Add StatementHandlerRegistry

Fix failing tests in tests/

.pre-commit-config.yaml

@@ -21,7 +21,7 @@ ci:
 repos:
 # Standard hooks
 - repo: https://github.com/pre-commit/pre-commit-hooks
-  rev: v4.6.0
+  rev: v6.0.0
   hooks:
   - id: check-added-large-files
   - id: check-case-conflict
@@ -36,7 +36,7 @@ repos:
   - id: trailing-whitespace
 
 - repo: https://github.com/astral-sh/ruff-pre-commit
-  rev: "v0.4.2"
+  rev: "v0.13.2"
   hooks:
     - id: ruff
       args: ["--fix", "--show-fixes"]
@@ -45,7 +45,7 @@ repos:
 
 # Checking static types
 - repo: https://github.com/pre-commit/mirrors-mypy
-  rev: "v1.10.0"
+  rev: "v1.18.2"
   hooks:
     - id: mypy
       exclude: ^(tests)|^(examples)

TODO.md

@@ -1,7 +1,6 @@
 ## Short term
 
 - Implement enough functionality to port the BCC tutorial examples in PythonBPF
-- Static Typing
 - Add all maps
 - XDP support in pylibbpf
 - ringbuf support

examples/clone-matplotlib.ipynb

@@ -12,7 +12,7 @@
     "from pythonbpf import bpf, map, section, bpfglobal, BPF\n",
     "from pythonbpf.helper import pid\n",
     "from pythonbpf.maps import HashMap\n",
-    "from pylibbpf import *\n",
+    "from pylibbpf import BpfMap\n",
     "from ctypes import c_void_p, c_int64, c_uint64, c_int32\n",
     "import matplotlib.pyplot as plt"
    ]

pythonbpf/binary_ops.py

@@ -9,6 +9,7 @@ logger: Logger = logging.getLogger(__name__)
 def recursive_dereferencer(var, builder):
     """dereference until primitive type comes out"""
     # TODO: Not worrying about stack overflow for now
+    logger.info(f"Dereferencing {var}, type is {var.type}")
     if isinstance(var.type, ir.PointerType):
         a = builder.load(var)
         return recursive_dereferencer(a, builder)
@@ -18,7 +19,7 @@ def recursive_dereferencer(var, builder):
         raise TypeError(f"Unsupported type for dereferencing: {var.type}")
 
 
-def get_operand_value(operand, module, builder, local_sym_tab):
+def get_operand_value(operand, builder, local_sym_tab):
     """Extract the value from an operand, handling variables and constants."""
     if isinstance(operand, ast.Name):
         if operand.id in local_sym_tab:
@@ -29,14 +30,14 @@ def get_operand_value(operand, module, builder, local_sym_tab):
             return ir.Constant(ir.IntType(64), operand.value)
         raise TypeError(f"Unsupported constant type: {type(operand.value)}")
     elif isinstance(operand, ast.BinOp):
-        return handle_binary_op_impl(operand, module, builder, local_sym_tab)
+        return handle_binary_op_impl(operand, builder, local_sym_tab)
     raise TypeError(f"Unsupported operand type: {type(operand)}")
 
 
-def handle_binary_op_impl(rval, module, builder, local_sym_tab):
+def handle_binary_op_impl(rval, builder, local_sym_tab):
     op = rval.op
-    left = get_operand_value(rval.left, module, builder, local_sym_tab)
+    left = get_operand_value(rval.left, builder, local_sym_tab)
-    right = get_operand_value(rval.right, module, builder, local_sym_tab)
+    right = get_operand_value(rval.right, builder, local_sym_tab)
     logger.info(f"left is {left}, right is {right}, op is {op}")
 
     # Map AST operation nodes to LLVM IR builder methods
@@ -61,6 +62,11 @@ def handle_binary_op_impl(rval, module, builder, local_sym_tab):
         raise SyntaxError("Unsupported binary operation")
 
 
-def handle_binary_op(rval, module, builder, var_name, local_sym_tab):
+def handle_binary_op(rval, builder, var_name, local_sym_tab):
-    result = handle_binary_op_impl(rval, module, builder, local_sym_tab)
+    result = handle_binary_op_impl(rval, builder, local_sym_tab)
-    builder.store(result, local_sym_tab[var_name].var)
+    if var_name and var_name in local_sym_tab:
+        logger.info(
+            f"Storing result {result} into variable {local_sym_tab[var_name].var}"
+        )
+        builder.store(result, local_sym_tab[var_name].var)
+    return result, result.type

pythonbpf/codegen.py

@@ -1,7 +1,7 @@
 import ast
 from llvmlite import ir
 from .license_pass import license_processing
-from .functions_pass import func_proc
+from .functions import func_proc
 from .maps import maps_proc
 from .structs import structs_proc
 from .globals_pass import globals_processing
@@ -48,7 +48,7 @@ def processor(source_code, filename, module):
     globals_processing(tree, module)
 
