Add compile to tests/failing_tests/conditionals/helper_cond.py

TODO.md

@@ -5,7 +5,6 @@
 - XDP support in pylibbpf
 - ringbuf support
 - recursive expression resolution
-- Add supoprt for BoolOp and short circuiting in conditions
 
 ## Long term
 

pythonbpf/expr/__init__.py

@@ -0,0 +1,4 @@
+from .expr_pass import eval_expr, handle_expr
+from .type_normalization import convert_to_bool
+
+__all__ = ["eval_expr", "handle_expr", "convert_to_bool"]

pythonbpf/expr/expr_pass.py

@@ -4,7 +4,8 @@ from logging import Logger
 import logging
 from typing import Dict
 
-from .type_deducer import ctypes_to_ir, is_ctypes
+from pythonbpf.type_deducer import ctypes_to_ir, is_ctypes
+from .type_normalization import convert_to_bool, handle_comparator
 
 logger: Logger = logging.getLogger(__name__)
 
@@ -43,7 +44,6 @@ def _handle_attribute_expr(
             var_ptr, var_type, var_metadata = local_sym_tab[var_name]
             logger.info(f"Loading attribute {attr_name} from variable {var_name}")
             logger.info(f"Variable type: {var_type}, Variable ptr: {var_ptr}")
-
             metadata = structs_sym_tab[var_metadata]
             if attr_name in metadata.fields:
                 gep = metadata.gep(builder, var_ptr, attr_name)
@@ -130,30 +130,6 @@ def _handle_ctypes_call(
     return val
 
 
-def _handle_comparator(builder, op, lhs, rhs):
-    """Handle comparison operations."""
-
-    # NOTE: For now assume same types
-
-    comparison_ops = {
-        ast.Eq: "==",
-        ast.NotEq: "!=",
-        ast.Lt: "<",
-        ast.LtE: "<=",
-        ast.Gt: ">",
-        ast.GtE: ">=",
-    }
-
-    if type(op) not in comparison_ops:
-        logger.error(f"Unsupported comparison operator: {type(op)}")
-        return None
-
-    predicate = comparison_ops[type(op)]
-    result = builder.icmp_signed(predicate, lhs, rhs)
-    logger.debug(f"Comparison result: {result}")
-    return result, ir.IntType(1)
-
-
 def _handle_compare(
     func, module, builder, cond, local_sym_tab, map_sym_tab, structs_sym_tab=None
 ):
@@ -187,17 +163,7 @@ def _handle_compare(
 
     lhs, _ = lhs
     rhs, _ = rhs
-    return _handle_comparator(builder, cond.ops[0], lhs, rhs)
+    return handle_comparator(func, builder, cond.ops[0], lhs, rhs)
-
-
-def convert_to_bool(builder, val):
-    if val.type == ir.IntType(1):
-        return val
-    if isinstance(val.type, ir.PointerType):
-        zero = ir.Constant(val.type, None)
-    else:
-        zero = ir.Constant(val.type, 0)
-    return builder.icmp_signed("!=", val, zero)
 
 
 def _handle_unary_op(
@@ -227,6 +193,136 @@ def _handle_unary_op(
     return result, ir.IntType(1)
 
