Merge branch 'code-inference-fixes' into 'main'

configurable weights

Closes #546

See merge request rewriting/ddisasm!1148

CHANGELOG.md

@@ -14,6 +14,8 @@
 * Fix size access of LDR instruction in ARM64.
 * Extend value_reg analysis to support memory loads using a register with
   constant address.
+* Refactor the code inference point system. Decouple heuristics from their weights.
+  Heuristic weights can now be modified by providing user hints.
 
 
 # 1.8.0

doc/source/GENERAL/5-AdvancedUsage.md

@@ -15,3 +15,19 @@ disassembly.invalid 0x100 definitely_not_code
 ```
 will add a fact `invalid(0x100,"definitely_not_code")` to the Datalog database of the disassembly pass.
 The fields need to be separated by tabs '\t'.
+
+You can consult the Datalog API reference https://grammatech.github.io/ddisasm/APIRef.html
+for details on the internal Ddisasm predicates.
+
+### Changing heuristic weights with hints
+
+The code inference algorithm uses several heuristics to determine what is code and what is data.
+Each heuristic has a weight associated to it, i.e. how many points a block candidate gets for a given
+heuristic.
+
+Users can modify the heuristic weights by providing hints with the `user_heuristic_weight` predicate.
+For example, the following hint:
+```
+disassembly.user_heuristic_weight   overlaps with relocation simple -4
+```
+changes the weight of the "overlaps with relocation" heuristic to -4.

src/CMakeLists.txt

@@ -148,6 +148,7 @@ set(DATALOG_BASE_SOURCES
     datalog/boundary_value_analysis.dl
     datalog/code_inference.dl
     datalog/code_inference_postprocess.dl
+    datalog/code_inference_weights.dl
     datalog/cfg.dl
     datalog/data.dl
     datalog/data_access_analysis.dl
@@ -309,6 +310,7 @@ if(DDISASM_ARM_32)
       datalog/arch/arm_binaries.dl
       datalog/arch/arm32_binaries.dl
       datalog/arch/arm32_code_inference.dl
+      datalog/arch/arm32_code_inference_weights.dl
       datalog/arch/arm32_jump_tables.dl
       datalog/arch/arm32_symbolization.dl)