewrite the test while we are here.
Audit the txn_do/scope_guard/txn_begin|txn_commit codepaths to make sure they can be interchangeably used and still just DTRT. Write a ton of tests to codify current behavior In the process revert d8a5a23 which introduces a regression (now poperly caught by tests)