MQ merging (was: corrupt repository after following mq merging instructions)

[Jesse Glick]

Bryan O'Sullivan wrote:
> the workflow around trying a merge of a patch queue is complicated
> and fragile, to the extent that I simply don't use it, and update
> patches by hand.

So are there any plans to fix MQ to do true merges? For example, MQ 
could record the parent node ID and patch SHA-1 hash whenever a patch is 
pushed successfully (no rejects, or even no fuzz). If subsequently 
application of the same patch would fail with rejects (or fuzz), and the 
current parent is a descendant of the recorded good parent, then instead 
of trying to apply the patch to the new parent, internally apply the 
patch to the former good parent and invoke a three-way merge using Hg's 
normal machinery. I think this would make MQ much more useful and more 
like regular Hg usage. ('hg import' could probably benefit from the same 
enhancement, in case the patch was from 'hg export' and recorded a 
usable parent node ID.)