Fix tab completion by mdboom · Pull Request #2055 · NVIDIA/cuda-python

mdboom · 2026-05-08T14:15:22Z

This is a possible fix for #2053.

I almost never reach for monkey-patching as a solution, however in this case:

The alternatives are way worse and intrusive and force us into API design corners for the niche use case of tab completion (see my comments about the agent's other suggestions).
The monkeypatch is quite narrow -- it doesn't patch builtins that might have broad implications. It should only affect autocompletion with readline. A project like that might exist, but it's probably pretty niche. IPython and Jupyter are not affected by the bug to begin with, and the patch has no effect on them. (The patch is /installed/ with IPython, but then the code never gets called. I experimented with detecting IPython and not installing the patch, but it was complex, specific to IPython, and ultimately doesn't matter).
This monkeypatch is idempotent. If it gets cargo-culted to multiple projects, and those projects get imported in the same process, it will still work and be effective.
(Pending the CI run passing) It should be fine across all versions of Python we support, and there is a test here to confirm that going forward. If Python "fixes" the issue such that this patch is no longer needed, the test here should catch that and we can gate it at that point (as part of the usual "bring up on a new Python version" process).

mdboom · 2026-05-08T14:28:00Z

+    dev = Device(0)
+    dev.set_current()
+    mr = DeviceMemoryResource(dev)
+    assert not mr.is_ipc_enabled, "test setup: mr should not be IPC-enabled"


This is only testing one specific class that we know broke things. If we accept this PR, we should probably add testing that /all/ of our Cython classes autocomplete correctly.

mdboom · 2026-05-08T14:36:46Z

Test failures indicate something wrong in the design of the test -- not a problem with the monkey-patch itself.

kkraus14 · 2026-05-08T15:01:38Z

+    if not (hasattr(sys, "ps1") or sys.flags.inspect):
+        # Plain `python script.py`, `python -c ...`, pytest, etc.
+        return


This doesn't seem quite correct, where if I just run python from a shell it drops me to a REPL that I should be able to get autocompletion with but sys.flags.inspect is 0.

Yes, but ps1 is there, right. It WFM with python.

kkraus14 · 2026-05-08T15:02:23Z

+    # This works by overriding the `property` built-in with a custom subclass of
+    # property, but only in the rlcompleter module.  This subclass overrides the
+    # `__instancecheck__` method to also return True for getset_descriptor and
+    # member_descriptor types, which are what Cython uses for properties on cdef
+    # classes.


Are we concerned about any implications of other modules Cython classes getset_descriptor objects?

No. If they wanted them to be private, raising an exception unconditionally would not be the way.

leofang · 2026-05-11T16:08:33Z

+    class _PatchedProperty(metaclass=_PatchedPropMeta):
+        pass
+
+    rlcompleter.property = _PatchedProperty


Note for posterity: This patches the built-in property but only within this module.

Co-authored-by: Leo Fang <leof@nvidia.com>

mdboom · 2026-05-11T16:43:56Z

This still had a test to confirm things aren't fixed when the patch isn't installed. Since we always install the patch now (except when the envvar is set), that test was failing. Removed the test -- hopefully things pass now.

leofang · 2026-05-11T20:03:53Z

Admin-merging this to unblock the release... There seems to exist a job scheduling problem that the L4 job refuses to run.

github-actions · 2026-05-12T00:34:10Z

Doc Preview CI
Preview removed because the pull request was closed or merged.

@leofang

* Fix tab completion * Fix tests * Always install the monkeypatch * Update release note * Apply suggestion from @leofang Co-authored-by: Leo Fang <leof@nvidia.com> * Fix test * Fix tests hanging on Windows --------- Co-authored-by: Leo Fang <leof@nvidia.com>

