Support Python 3.10 #1022
Assignees
Labels
Comments
rchen152
added a commit
that referenced
this issue
Oct 6, 2021
Generated by following the mostly automated steps at the beginning of https://github.com/google/pytype/blob/master/docs/developers/python_version_upgrades.md. With this, we actually get useful test results rather than a bunch of key errors for missing Python 3.10 magic numbers. For #1022.
rchen152
added a commit
that referenced
this issue
Oct 7, 2021
Generated by following the mostly automated steps at the beginning of https://github.com/google/pytype/blob/master/docs/developers/python_version_upgrades.md. With this, we actually get useful test results rather than a bunch of key errors for missing Python 3.10 magic numbers. For #1022. Resolves #1023 PiperOrigin-RevId: 401326143
1 task
rchen152
added a commit
that referenced
this issue
Feb 17, 2022
* Change builtins.NotImplementedType to _NotImplementedType to match typeshed. * Add Concatenate, ParamSpec, TypeAlias, TypeGuard, and is_typeddict to typing. These changes get pytype.load_pytd_test and typeshed's pytype_test passing in 3.10. For #1022. PiperOrigin-RevId: 429376959
rchen152
added a commit
that referenced
this issue
Feb 18, 2022
The script missed an opcode (RERAISE) whose index changed in 3.10. Technically speaking, we should be using different class definitions for RERAISE in pyc/opcodes.py depending on the target version, but using the 3.10 definition in older versions doesn't seem to cause any issues, so it's simpler to just always use the newest definition. For #1022. PiperOrigin-RevId: 429463279
rchen152
added a commit
that referenced
this issue
Feb 18, 2022
This bug manifested in tests as an IndexError on the `op.arg = op.pretty_arg = offset_to_index[op.arg]` line in pyc/opcodes.py. Skimming through the Python changelog (https://docs.python.org/3/whatsnew/changelog.html), I found this entry: bpo-27129: The bytecode interpreter uses instruction, rather byte, offsets internally. This reduces the number of EXTENDED_ARG instructions needed and streamlines instruction dispatch a bit. The corresponding CPython commit (python/cpython@fcb55c0) helpfully includes a change to the stdlib `dis` library that shows how to adjust oparg calculation to account for this change. For #1022. PiperOrigin-RevId: 429488804
rchen152
added a commit
that referenced
this issue
Mar 1, 2022
This brings down the number of 3.10 test failures from 106 to 87. For #1022. PiperOrigin-RevId: 431061841
rchen152
added a commit
that referenced
this issue
Mar 23, 2022
* Handles a change in how function annotations are stored. * Adds the NO_NEXT flag to the new RERAISE opcode. * Implements NOP. This opcode was present in 3.9 as well but seems to have gone unused until 3.10. These fixes reduce the number of tracebacks in our test logs from 966 to 320 (and the remaining ones seem to be assertion errors for test failures). For #1022. PiperOrigin-RevId: 436792541
rchen152
added a commit
that referenced
this issue
Mar 25, 2022
… parser. In our line number calculations, we were skipping the first delta, leading to our line numbers sometimes being off by one. This fix brings down the test failure count from 320 to 69. For #1022. PiperOrigin-RevId: 437330907
rchen152
added a commit
that referenced
this issue
Mar 29, 2022
* Copies some logic for adjusting line deltas from the CPython source. I don't understand it, but it works? * Fixes an expected error message in the pyi parser test. The pre-3.10 error message was better IMO, but the error message comes from the stdlib ast parser, so we don't have control over it. We now have just 26 test failures in 8 test files left. For #1022. PiperOrigin-RevId: 437374111
rchen152
added a commit
that referenced
this issue
Mar 31, 2022
…s allowed. The immediate reason for this change is to fix a test that was failing in 3.10 because we were failing to log [not-supported-yet] for typing_extensions.TypeGuard. I took the opportunity to simplify some things in typing_overlay and typing_extensions_overlay to make future development easier and enable us to log more accurate and helpful error messages. For #1022. PiperOrigin-RevId: 438436286
rchen152
added a commit
that referenced
this issue
Apr 1, 2022
Updates some tests in cases where it was obvious to me that the problem was in the test and not the functionality under test. Only 21 failures in 5 test modules left. For #1022. PiperOrigin-RevId: 438635996
rchen152
added a commit
that referenced
this issue
Apr 6, 2022
* tests/test_errors1: All that needed to be done here was adjusting some of the expected error counts and messages. While I was in the file, I also changed some unnecessary usages of InferWithErrors to CheckWithErrors. * constant_folding_test: In 3.10, LIST_APPEND, MAP_ADD, and DICT_UPDATE are used to construct large constant lists and dicts, so constant_folding needs to know what to do with them. Only four failures in three test files left! For #1022. PiperOrigin-RevId: 439487010
rchen152
added a commit
that referenced
this issue
Apr 6, 2022
The last few test failures were due to the line numbers of RETURN_VALUE opcodes changing. The only real bugs were in pytype's handling of function type comments, and since (1) no one should have any reason to use function type comments anymore and (2) the failures are in what look like pretty uncommon edge cases, I just skipped those tests in 3.10. I also updated the upper version bound and fixed a type annotation so that the GitHub pytype check runs successfully in 3.10. For #1022. PiperOrigin-RevId: 439905120
Merged
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Python 3.10.0 came out yesterday.
The text was updated successfully, but these errors were encountered: