[Monotone-devel] Re: Annoying behaviour with merge_into_dir

[Bruce Stephens]

Nathaniel Smith <address@hidden> writes:

[...]

> What about it is infeasible?  With long-lived branches it can create
> more data transfer overhead, which is a little annoying, but nothing
> specific to merge_into_dir...

Ah, OK.  My mistake is in making this merged_into branch and then
continuing to propagate the regular changes in to it.  (My reason for
doing that is just that that's how I was converting from subversion.)

So if I abandon the branch that I called "main" in the example: the
one that has most common changes, and simply make those changes to
"all", then most revisions will just have one ancestor, and
everything's OK.  That requires a rearrangement of my Tailor set up,
but nothing impossible, I think.

That leaves cases where I need to propagate (because one of the
subbranches changes) rather expensive (1.6M revision uncompressed, by
the looks of it), but it keeps most changes OK.

Presuming the expensive changes are O(0.5M), I'll have to look and see
how common they in fact happen, and see how costly that'll be.  (Most
of the code doesn't change much, but part of the reason we commit
these is to commit our build rules and things, and they change more
often than the basic code.)