 
-def compile_to_ir(filename: str, output: str, loglevel=logging.WARNING):
+def compile_to_ir(filename: str, output: str, loglevel=logging.INFO):
     logging.basicConfig(
         level=loglevel, format="%(asctime)s [%(levelname)s] %(name)s: %(message)s"
     )
@@ -121,7 +121,7 @@ def compile_to_ir(filename: str, output: str, loglevel=logging.WARNING):
     return output
 
 
-def compile(loglevel=logging.WARNING) -> bool:
+def compile(loglevel=logging.INFO) -> bool:
     # Look one level up the stack to the caller of this function
     caller_frame = inspect.stack()[1]
     caller_file = Path(caller_frame.filename).resolve()
@@ -154,7 +154,7 @@ def compile(loglevel=logging.WARNING) -> bool:
     return success
 
 
-def BPF(loglevel=logging.WARNING) -> BpfProgram:
+def BPF(loglevel=logging.INFO) -> BpfProgram:
     caller_frame = inspect.stack()[1]
     src = inspect.getsource(caller_frame.frame)
     with tempfile.NamedTemporaryFile(

pythonbpf/functions/__init__.py

@@ -0,0 +1,3 @@
+from .functions_pass import func_proc
+
+__all__ = ["func_proc"]

pythonbpf/functions/func_registry_handlers.py

@@ -0,0 +1,22 @@
+from typing import Dict
+
+
+class StatementHandlerRegistry:
+    """Registry for statement handlers."""
+
+    _handlers: Dict = {}
+
+    @classmethod
+    def register(cls, stmt_type):
+        """Register a handler for a specific statement type."""
+
+        def decorator(handler):
+            cls._handlers[stmt_type] = handler
+            return handler
+
+        return decorator
+
+    @classmethod
+    def get_handler(cls, stmt_type):
+        """Get the handler for a specific statement type."""
+        return cls._handlers.get(stmt_type, None)

pythonbpf/functions/functions_pass.py

@@ -4,10 +4,10 @@ import logging
 from typing import Any
 from dataclasses import dataclass
 
-from .helper import HelperHandlerRegistry, handle_helper_call
+from pythonbpf.helper import HelperHandlerRegistry, handle_helper_call
-from .type_deducer import ctypes_to_ir
+from pythonbpf.type_deducer import ctypes_to_ir
-from .binary_ops import handle_binary_op
+from pythonbpf.binary_ops import handle_binary_op
-from .expr_pass import eval_expr, handle_expr
+from pythonbpf.expr_pass import eval_expr, handle_expr
 
 logger = logging.getLogger(__name__)
 
@@ -146,8 +146,7 @@ def handle_assign(
                     local_sym_tab[var_name].var,
                 )
                 logger.info(
-                    f"Assigned {call_type} constant "
+                    f"Assigned {call_type} constant {rval.args[0].value} to {var_name}"
-                    f"{rval.args[0].value} to {var_name}"
                 )
             elif HelperHandlerRegistry.has_handler(call_type):
                 # var = builder.alloca(ir.IntType(64), name=var_name)
@@ -233,7 +232,7 @@ def handle_assign(
         else:
             logger.info("Unsupported assignment call function type")
     elif isinstance(rval, ast.BinOp):
-        handle_binary_op(rval, module, builder, var_name, local_sym_tab)
+        handle_binary_op(rval, builder, var_name, local_sym_tab)
     else:
         logger.info("Unsupported assignment value type")
 
