peak.util.assembler is a simple bytecode assembler module that handles most
low-level bytecode generation details like jump offsets, stack size tracking,
line number table generation, constant and variable name index tracking, etc.
That way, you can focus your attention on the desired semantics of your
bytecode instead of on these mechanical issues.
In addition to a low-level opcode-oriented API for directly generating specific
Python bytecodes, this module also offers an extensible mini-AST framework for
generating code from high-level specifications. This framework does most of
the work needed to transform tree-like structures into linear bytecode
instructions, and includes the ability to do compile-time constant folding.
Please see the BytecodeAssembler reference manual for more details.
Changes since version 0.6:
- Fix bad stack calculations for BUILD_CLASS opcode
Changes since version 0.5.2:
- Symbolic disassembly with full emulation of backward-compatible
JUMP_IF_TRUE and JUMP_IF_FALSE opcodes on Python 2.7 – tests now
run clean on Python 2.7.
- Support for backward emulation of Python 2.7’s JUMP_IF_TRUE_OR_POP and
JUMP_IF_FALSE_OR_POP instructions on earlier Python versions; these
emulations are also used in BytecodeAssembler’s internal code generation,
for maximum performance on 2.7+ (with no change to performance on older
Changes since version 0.5.1:
- Initial support for Python 2.7’s new opcodes and semantics changes, mostly
by emulating older versions’ behavior with macros. (0.5.2 is really just
a quick-fix release to allow packages using BytecodeAssembler to run on 2.7
without having to change any of their code generation; future releases will
provide proper support for the new and changed opcodes, as well as a test
suite that doesn’t show spurious differences in the disassembly listings
under Python 2.7.)
Changes since version 0.5:
- Fix incorrect stack size calculation for MAKE_CLOSURE on Python 2.5+
Changes since version 0.3:
- New node types:
- For(iterable, assign, body) – define a “for” loop over iterable
- UnpackSequence(nodes) – unpacks a sequence that’s len(nodes) long,
and then generates the given nodes.
- LocalAssign(name) – issues a STORE_FAST, STORE_DEREF or
STORE_LOCAL as appropriate for the given name.
- Function(body, name='<lambda>', args=(), var=None, kw=None, defaults=())
– creates a nested function from body and puts it on the stack.
- If(cond, then_, else_=Pass) – “if” statement analogue
- ListComp(body) and LCAppend(value) – implement list comprehensions
- YieldStmt(value) – generates a YIELD_VALUE (plus a POP_TOP in
- Code objects are now iterable, yielding (offset, op, arg) triples,
where op is numeric and arg is either numeric or None.
- Code objects’ .code() method can now take a “parent” Code object,
to link the child code’s free variables to cell variables in the parent.
- Added Code.from_spec() classmethod, that initializes a code object from a
name and argument spec.
- Code objects now have a .nested(name, args, var, kw) method, that
creates a child code object with the same co_filename and the supplied
- Fixed incorrect stack tracking for the FOR_ITER and YIELD_VALUE
- Ensure that CO_GENERATOR flag is set if YIELD_VALUE opcode is used
- Change tests so that Python 2.3’s broken line number handling in dis.dis
and constant-folding optimizer don’t generate spurious failures in this
package’s test suite.
Changes since version 0.2:
- Added Suite, TryExcept, and TryFinally node types
- Added a Getattr node type that does static or dynamic attribute access
and constant folding
- Fixed code.from_function() not copying the co_filename attribute when
copy_lineno was specified.
- The repr() of AST nodes doesn’t include a trailing comma for 1-argument
node types any more.
- Added a Pass symbol that generates no code, a Compare() node type
that does n-way comparisons, and And() and Or() node types for doing
- The COMPARE_OP() method now accepts operator strings like "<=",
"not in", "exception match", and so on, as well as numeric opcodes.
See the standard library’s opcode module for a complete list of the
strings accepted (in the cmp_op tuple). "<>" is also accepted as an
alias for "!=".
- Added code to verify that forward jump offsets don’t exceed a 64KB span, and
support absolute backward jumps to locations >64KB.
Changes since version 0.1:
- Constant handling has been fixed so that it doesn’t confuse equal values of
differing types (e.g. 1.0 and True), or equal unhashable objects
(e.g. two empty lists).
- Removed nil, ast_curry() and folding_curry(), replacing them with
the nodetype() decorator and fold_args(); please see the docs for
- Added stack tracking across jumps, globally verifying stack level prediction
consistency and automatically rejecting attempts to generate dead code. It
should now be virtually impossible to accidentally generate bytecode that can
crash the interpreter. (If you find a way, let me know!)
Changes since version 0.0.1:
- Added massive quantities of new documentation and examples
- Full block, loop, and closure support
- High-level functional code generation from trees, with smart labels and
blocks, constant folding, extensibility, smart local variable names, etc.
- The .label() method was renamed to .here() to distinguish it from
the new smart Label objects.
- Docs and tests were moved to README.txt instead of assembler.txt
- Added a demo that implements a “switch”-like statement template that shows
how to extend the code generation system and how to abuse END_FINALLY
to implement a “computed goto” in bytecode.
- Various bug fixes
There are a few features that aren’t tested yet, and not all opcodes may be
fully supported. Also note the following limitations:
- Jumps to as-yet-undefined labels cannot span a distance greater than 65,535
- The dis() function in Python 2.3 has a bug that makes it show incorrect
line numbers when the difference between two adjacent line numbers is
greater than 255. (To work around this, the test_suite uses a later version
of dis(), but do note that it may affect your own tests if you use
dis() with Python 2.3 and use widely separated line numbers.)
If you find any other issues, please let me know.
Please also keep in mind that this is a work in progress, and the API may
change if I come up with a better way to do something.
Questions and discussion regarding this software should be directed to the
PEAK Mailing List.