 
+def _handle_and_op(func, builder, expr, local_sym_tab, map_sym_tab, structs_sym_tab):
+    """Handle `and` boolean operations."""
+
+    logger.debug(f"Handling 'and' operator with {len(expr.values)} operands")
+
+    merge_block = func.append_basic_block(name="and.merge")
+    false_block = func.append_basic_block(name="and.false")
+
+    incoming_values = []
+
+    for i, value in enumerate(expr.values):
+        is_last = i == len(expr.values) - 1
+
+        # Evaluate current operand
+        operand_result = eval_expr(
+            func, None, builder, value, local_sym_tab, map_sym_tab, structs_sym_tab
+        )
+        if operand_result is None:
+            logger.error(f"Failed to evaluate operand {i} in 'and' expression")
+            return None
+
+        operand_val, operand_type = operand_result
+
+        # Convert to boolean if needed
+        operand_bool = convert_to_bool(builder, operand_val)
+        current_block = builder.block
+
+        if is_last:
+            # Last operand: result is this value
+            builder.branch(merge_block)
+            incoming_values.append((operand_bool, current_block))
+        else:
+            # Not last: check if true, continue or short-circuit
+            next_check = func.append_basic_block(name=f"and.check_{i + 1}")
+            builder.cbranch(operand_bool, next_check, false_block)
+            builder.position_at_end(next_check)
+
+    # False block: short-circuit with false
+    builder.position_at_end(false_block)
+    builder.branch(merge_block)
+    false_value = ir.Constant(ir.IntType(1), 0)
+    incoming_values.append((false_value, false_block))
+
+    # Merge block: phi node
+    builder.position_at_end(merge_block)
+    phi = builder.phi(ir.IntType(1), name="and.result")
+    for val, block in incoming_values:
+        phi.add_incoming(val, block)
+
+    logger.debug(f"Generated 'and' with {len(incoming_values)} incoming values")
+    return phi, ir.IntType(1)
+
+
+def _handle_or_op(func, builder, expr, local_sym_tab, map_sym_tab, structs_sym_tab):
+    """Handle `or` boolean operations."""
+
+    logger.debug(f"Handling 'or' operator with {len(expr.values)} operands")
+
+    merge_block = func.append_basic_block(name="or.merge")
+    true_block = func.append_basic_block(name="or.true")
+
+    incoming_values = []
+
+    for i, value in enumerate(expr.values):
+        is_last = i == len(expr.values) - 1
+
+        # Evaluate current operand
+        operand_result = eval_expr(
+            func, None, builder, value, local_sym_tab, map_sym_tab, structs_sym_tab
+        )
+        if operand_result is None:
+            logger.error(f"Failed to evaluate operand {i} in 'or' expression")
+            return None
+
+        operand_val, operand_type = operand_result
+
+        # Convert to boolean if needed
+        operand_bool = convert_to_bool(builder, operand_val)
+        current_block = builder.block
+
+        if is_last:
+            # Last operand: result is this value
+            builder.branch(merge_block)
+            incoming_values.append((operand_bool, current_block))
+        else:
+            # Not last: check if false, continue or short-circuit
+            next_check = func.append_basic_block(name=f"or.check_{i + 1}")
+            builder.cbranch(operand_bool, true_block, next_check)
+            builder.position_at_end(next_check)
+
+    # True block: short-circuit with true
+    builder.position_at_end(true_block)
+    builder.branch(merge_block)
+    true_value = ir.Constant(ir.IntType(1), 1)
+    incoming_values.append((true_value, true_block))
+
+    # Merge block: phi node
+    builder.position_at_end(merge_block)
+    phi = builder.phi(ir.IntType(1), name="or.result")
+    for val, block in incoming_values:
+        phi.add_incoming(val, block)
+
+    logger.debug(f"Generated 'or' with {len(incoming_values)} incoming values")
+    return phi, ir.IntType(1)
+
+
+def _handle_boolean_op(
+    func,
+    module,
+    builder,
+    expr: ast.BoolOp,
+    local_sym_tab,
+    map_sym_tab,
+    structs_sym_tab=None,
+):
+    """Handle `and` and `or` boolean operations."""
+
+    if isinstance(expr.op, ast.And):
+        return _handle_and_op(
+            func, builder, expr, local_sym_tab, map_sym_tab, structs_sym_tab
+        )
+    elif isinstance(expr.op, ast.Or):
+        return _handle_or_op(
+            func, builder, expr, local_sym_tab, map_sym_tab, structs_sym_tab
+        )
+    else:
+        logger.error(f"Unsupported boolean operator: {type(expr.op).__name__}")
+        return None
+
+
 def eval_expr(
     func,
     module,
@@ -315,6 +411,10 @@ def eval_expr(
         return _handle_unary_op(
             func, module, builder, expr, local_sym_tab, map_sym_tab, structs_sym_tab
         )
+    elif isinstance(expr, ast.BoolOp):
+        return _handle_boolean_op(
+            func, module, builder, expr, local_sym_tab, map_sym_tab, structs_sym_tab
+        )
     logger.info("Unsupported expression evaluation")
     return None
 

