Python 3.15 support by Quinntyx · Pull Request #186 · rocky/python-xdis

Quinntyx · 2026-05-29T20:17:01Z

Implements Python 3.15 support.

This pull request aims to accomplish the following:

Implement unmarshal.py support for the new frozendict builtin type (and add tests for it)
- Create a polyfill type in cross_types.py to facilitate disassembling 3.15.0 code on pre-3.15 versions
Implement support for lazy/eager imports
Renumber opcodes to match 3.15's new opcode numbers
Update stack effect numbers to match new changes (affected: END_SEND, POP_ITER, GET_ITER, CALL_LIST_APPEND, SEND_GEN, LOAD_ATTR_GETATTRIBUTE_OVERRIDDEN)
Implement support for new opcodes/instructions (BINARY_OP_SUBSCR_USTR_INT, CALL_EX_NON_PY_GENERAL, CALL_EX_PY, FOR_ITER_VIRTUAL, GET_ITER_SELF, GET_ITER_VIRTUAL, RESUME_CHECK_JIT, SEND_ASYNC_GEN, SEND_VIRTUAL)
Add the 3.15.0b1 magic number (3666), which should hopefully be the 3.15.0 magic number as well

There may be things I've missed. In particular, I haven't really touched any of the formatting code. I'll also keep testing this in the coming weeks to make sure it's fully correct.

….15+ bytecode on pre-3.15 versions

…t of None out of co_consts and into LOAD_COMMON_CONSTS

…3 supports surrogate code points

…plementation.CPython

Frozendict and big_dict disassembly reference files for 3.15.

…orts

rocky · 2026-05-31T15:25:47Z

+import struct
+import time
+
+# 1. Compile a simple dummy script


It would be nice to remove the (Gemini?) AI numbers. I think the motivation for these is so that when you are in conversation with Gemini or AI, it is easy for you to refer to various steps. However, in the long term, the numbers are going to get in the way and become out of order.

I will do this after merge.

rocky · 2026-05-31T15:26:24Z

 # ----- configure this for your needs

-lib_prefix = "/usr/lib"
+lib_prefix = os.environ.get("XDIS_LIB_PREFIX", "/usr/lib")


Probably a good idea. Thanks!

rocky · 2026-05-31T15:26:54Z

-         source_size) = load_module(mod_file)
-        self.assertEqual(version, 2.5,
+         source_size, *_) = load_module(mod_file)
+        self.assertEqual(version, (2, 5),


Thanks for the fix!

rocky · 2026-05-31T15:28:24Z

 add_canonic_versions("3.14-dev", "3.14b3")
-add_canonic_versions("3.14 3.14.0 3.14.1, 3.14.2 3.14.3 3.14.4", "3.14rc3")
+add_canonic_versions("3.14 3.14.0 3.14.1, 3.14.2 3.14.3 3.14.4 3.14.5", "3.14rc3")


Thanks for the update.

rocky · 2026-05-31T15:31:52Z

@@ -0,0 +1,414 @@
+# (C) Copyright 2026 by Rocky Bernstein


You can substitute your name here. You did the work!

Oh, sure, I wasn't sure about that lol
I don't know if you want me to make another commit on my branch (maybe awkward since this PR is already merged) or...
I have not that much experience working on PRs for projects like this so I'm not entirely sure what the process is usually like

Oh, sure, I wasn't sure about that lol I don't know if you want me to make another commit on my branch (maybe awkward since this PR is already merged) or...

I'd be grateful if you just add a new PR with the change .

I have not that much experience working on PRs for projects like this so I'm not entirely sure what the process is usually like

I have much experience and am stil not entire sure what the process is usually like. :-) It tends to change from project to project.

Done, I opened it as PR #187 with a single commit

rocky · 2026-05-31T15:55:00Z

There may be things I've missed. In particular, I haven't really touched any of the formatting code.

Custom formatting can wait.

(In the long run it should be generalized more, but this kind of idea will always be around.)

I'll also keep testing this in the coming weeks to make sure it's fully correct.

Thanks.

Be aware that Python 3.15-dev may change in incompatible ways, and it typically does. I tried running tests on the 3.15-dev, and 3.15a7, and I got failures:

stream_fixture = <_io.StringIO object at 0x7bcba94477f0>

    def test_show_code(stream_fixture):
        dis.show_code(TEST_SOURCE_CODE, file=stream_fixture)
        actual = stream_fixture.getvalue()
>       assert actual == EXPECTED_CODE_INFO + "\n"
E       AssertionError: assert '# Method Nam...\n#    0: a\n' == '# Method Nam...\n#    0: a\n'
E         
E         Skipping 273 identical leading characters in diff, use -v to show
E             0: 10
E         + #    1: None
E           # Names:
E           #    0: a

pytest/test_std.py:138: AssertionError
======================================================================================== short test summary info =========================================================================================
FAILED pytest/test_stack_effect.py::test_stack_effect_vs_dis - AssertionError: 10 (END_SEND)  not okay; effect -2 vs -1
FAILED pytest/test_std.py::test_bytecode_info - AssertionError: assert '# Method Nam...s:\n#    0: a' == '# Method Nam...s:\n#    0: a'
FAILED pytest/test_std.py::test_code_info - AssertionError: assert '# Method Nam...s:\n#    0: a' == '# Method Nam...s:\n#    0: a'
FAILED pytest/test_std.py::test_show_code - AssertionError: assert '# Method Nam...\n#    0: a\n' == '# Method Nam...\n#    0: a\n'
================================================================================ 4 failed, 28 passed, 4 skipped in 0.15s =================================================================================

The other thing to note is that we support disassembling from an old Python to a current one! And this is done in branches python-3.6-to-3.10 for example. But I've usually been handling the merges from master into these branches.

