[MERGE/RFC] log -n0 -r 1..1.1.1 skips non-ancestrial revisions

[Ian Clatworthy]

Marius Kruger wrote:
>
> So I thought it might be reasonable to not delay_graph_generation
> if the user wants to see merge_revisions.
> This has a performance impact, but I wonder if it is acceptable either
> way with or without this patch.
>
> If performance is critical here we may be able to cheat a little for
> this special case
> and only do this if the user asked for a log up to a dotted revision
> (if end_rev.find('.'))
>
The only way to know whether delaying the graph generation or not is the
right choice is to benchmark. If you haven't already, try installing and
using bzr-usertest. I can provide some large data sets if you need them.

The reason for delaying graph generation is because "log --limit 10" is
a really popular use of log. I have that aliased to "top" for example.
If those 10 revisions don't have any merges, then rebuilding the whole
graph is a waste of time.

Ian C.