pythonbpf/expr/type_normalization.py

@@ -0,0 +1,128 @@
+from llvmlite import ir
+import logging
+import ast
+
+logger = logging.getLogger(__name__)
+
+COMPARISON_OPS = {
+    ast.Eq: "==",
+    ast.NotEq: "!=",
+    ast.Lt: "<",
+    ast.LtE: "<=",
+    ast.Gt: ">",
+    ast.GtE: ">=",
+    ast.Is: "==",
+    ast.IsNot: "!=",
+}
+
+
+def _get_base_type_and_depth(ir_type):
+    """Get the base type for pointer types."""
+    cur_type = ir_type
+    depth = 0
+    while isinstance(cur_type, ir.PointerType):
+        depth += 1
+        cur_type = cur_type.pointee
+    return cur_type, depth
+
+
+def _deref_to_depth(func, builder, val, target_depth):
+    """Dereference a pointer to a certain depth."""
+
+    cur_val = val
+    cur_type = val.type
+
+    for depth in range(target_depth):
+        if not isinstance(val.type, ir.PointerType):
+            logger.error("Cannot dereference further, non-pointer type")
+            return None
+
+        # dereference with null check
+        pointee_type = cur_type.pointee
+        null_check_block = builder.block
+        not_null_block = func.append_basic_block(name=f"deref_not_null_{depth}")
+        merge_block = func.append_basic_block(name=f"deref_merge_{depth}")
+
+        null_ptr = ir.Constant(cur_type, None)
+        is_not_null = builder.icmp_signed("!=", cur_val, null_ptr)
+        logger.debug(f"Inserted null check for pointer at depth {depth}")
+
+        builder.cbranch(is_not_null, not_null_block, merge_block)
+
+        builder.position_at_end(not_null_block)
+        dereferenced_val = builder.load(cur_val)
+        logger.debug(f"Dereferenced to depth {depth - 1}, type: {pointee_type}")
+        builder.branch(merge_block)
+
+        builder.position_at_end(merge_block)
+        phi = builder.phi(pointee_type, name=f"deref_result_{depth}")
+
+        zero_value = (
+            ir.Constant(pointee_type, 0)
+            if isinstance(pointee_type, ir.IntType)
+            else ir.Constant(pointee_type, None)
+        )
+        phi.add_incoming(zero_value, null_check_block)
+
+        phi.add_incoming(dereferenced_val, not_null_block)
+
+        # Continue with phi result
+        cur_val = phi
+        cur_type = pointee_type
+    return cur_val
+
+
+def _normalize_types(func, builder, lhs, rhs):
+    """Normalize types for comparison."""
+
+    logger.info(f"Normalizing types: {lhs.type} vs {rhs.type}")
+    if isinstance(lhs.type, ir.IntType) and isinstance(rhs.type, ir.IntType):
+        if lhs.type.width < rhs.type.width:
+            lhs = builder.sext(lhs, rhs.type)
+        else:
+            rhs = builder.sext(rhs, lhs.type)
+        return lhs, rhs
+    elif not isinstance(lhs.type, ir.PointerType) and not isinstance(
+        rhs.type, ir.PointerType
+    ):
+        logger.error(f"Type mismatch: {lhs.type} vs {rhs.type}")
+        return None, None
+    else:
+        lhs_base, lhs_depth = _get_base_type_and_depth(lhs.type)
+        rhs_base, rhs_depth = _get_base_type_and_depth(rhs.type)
+        if lhs_base == rhs_base:
+            if lhs_depth < rhs_depth:
+                rhs = _deref_to_depth(func, builder, rhs, rhs_depth - lhs_depth)
+            elif rhs_depth < lhs_depth:
+                lhs = _deref_to_depth(func, builder, lhs, lhs_depth - rhs_depth)
+            return _normalize_types(func, builder, lhs, rhs)
+
+
+def convert_to_bool(builder, val):
+    """Convert a value to boolean."""
+    if val.type == ir.IntType(1):
+        return val
+    if isinstance(val.type, ir.PointerType):
+        zero = ir.Constant(val.type, None)
+    else:
+        zero = ir.Constant(val.type, 0)
+    return builder.icmp_signed("!=", val, zero)
+
+
+def handle_comparator(func, builder, op, lhs, rhs):
+    """Handle comparison operations."""
+
+    if lhs.type != rhs.type:
+        lhs, rhs = _normalize_types(func, builder, lhs, rhs)
+
+    if lhs is None or rhs is None:
+        return None
+
+    if type(op) not in COMPARISON_OPS:
+        logger.error(f"Unsupported comparison operator: {type(op)}")
+        return None
+
+    predicate = COMPARISON_OPS[type(op)]
+    result = builder.icmp_signed(predicate, lhs, rhs)
+    logger.debug(f"Comparison result: {result}")
+    return result, ir.IntType(1)

