Consider to use more formal and modern intermediate representation

[XVilka]

To be able to use something like Falcon RE IL (based in turn on RREIL and Binary Ninja's IL): https://github.com/falconre/falcon Or even grab some ideas from MLIR: https://github.com/tensorflow/mlir/blob/master/g3doc/LangRef.md

My proposal is to define two forms of it:

• bitstream (like for LLVM bytecode)
• textual representation

It should support also:

• [ ] Floating point operations
• [ ] Vector operations (SIMD)
• [ ] Parallel execution
• [ ] Embedding annotations (e.g. types information, references, inheritance, etc).

See also Radeco's IL: https://github.com/radareorg/radeco-lib/blob/master/src/middle/ir.rs + https://github.com/radareorg/radeco-lib/tree/master/src/middle/ir_reader

https://github.com/radare/radare2/issues/11421 https://github.com/radare/radare2/issues/4799 https://github.com/radare/radare2/issues/7437 https://github.com/radare/radare2/issues/6093 https://github.com/radare/radare2/issues/4327

It is even possible to make an intermediate ESIL->R2IL (this hypothetical new IL) as a temporary solution, and move uplifters (architectures) to the new infrastructure one by one.

[XVilka]

Good example on how good IR/IL boosts decompiler and extension development https://twitter.com/josh_watson/status/1127364618429452288?s=19