Quinntyx · 2026-06-02T15:16:25Z

I'm aware that Python 3.15-dev may change in incompatible ways, but I do believe that now that the beta is released it shouldn't happen? At least, the bytecode is supposed to be frozen now.

rocky · 2026-06-02T15:37:03Z

I'm aware that Python 3.15-dev may change in incompatible ways, but I do believe that now that the beta is released it shouldn't happen? At least, the bytecode is supposed to be frozen now.

Weirder things have happened in Python bytecode.

Quinntyx · 2026-06-02T19:23:10Z

Well, I guess I just hope that it doesn't change and if it does I'll have to fix it then

rocky · 2026-06-03T11:19:46Z

@Quinntyx A couple of things I noticed after merge.

The main thing is that new code should not use any Python 3.15isms just yet; it needs to work on Python 3.11. This may change depending on how different 3.15 is from 3.14.

In particular, instead of list[type] and dict[type] use List[type] and Dict[type].

Quinntyx · 2026-06-04T20:47:07Z

I'm relatively sure that I haven't used any of the Python 3.15isms as I was testing my branch of xdis using Python 3.14 to run (decompiling 3.15-0b1 bytecode).

However, I wasn't aware that list[type] was not available in some older Python versions, thanks for pointing it out! I'll probably also test decompiling 3.15 bytecode from 3.11 and maybe some earlier versions to make sure that I didn't accidentally use any newer features that break compatibility with old versions.

Quinntyx · 2026-06-04T20:59:41Z

After some research, I see that list[type] and dict[type] etc were added in Python 3.9, and we are targeting 3.11, so they should be available, no? In my testing, my branch of xdis seems to run just fine in a 3.11 environment. Is there something I'm missing?

rocky · 2026-06-04T23:14:25Z

After some research, I see that list[type] and dict[type] etc were added in Python 3.9, and we are targeting 3.11, so they should be available, no? In my testing, my branch of xdis seems to run just fine in a 3.11 environment. Is there something I'm missing?

Ok. My mistake then. Thanks for checking. xdis supports Python going back before 3.9, and probably I got confused in working with one of those other branches.. In the future, then list[type] is okay for the master branch.

Quinntyx · 2026-06-05T16:09:06Z

Okay, thanks for the clarification!

Quinntyx added 20 commits May 27, 2026 16:15

feat: update magic int from 3655 (CPython 3.15-dev) to 3666.

28bc162

feat: added support for new opcodes from CPython 3.15.0b1

203fa1a

feat: implemented frozendict unmarshal for CPython 3.15+

13f45d8

feat: added frozendict polyfill type to allow unmarshalling CPython 3…

5eea3e2

….15+ bytecode on pre-3.15 versions

feat: implemented 3.15 lazy/eager import instruction decoding and offset

7836e73

fix: updated test to skip 1: None requirement on 3.15+ due to movemen…

dd8d9b7

…t of None out of co_consts and into LOAD_COMMON_CONSTS

fix: configurable lib_prefix via env variable

bb477df

chore: added 3.15 to version list

89bc9b0

feat: created frozendict unmarshalling test

248cc13

fix: t_long now uses read_int32 instead of read_uint32

7de2db2

fix: t_dict can no longer terminate on null values, only null keys

bbade08

fix: t_unicode now unmarshalls with surrogatepass enabled, as Python …

b9f70e9

…3 supports surrogate code points

chore: generated pyc test cases for CPython 3.15.0b1

e8d33d0

feat: updated opcodes to match new dis hastable behavior

7de2f0a

chore: added 3.15 tests to test infrastructure

4ca26a1

feat: added 3.15 frozendict marshalling test.

b6f218b

fix: updated get_opcode_module() to use sys.version_info and PythonIm…

af9fb41

…plementation.CPython

chore: added verified 3.15 .right test data files

93a897f

Frozendict and big_dict disassembly reference files for 3.15.

fix: register Python 3.14.5 as canonical version in magics and op_imp…

bc30431

…orts

chore: add Python 3.15 to cross-version tox environments

2692140

rocky reviewed May 31, 2026

View reviewed changes

rocky merged commit a79d186 into rocky:master Jun 2, 2026
12 checks passed

Uh oh!

Conversation

Quinntyx commented May 29, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

rocky commented May 31, 2026

Uh oh!

Uh oh!

Quinntyx commented Jun 2, 2026

Uh oh!

rocky commented Jun 2, 2026

Uh oh!

Quinntyx commented Jun 2, 2026

Uh oh!

rocky commented Jun 3, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Quinntyx commented Jun 4, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Quinntyx commented Jun 4, 2026

Uh oh!

rocky commented Jun 4, 2026

Uh oh!

Quinntyx commented Jun 5, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

rocky commented Jun 3, 2026 •

edited

Loading

Quinntyx commented Jun 4, 2026 •

edited

Loading