This isn’t really a roadmap, it is more of a list of signs pointing in the right direction.

There are various tracks that can be followed to improve the future of GEGL in GIMP, many of these things, at least on the GEGL side are covered by bugs in bugzilla, properly git formatted patches adressing these issues are most welcome contributions. Doing things on this list would help speed up the arrival of the next and more powerful generation of GIMP.

Allow GEGL to run untrusted ops out-of process - ensuring that badly behaved ops will not bring GIMP/GEGL down. -Implement previews working on mipmap levels, allowing GEGL to do a preview at 12.5%, 25% or 50% of actual size - rendering the full size view in the background or on demand.

Allow GEGL to connect any property of ops as inputs/outputs in the graph, the infrastructure to do so is mostly already in place, but not properly exposed. For some uses, in particular for exposing more powerful meta-ops this would be very nice.

Improve the gparamspec vocabulary used to describe properties of of ops, providing more semantics for better auto-created user interfaces.

Writing GEGL ops is simpler than writing GIMP plug-ins, plug-ins can support higher bit depth in the future. Floating point code is easier to write, less code is needed. You can get auto generated UI. Writing more GIMP legacy plug-ins only increases the future porting work-load/leaves the new plug-in outdated, so please write new code as GEGL plug-ins. Even after a couple of gsocs, there is still unported such.

GIMP currently has many file loaders/savers that would be good to have in GEGL.

the GEGL website is generated as part of the GEGL build, patches to the documentation that fully integrates the changes also with the website are welcome. Documentation for the structure of GEGL and its internal architecture is also welcome.

Use GEGL in other places than GIMP

Commandline tools for doing batch processing, UIs for compositing or image manipulation, video effects rendering and more are all things that are possible to implement with GEGL, that would end up sharing their available ops among each other.

to provide incentive for yourself and others to work on the above points. GEGL is designed to be a generically usable API and does not cater especially for GIMP.

The pluggable GeglBuffers of GEGL permit other tiled buffer systems (krita, mypaint?, a future blender?, a plug-in for GIMP 1.x, OpenStreetMap ?) to efficiently reuse GEGL operations as plug-ins without unneccesary buffer copying.