add array field generation support

pythonbpf/vmlinux_parser/ir_gen/ir_generation.py

@@ -1,8 +1,10 @@
+import ctypes
 import logging
 from ..dependency_handler import DependencyHandler
 from .debug_info_gen import debug_info_generation
 from ..dependency_node import DependencyNode
 import llvmlite.ir as ir
+from typing import Optional
 
 logger = logging.getLogger(__name__)
 
@@ -20,19 +22,50 @@ class IRGenerator:
         for struct in handler:
             self.struct_processor(struct)
 
-    def struct_processor(self, struct):
-        if struct.name not in self.generated:
-            print(f"IR generating for {struct.name}")
+    def struct_processor(self, struct, processing_stack=None):
+        # Initialize processing stack on first call
+        if processing_stack is None:
+            processing_stack = set()
+
+        # If already generated, skip
+        if struct.name in self.generated:
+            return
+
+        # Detect circular dependency
+        if struct.name in processing_stack:
+            logger.info(f"Circular dependency detected for {struct.name}, skipping recursive processing")
+            # For circular dependencies, we can either:
+            # 1. Use forward declarations (opaque pointers)
+            # 2. Mark as incomplete and process later
+            # 3. Generate a placeholder type
+            # Here we'll just skip and let it be processed in its own call
+            return
+
+        logger.info(f"IR generating for {struct.name}")
+
+        # Add to processing stack before processing dependencies
+        processing_stack.add(struct.name)
+
+        try:
+            # Process all dependencies first
             for dependency in struct.depends_on:
                 if dependency not in self.generated:
-                    dep_node_from_dependency = self.handler[dependency]
-                    self.struct_processor(dep_node_from_dependency)
-                    self.generated.append(dependency)
-            # actual processor logic here after assuming all dependencies are resolved
-            # this part cannot yet resolve circular dependencies. Gets stuck on an infinite loop during that.
+                    # Check if dependency exists in handler
+                    if dependency in self.handler.nodes:
+                        dep_node_from_dependency = self.handler[dependency]
+                        # Pass the processing_stack down to track circular refs
+                        self.struct_processor(dep_node_from_dependency, processing_stack)
+                    else:
+                        raise RuntimeError(f"Warning: Dependency {dependency} not found in handler")
+
+            # Actual processor logic here after dependencies are resolved
             self.gen_ir(struct)
             self.generated.append(struct.name)
 
+        finally:
+            # Remove from processing stack after we're done
+            processing_stack.discard(struct.name)
+
     def gen_ir(self, struct):
         # TODO: we add the btf_ama attribute by monkey patching in the end of compilation, but once llvmlite
         #  accepts our issue, we will resort to normal accessed attribute based attribute addition
@@ -41,19 +74,54 @@ class IRGenerator:
         field_index = 0
         for field_name, field in struct.fields.items():
             # does not take arrays and similar types into consideration yet.
-            field_co_re_name = self._struct_name_generator(struct, field, field_index)
-            field_index += 1
-            globvar = ir.GlobalVariable(
-                self.llvm_module, ir.IntType(64), name=field_co_re_name
-            )
-            globvar.linkage = "external"
-            globvar.set_metadata("llvm.preserve.access.index", debug_info)
-        print()
+            if field.ctype_complex_type is not None and issubclass(field.ctype_complex_type, ctypes.Array):
+                array_size = field.type_size
+                containing_type = field.containing_type
+                if containing_type.__module__ == ctypes.__name__:
+                    containing_type_size = ctypes.sizeof(containing_type)
+                    for i in range(0,array_size):
+                        field_co_re_name = self._struct_name_generator(struct, field, field_index, True, i, containing_type_size)
+                        globvar = ir.GlobalVariable(
+                            self.llvm_module, ir.IntType(64), name=field_co_re_name
+                        )
+                        globvar.linkage = "external"
+                        globvar.set_metadata("llvm.preserve.access.index", debug_info)
+                    field_index += 1
+            elif field.type_size is not None:
+                array_size = field.type_size
+                containing_type = field.containing_type
+                if containing_type.__module__ == "vmlinux":
+                    containing_type_size = self.handler[containing_type.__name__].current_offset
+                    for i in range(0,array_size):
+                        field_co_re_name = self._struct_name_generator(struct, field, field_index, True, i, containing_type_size)
+                        globvar = ir.GlobalVariable(
+                            self.llvm_module, ir.IntType(64), name=field_co_re_name
+                        )
+                        globvar.linkage = "external"
+                        globvar.set_metadata("llvm.preserve.access.index", debug_info)
+                    field_index += 1
+            else:
+                field_co_re_name = self._struct_name_generator(struct, field, field_index)
+                field_index += 1
+                globvar = ir.GlobalVariable(
+                    self.llvm_module, ir.IntType(64), name=field_co_re_name
+                )
+                globvar.linkage = "external"
+                globvar.set_metadata("llvm.preserve.access.index", debug_info)
 
     def _struct_name_generator(
-        self, struct: DependencyNode, field, field_index: int
+        self, struct: DependencyNode, field, field_index: int, is_indexed: bool=False, index: Optional[int]=None, containing_type_size: Optional[int]=None
     ) -> str:
-        if struct.name.startswith("struct_"):
+        if is_indexed:
+            name = (
+                    "llvm."
+                    + struct.name.removeprefix("struct_")
+                    + f":0:{field.offset + index*containing_type_size}"
+                    + "$"
+                    + f"0:{field_index}:{index}"
+            )
+            return name
+        elif struct.name.startswith("struct_"):
             name = (
                 "llvm."
                 + struct.name.removeprefix("struct_")