xDSL

0.63.0 verified Fri May 01 auth: no python

xDSL is a Python library for defining and manipulating domain-specific languages (DSLs) with a focus on compiler infrastructure. It provides IR building blocks, pattern rewrites, and verification passes. Current version: 0.63.0. Released monthly.

pip install xdsl

Common errors

error AttributeError: module 'xdsl' has no attribute 'ir' ↓

cause Import path is wrong; the package does not expose submodules at top level.

fix

Use from xdsl.ir import Block, Region, Operation, SSAValue.

error ValueError: Operation must have at least one operand or result ↓

cause Creating an operation without specifying operands and results in the new API (post 0.60).

fix

Pass operands and result_types when constructing an operation: arith.Addi(operands=[...], result_types=[...]) or use helper methods.

error TypeError: cannot instantiate 'Block' with positional arguments ↓

cause Block no longer accepts operations as positional arguments in newer versions.

fix

Use Block.from_ops([op1, op2]) or Block(ops=[op1, op2]) keyword argument.

Warnings

breaking In version 0.60.0, the entire IR API was refactored. Operations now require explicit operands and results; the old mutable operation pattern is removed. ↓

fix Use Block.from_ops() with SSAValue references and explicit operands. See migration guide.

deprecated The xdsl.util module is deprecated; use xdsl.utils or xdsl.ir utilities instead. ↓

fix Replace xdsl.util imports with xdsl.utils equivalents.

gotcha SSAValue indices are positional and must match the block's arguments. A common mistake is to assume SSAValue(0) refers to the first operation's result; it refers to the first block argument. ↓

fix Ensure block arguments are defined first, then use SSAValue(index) where index corresponds to the argument list order.

Imports

IR builtins

wrong

from xdsl import Block

correct

from xdsl.ir import Block, Region, Operation, SSAValue

Top-level imports are not available; use xdsl.ir submodule.

MLIR type conversion (legacy)

wrong

from xdsl.mlir import MLIRConverter

correct

from xdsl.interpreters.mlir import MLIRConverter

MLIR converter moved to xdsl.interpreters.mlir.

Quickstart

Construct an MLIR-like module with a function containing an add operation.

from xdsl.ir import Block, Region, Operation, SSAValue
from xdsl.dialects.builtin import ModuleOp, IntegerType, FunctionType
from xdsl.dialects import arith, func

# Create a simple function: i32 add
int32 = IntegerType(32)
func_type = FunctionType.from_lists([int32, int32], [int32])

# Build region with add operation
block = Block.from_ops([
    arith.Addi(SSAValue(0), SSAValue(1)),
    func.Return()
])
region = Region(block)

# Wrap in FunctionOp
func_op = func.FuncOp("add", func_type, region)
module = ModuleOp([func_op])
print(module)