@@ -385,24 +384,48 @@ def process_stmt(
         )
     elif isinstance(stmt, ast.Return):
         if stmt.value is None:
-            builder.ret(ir.Constant(ir.IntType(32), 0))
+            builder.ret(ir.Constant(ir.IntType(64), 0))
             did_return = True
         elif (
             isinstance(stmt.value, ast.Call)
             and isinstance(stmt.value.func, ast.Name)
             and len(stmt.value.args) == 1
-            and isinstance(stmt.value.args[0], ast.Constant)
-            and isinstance(stmt.value.args[0].value, int)
         ):
-            call_type = stmt.value.func.id
+            if isinstance(stmt.value.args[0], ast.Constant) and isinstance(
-            if ctypes_to_ir(call_type) != ret_type:
+                stmt.value.args[0].value, int
-                raise ValueError(
+            ):
-                    "Return type mismatch: expected"
+                call_type = stmt.value.func.id
-                    f"{ctypes_to_ir(call_type)}, got {call_type}"
+                if ctypes_to_ir(call_type) != ret_type:
-                )
+                    raise ValueError(
-            else:
+                        "Return type mismatch: expected"
-                builder.ret(ir.Constant(ret_type, stmt.value.args[0].value))
+                        f"{ctypes_to_ir(call_type)}, got {call_type}"
+                    )
+                else:
+                    builder.ret(ir.Constant(ret_type, stmt.value.args[0].value))
+                    did_return = True
+            elif isinstance(stmt.value.args[0], ast.BinOp):
+                # TODO: Should be routed through eval_expr
+                val = handle_binary_op(stmt.value.args[0], builder, None, local_sym_tab)
+                if val is None:
+                    raise ValueError("Failed to evaluate return expression")
+                if val[1] != ret_type:
+                    raise ValueError(
+                        f"Return type mismatch: expected {ret_type}, got {val[1]}"
+                    )
+                builder.ret(val[0])
                 did_return = True
+            elif isinstance(stmt.value.args[0], ast.Name):
+                if stmt.value.args[0].id in local_sym_tab:
+                    var = local_sym_tab[stmt.value.args[0].id].var
+                    val = builder.load(var)
+                    if val.type != ret_type:
+                        raise ValueError(
+                            f"Return type mismatch: expected {ret_type}, got {val.type}"
+                        )
+                    builder.ret(val)
+                    did_return = True
+                else:
+                    raise ValueError("Failed to evaluate return expression")
         elif isinstance(stmt.value, ast.Name):
             if stmt.value.id == "XDP_PASS":
                 builder.ret(ir.Constant(ret_type, 2))
@@ -455,6 +478,9 @@ def allocate_mem(
                 continue
             var_name = target.id
             rval = stmt.value
+            if var_name in local_sym_tab:
+                logger.info(f"Variable {var_name} already allocated")
+                continue
             if isinstance(rval, ast.Call):
                 if isinstance(rval.func, ast.Name):
                     call_type = rval.func.id
@@ -483,8 +509,7 @@ def allocate_mem(
                         var = builder.alloca(ir_type, name=var_name)
                         has_metadata = True
                         logger.info(
-                            f"Pre-allocated variable {var_name} "
+                            f"Pre-allocated variable {var_name} for struct {call_type}"
-                            f"for struct {call_type}"
                         )
                 elif isinstance(rval.func, ast.Attribute):
                     ir_type = ir.PointerType(ir.IntType(64))
@@ -568,7 +593,7 @@ def process_func_body(
         )
 
     if not did_return:
-        builder.ret(ir.Constant(ir.IntType(32), 0))
+        builder.ret(ir.Constant(ir.IntType(64), 0))
 
 
 def process_bpf_chunk(func_node, module, return_type, map_sym_tab, structs_sym_tab):

pythonbpf/helper/bpf_helper_handler.py

@@ -62,7 +62,7 @@ def bpf_map_lookup_elem_emitter(
     """
     if not call.args or len(call.args) != 1:
         raise ValueError(
-            "Map lookup expects exactly one argument (key), got " f"{len(call.args)}"
+            f"Map lookup expects exactly one argument (key), got {len(call.args)}"
         )
     key_ptr = get_or_create_ptr_from_arg(call.args[0], builder, local_sym_tab)
     map_void_ptr = builder.bitcast(map_ptr, ir.PointerType())
@@ -145,8 +145,7 @@ def bpf_map_update_elem_emitter(
     """
     if not call.args or len(call.args) < 2 or len(call.args) > 3:
         raise ValueError(
-            "Map update expects 2 or 3 args (key, value, flags), "
+            f"Map update expects 2 or 3 args (key, value, flags), got {len(call.args)}"
-            f"got {len(call.args)}"
         )
 