pythonbpf/functions/functions_pass.py

@@ -7,7 +7,7 @@ from dataclasses import dataclass
 from pythonbpf.helper import HelperHandlerRegistry, handle_helper_call
 from pythonbpf.type_deducer import ctypes_to_ir
 from pythonbpf.binary_ops import handle_binary_op
-from pythonbpf.expr_pass import eval_expr, handle_expr, convert_to_bool
+from pythonbpf.expr import eval_expr, handle_expr, convert_to_bool
 
 from .return_utils import _handle_none_return, _handle_xdp_return, _is_xdp_name
 
@@ -240,72 +240,13 @@ def handle_assign(
         logger.info("Unsupported assignment value type")
 
 
-def handle_cond(func, module, builder, cond, local_sym_tab, map_sym_tab):
+def handle_cond(
-    if True:
+    func, module, builder, cond, local_sym_tab, map_sym_tab, structs_sym_tab=None
-        val = eval_expr(func, module, builder, cond, local_sym_tab, map_sym_tab)[0]
+):
-        return convert_to_bool(builder, val)
+    val = eval_expr(
-    if isinstance(cond, ast.Constant):
+        func, module, builder, cond, local_sym_tab, map_sym_tab, structs_sym_tab
-        if isinstance(cond.value, bool) or isinstance(cond.value, int):
+    )[0]
-            return ir.Constant(ir.IntType(1), int(cond.value))
+    return convert_to_bool(builder, val)
-        else:
-            raise ValueError("Unsupported constant type in condition")
-            return None
-    elif isinstance(cond, ast.Name):
-        if cond.id in local_sym_tab:
-            var = local_sym_tab[cond.id].var
-            val = builder.load(var)
-            if val.type != ir.IntType(1):
-                # Convert nonzero values to true, zero to false
-                if isinstance(val.type, ir.PointerType):
-                    # For pointer types, compare with null pointer
-                    zero = ir.Constant(val.type, None)
-                else:
-                    # For integer types, compare with zero
-                    zero = ir.Constant(val.type, 0)
-                val = builder.icmp_signed("!=", val, zero)
-            return val
-        else:
-            logger.info(f"Undefined variable {cond.id} in condition")
-            return None
-    elif isinstance(cond, ast.Compare):
-        lhs = eval_expr(func, module, builder, cond.left, local_sym_tab, map_sym_tab)[0]
-        if len(cond.ops) != 1 or len(cond.comparators) != 1:
-            logger.info("Unsupported complex comparison")
-            return None
-        rhs = eval_expr(
-            func, module, builder, cond.comparators[0], local_sym_tab, map_sym_tab
-        )[0]
-        op = cond.ops[0]
-
-        if lhs.type != rhs.type:
-            if isinstance(lhs.type, ir.IntType) and isinstance(rhs.type, ir.IntType):
-                # Extend the smaller type to the larger type
-                if lhs.type.width < rhs.type.width:
-                    lhs = builder.sext(lhs, rhs.type)
-                elif lhs.type.width > rhs.type.width:
-                    rhs = builder.sext(rhs, lhs.type)
-            else:
-                logger.info("Type mismatch in comparison")
-                return None
-
-        if isinstance(op, ast.Eq):
-            return builder.icmp_signed("==", lhs, rhs)
-        elif isinstance(op, ast.NotEq):
-            return builder.icmp_signed("!=", lhs, rhs)
-        elif isinstance(op, ast.Lt):
-            return builder.icmp_signed("<", lhs, rhs)
-        elif isinstance(op, ast.LtE):
-            return builder.icmp_signed("<=", lhs, rhs)
-        elif isinstance(op, ast.Gt):
-            return builder.icmp_signed(">", lhs, rhs)
-        elif isinstance(op, ast.GtE):
-            return builder.icmp_signed(">=", lhs, rhs)
-        else:
-            logger.info("Unsupported comparison operator")
-            return None
-    else:
-        logger.info("Unsupported condition expression")
-        return None
 
