The essential concept behind all this is to define a super simple patchscript AST, operations on it (application, optimization passes), and figure out how to stuff all existing patch formats into it.
How? Any exist for binary?? Probably (definitely) danger for copyright.
bytepatch supports this due to its simplicity. If you provide exact matches for all expected source strings in the patchscript, you can reverse just fine. (The various extensions like binrepping things will break stuff, but there are ways to support them too.)
Can I, for any patchscript, read just the patchscript and determine
- the minimum size of the input (
Maybefor weird formats like VCDIFF), and - the precise size of the output?
That would be cool. Gives way to efficient, safe patching.
I've focused on in-place patching so far for bytepatch. For that, minimum input size is easy. Output size depends on a secret implicit instruction in bytepatch, which is "copy to the end". With that, the output size is always precisely equal to the input size. Without it, the exact size is at the end of the final linearized patch.
The problem with developing this is... the patch formats out there are just poorly designed. VCDIFF is nice and popular, but the format is overly complex and it has a weird thing that needs consideration for converting to the unified AST. BPS is simple, but poorly documented, not well used, and has an irritating design choice that makes it slightly painful to parse.
For both of them, I'd probably do better by annotating a patch file and using that , rather than the specifications.