     key_arg = call.args[0]
@@ -196,7 +195,7 @@ def bpf_map_delete_elem_emitter(
     """
     if not call.args or len(call.args) != 1:
         raise ValueError(
-            "Map delete expects exactly one argument (key), got " f"{len(call.args)}"
+            f"Map delete expects exactly one argument (key), got {len(call.args)}"
         )
     key_ptr = get_or_create_ptr_from_arg(call.args[0], builder, local_sym_tab)
     map_void_ptr = builder.bitcast(map_ptr, ir.PointerType())
@@ -255,7 +254,7 @@ def bpf_perf_event_output_handler(
 ):
     if len(call.args) != 1:
         raise ValueError(
-            "Perf event output expects exactly one argument, " f"got {len(call.args)}"
+            f"Perf event output expects exactly one argument, got {len(call.args)}"
         )
     data_arg = call.args[0]
     ctx_ptr = func.args[0]  # First argument to the function is ctx

pythonbpf/helper/helper_utils.py

@@ -270,7 +270,7 @@ def _prepare_expr_args(expr, func, module, builder, local_sym_tab, struct_sym_ta
                 val = builder.sext(val, ir.IntType(64))
         else:
             logger.warning(
-                "Only int and ptr supported in bpf_printk args. " "Others default to 0."
+                "Only int and ptr supported in bpf_printk args. Others default to 0."
             )
             val = ir.Constant(ir.IntType(64), 0)
         return val

pythonbpf/maps/maps_pass.py

@@ -278,9 +278,7 @@ def process_bpf_map(func_node, module):
         if handler:
             return handler(map_name, rval, module)
         else:
-            logger.warning(
+            logger.warning(f"Unknown map type {rval.func.id}, defaulting to HashMap")
-                f"Unknown map type " f"{rval.func.id}, defaulting to HashMap"
-            )
             return process_hash_map(map_name, rval, module)
     else:
         raise ValueError("Function under @map must return a map")

tests/failing_tests/direct_assign.py

@@ -4,6 +4,18 @@ from pythonbpf.maps import HashMap
 
 from ctypes import c_void_p, c_int64
 
+# NOTE: I have decided to not fix this example for now.
+# The issue is in line 31, where we are passing an expression.
+# The update helper expects a pointer type. But the problem is
+# that we must allocate the space for said pointer in the first
+# basic block. As that usage is in a different basic block, we
+# are unable to cast the expression to a pointer type. (as we never
+# allocated space for it).
+# Shall we change our space allocation logic? That allows users to
+# spam the same helper with the same args, and still run out of
+# stack space. So we consider this usage invalid for now.
+# Might fix it later.
+
 
 @bpf
 @map
@@ -14,12 +26,12 @@ def count() -> HashMap:
 @bpf
 @section("xdp")
 def hello_world(ctx: c_void_p) -> c_int64:
-    prev = count().lookup(0)
+    prev = count.lookup(0)
     if prev:
-        count().update(0, prev + 1)
+        count.update(0, prev + 1)
         return XDP_PASS
     else:
-        count().update(0, 1)
+        count.update(0, 1)
 
     return XDP_PASS
 

tests/failing_tests/named_arg.py

@@ -0,0 +1,40 @@
+from pythonbpf import bpf, map, section, bpfglobal, compile
+from pythonbpf.helper import XDP_PASS
+from pythonbpf.maps import HashMap
+
+from ctypes import c_void_p, c_int64
+
+# NOTE: This example exposes the problems with our typing system.
+# We can't do steps on line 25 and 27.
+# prev is of type i64**. For prev + 1, we deref it down to i64
+# To assign it back to prev, we need to go back to i64**.
+# We cannot allocate space for the intermediate type now.
+# We probably need to track the ref/deref chain for each variable.
+
+@bpf
+@map
+def count() -> HashMap:
+    return HashMap(key=c_int64, value=c_int64, max_entries=1)
+
+
+@bpf
+@section("xdp")
+def hello_world(ctx: c_void_p) -> c_int64:
+    prev = count.lookup(0)
+    if prev:
+        prev = prev + 1
+        count.update(0, prev)
+        return XDP_PASS
+    else:
+        count.update(0, 1)
+
+    return XDP_PASS
+
+
+@bpf
+@bpfglobal
+def LICENSE() -> str:
+    return "GPL"
+
+
+compile()

tests/passing_tests/var_rval.py

@@ -0,0 +1,20 @@
+import logging
+
+from pythonbpf import compile, bpf, section, bpfglobal
+from ctypes import c_void_p, c_int64
+
+
+@bpf
+@section("sometag1")
+def sometag(ctx: c_void_p) -> c_int64:
+    a = 1 - 1
+    return c_int64(a)
+
+
+@bpf
+@bpfglobal
+def LICENSE() -> str:
+    return "GPL"
+
+
+compile(loglevel=logging.INFO)