 
 def handle_if(
@@ -321,7 +262,9 @@ def handle_if(
     else:
         else_block = None
 
-    cond = handle_cond(func, module, builder, stmt.test, local_sym_tab, map_sym_tab)
+    cond = handle_cond(
+        func, module, builder, stmt.test, local_sym_tab, map_sym_tab, structs_sym_tab
+    )
     if else_block:
         builder.cbranch(cond, then_block, else_block)
     else:

pythonbpf/helper/helper_utils.py

@@ -3,7 +3,7 @@ import logging
 from collections.abc import Callable
 
 from llvmlite import ir
-from pythonbpf.expr_pass import eval_expr
+from pythonbpf.expr import eval_expr
 
 logger = logging.getLogger(__name__)
 

pythonbpf/maps/maps_pass.py

@@ -3,7 +3,7 @@ from logging import Logger
 from llvmlite import ir
 from enum import Enum
 from .maps_utils import MapProcessorRegistry
-from ..debuginfo import DebugInfoGenerator
+from pythonbpf.debuginfo import DebugInfoGenerator
 import logging
 
 logger: Logger = logging.getLogger(__name__)

tests/failing_tests/conditionals/helper_cond.py

@@ -0,0 +1,34 @@
+from pythonbpf import bpf, map, section, bpfglobal, compile
+from ctypes import c_void_p, c_int64, c_uint64
+from pythonbpf.maps import HashMap
+
+# NOTE: Decided against fixing this
+# as a workaround is assigning the result of lookup to a variable
+# and then using that variable in the if statement.
+# Might fix in future.
+
+
+@bpf
+@map
+def last() -> HashMap:
+    return HashMap(key=c_uint64, value=c_uint64, max_entries=3)
+
+
+@bpf
+@section("tracepoint/syscalls/sys_enter_execve")
+def hello_world(ctx: c_void_p) -> c_int64:
+    last.update(0, 1)
+    if last.lookup(0) > 0:
+        print("Hello, World!")
+    else:
+        print("Goodbye, World!")
+    return
+
+
+@bpf
+@bpfglobal
+def LICENSE() -> str:
+    return "GPL"
+
+
+compile()

