Re: [Gimp-user] Large XCF filesize

[Johannes Bauer <dfnsonfsduifb gmx de>]

On 14.06.19 10:49, Simon Budig wrote:

> > Well, I opened that and then saved it as a .xcf.bz2 file and it was
> > 10% of the original.
> >
> > Pretty reasonable to me?
>
> Well, yeah. But this is not what Johannes is talking about, he is
> questioning the performance of the builtin compression of XCF.

Yes, exactly. Especially the increased file size when reducing a layer
size puzzled me.

> One would probably have to dissect the XCF to figure out where all the
> file size is used for.

Well, I'm curious:
https://gitlab.gnome.org/GNOME/gimp/blob/master/devel-docs/xcf.txt

Maybe I'll have a look and check it out, just for fun.

> One key difference I vaguely recall is, that XCF compresses the image
> on the tile-by-tile level (i.e. small square chunks of image data). Each
> Tile is compressed individually, making it basically impossible to
> benefit from global similiarities.

Very interesting. I would have thought that many layers in XCF contain
usually many tiles that are exactly identical (e.g., 64x64 pixel that
are completely transparent or for layer masks similar blocks that are
entirely black or white) and assumed that XCF would have a means of
"referencing" those, if it makes sense. Apparently that's not the case.

> On the other hand this makes it possible to access specific parts of
> image data and adress certain regions of interest without having to
> decompress all of the image. But it certainly hurts the compression
> ratio.
>
> If that is enough to explain the difference I don't know.

True, all ups and downs. I'm mainly interested in the why.

On the other hand, for practical applications, I had already mentioned
that I gzipped my XCF and it reduced the size significantly. What I
learned since my last post is that GIMP can open .xcf.gz files (and
probably other compressions as well) natively without the need to gunzip
them first. That's a nice feature and I think one that I'll use.

All the best,
Joe