Re: [PATCH] Speedup recursive by flushing index only once for all entries

[Junio C Hamano <junkio@cox.net>]

"Alex Riesen" <raa.lkml@gmail.com> writes:

> Well, tried it now. Strangely enough - there is almost no speedup with
> Junio's patch on top of mine.

Mine is primarily meant to be a conceptual clean-up.  While it
does save unnecessary write-tree at the end, and it also saves
unnecessary write-cache/read-cache for the recursive part when
you have more than one merge base, I would not be surprised if
the effect of tons of unnecessary write-index, once per path
involved in the merge, which was what your patch removed, would
drawf anything else.

Since single merge base cases dominate in practice, you might
see improvements from the avoidance of the final write-tree but
probably would not see much benefit of write-cache/read-cache
avoidance unless your merge has many merge bases.