tests/failing_tests/conditionals/struct_ptr.py

@@ -0,0 +1,34 @@
+from pythonbpf import bpf, struct, section, bpfglobal, compile
+from ctypes import c_void_p, c_int64, c_uint64
+
+# NOTE: Decided against fixing this
+# as one workaround is to just check any field of the struct
+# in the if statement. Ugly but works.
+# Might fix in future.
+
+
+@bpf
+@struct
+class data_t:
+    pid: c_uint64
+    ts: c_uint64
+
+
+@bpf
+@section("tracepoint/syscalls/sys_enter_execve")
+def hello_world(ctx: c_void_p) -> c_int64:
+    dat = data_t()
+    if dat:
+        print("Hello, World!")
+    else:
+        print("Goodbye, World!")
+    return
+
+
+@bpf
+@bpfglobal
+def LICENSE() -> str:
+    return "GPL"
+
+
+compile()

tests/passing_tests/conditionals/and.py

@@ -0,0 +1,32 @@
+from pythonbpf import bpf, map, section, bpfglobal, compile
+from ctypes import c_void_p, c_int64, c_uint64
+from pythonbpf.maps import HashMap
+
+
+@bpf
+@map
+def last() -> HashMap:
+    return HashMap(key=c_uint64, value=c_uint64, max_entries=3)
+
+
+@bpf
+@section("tracepoint/syscalls/sys_enter_execve")
+def hello_world(ctx: c_void_p) -> c_int64:
+    last.update(0, 1)
+    last.update(1, 2)
+    x = last.lookup(0)
+    y = last.lookup(1)
+    if x and y:
+        print("Hello, World!")
+    else:
+        print("Goodbye, World!")
+    return
+
+
+@bpf
+@bpfglobal
+def LICENSE() -> str:
+    return "GPL"
+
+
+compile()

tests/passing_tests/conditionals/map_comp.py

@@ -14,7 +14,7 @@ def last() -> HashMap:
 def hello_world(ctx: c_void_p) -> c_int64:
     last.update(0, 1)
     tsp = last.lookup(0)
-    if not (tsp > 0):
+    if tsp > 0:
         print("Hello, World!")
     else:
         print("Goodbye, World!")

tests/passing_tests/conditionals/or.py

@@ -0,0 +1,32 @@
+from pythonbpf import bpf, map, section, bpfglobal, compile
+from ctypes import c_void_p, c_int64, c_uint64
+from pythonbpf.maps import HashMap
+
+
+@bpf
+@map
+def last() -> HashMap:
+    return HashMap(key=c_uint64, value=c_uint64, max_entries=3)
+
+
+@bpf
+@section("tracepoint/syscalls/sys_enter_execve")
+def hello_world(ctx: c_void_p) -> c_int64:
+    last.update(0, 1)
+    # last.update(1, 2)
+    x = last.lookup(0)
+    y = last.lookup(1)
+    if x or y:
+        print("Hello, World!")
+    else:
+        print("Goodbye, World!")
+    return
+
+
+@bpf
+@bpfglobal
+def LICENSE() -> str:
+    return "GPL"
+
+
+compile()

tests/passing_tests/conditionals/struct_access.py

@@ -13,7 +13,7 @@ class data_t:
 @section("tracepoint/syscalls/sys_enter_execve")
 def hello_world(ctx: c_void_p) -> c_int64:
     dat = data_t()
-    if dat.pid:
+    if dat.ts:
         print("Hello, World!")
     else:
         print("Goodbye, World!")

tests/passing_tests/conditionals/type_mismatch.py

@@ -0,0 +1,23 @@
+from pythonbpf import bpf, section, bpfglobal, compile
+from ctypes import c_void_p, c_int64, c_int32
+
+
+@bpf
+@section("tracepoint/syscalls/sys_enter_execve")
+def hello_world(ctx: c_void_p) -> c_int64:
+    x = 0
+    y = c_int32(0)
+    if x == y:
+        print("Hello, World!")
+    else:
+        print("Goodbye, World!")
+    return
+
+
+@bpf
+@bpfglobal
+def LICENSE() -> str:
+    return "GPL"
+
+
+compile()