[Kde-scm-interest] Have we arrived to a dead end?

[Oswald Buddenhagen]

On Wed, Feb 17, 2010 at 05:36:43PM +0000, John Tapsell wrote:
> I'm not all that convinced we even need shallow clones.
> 
> The Gnome guys found shallow clones only saved a small amount of space:
> 
> The first size, in MB, is the full checkout, and second number is a
> shallow clone of depth 1.
> 
> [...]
>
> Hardly seems worth the complication, let alone being a blocker wish,
> for a 10% saving.
>
these numbers were to be expected, as a shallow clone needs to contain
the head revisions of all files in the tip tree. the rest is stored in
pretty efficient xdeltas. and as most files mostly only gain code, the
inverse deltas are small. so a project must have a rather impressive
history to make its length contribute significantly to the repo size.
however, things are radically different for badly compressible binary
data, especially when files are often completely discarded - artwork.

narrow clones will be a significantly bigger gain, as *all* blobs for
the irrelevant files are omitted, including the big tip ones, obviously.

and just in case it's not obvious: one consequence of the delta
compression is that any file's bare presence at any point in time has a
significantly bigger impact than another file's long history. which is
one of the reasons why i want a clean splitting which enables moving
without leaving behind artificts. shallow/narrow clones would
theoretically compensate most of the problem, but i predict that in the
end everyone who is actually using the history will end up with more or
less complete clones sooner or later anyway.