`cuda/core/__init__.py` reads `CUDA_CORE_DONT_FIX_TAB_COMPLETION` with a bare `int(os.environ.get(..., "0"))` at import time. `int()` raises for any value that is not a base-10 integer, and `os.environ.get` returns the empty string (not the `"0"` default) when the variable is set but empty, so: export CUDA_CORE_DONT_FIX_TAB_COMPLETION= python -c "import cuda.core" ValueError: invalid literal for int() with base 10: '' Clearing a variable with `export VAR=` is the usual way to neutralize it in a shell profile, a Dockerfile, or a CI job spec, and `=true` / `=yes` are the obvious guesses for a boolean-looking opt-out. All of them make the whole package unimportable, which is a hard failure for a knob whose only purpose is to skip an optional `rlcompleter` patch. Parse the value leniently instead. Integer values keep their existing meaning (non-zero opts out, so `0` and `00` still install the patch), while a non-integer, non-empty value is honored as an opt-out rather than being silently ignored. Unset and empty/whitespace-only both mean "not set". Also document the variable, which was not listed on the environment variables page, and drop the stale "only installed in interactive mode" comment: the interactivity gate was intentionally removed in NVIDIA#2055 ("Always install the monkeypatch"), so the patch has been unconditional since then. The new parametrized test asserts the resulting behavior for eight values; four of them ("", " ", "true", "yes") fail on main because the subprocess exits non-zero with the ValueError above.

…lue (#2535) `cuda/core/__init__.py` reads `CUDA_CORE_DONT_FIX_TAB_COMPLETION` with a bare `int(os.environ.get(..., "0"))` at import time. `int()` raises for any value that is not a base-10 integer, and `os.environ.get` returns the empty string (not the `"0"` default) when the variable is set but empty, so: export CUDA_CORE_DONT_FIX_TAB_COMPLETION= python -c "import cuda.core" ValueError: invalid literal for int() with base 10: '' Clearing a variable with `export VAR=` is the usual way to neutralize it in a shell profile, a Dockerfile, or a CI job spec, and `=true` / `=yes` are the obvious guesses for a boolean-looking opt-out. All of them make the whole package unimportable, which is a hard failure for a knob whose only purpose is to skip an optional `rlcompleter` patch. Parse the value leniently instead. Integer values keep their existing meaning (non-zero opts out, so `0` and `00` still install the patch), while a non-integer, non-empty value is honored as an opt-out rather than being silently ignored. Unset and empty/whitespace-only both mean "not set". Also document the variable, which was not listed on the environment variables page, and drop the stale "only installed in interactive mode" comment: the interactivity gate was intentionally removed in #2055 ("Always install the monkeypatch"), so the patch has been unconditional since then. The new parametrized test asserts the resulting behavior for eight values; four of them ("", " ", "true", "yes") fail on main because the subprocess exits non-zero with the ValueError above. Co-authored-by: Michael Droettboom <mdboom@gmail.com>

Fix tab completion

a4bc060

mdboom self-assigned this May 8, 2026

mdboom added the cuda.core Everything related to the cuda.core module label May 8, 2026

mdboom added this to the cuda.core v1.0.0 milestone May 8, 2026

mdboom added bug Something isn't working P0 High priority - Must do! labels May 8, 2026

mdboom commented May 8, 2026

View reviewed changes

This comment has been minimized.

Sign in to view

Fix tests

6ce0cad

mdboom requested a review from leofang May 8, 2026 14:38

kkraus14 reviewed May 8, 2026

View reviewed changes

mdboom added 2 commits May 11, 2026 11:57

Always install the monkeypatch

ddac793

Update release note

3e57505

leofang mentioned this pull request May 11, 2026

Use Python properties for Cython descriptors #2056

Closed

leofang approved these changes May 11, 2026

View reviewed changes

leofang requested a review from kkraus14 May 11, 2026 16:09

mdboom and others added 2 commits May 11, 2026 12:41

Apply suggestion from @leofang

4a8cad4

Co-authored-by: Leo Fang <leof@nvidia.com>

Fix test

fa01701

leofang enabled auto-merge (squash) May 11, 2026 17:02

Fix tests hanging on Windows

8cad1b3

leofang disabled auto-merge May 11, 2026 20:03

leofang merged commit d62ff30 into NVIDIA:main May 11, 2026
179 of 181 checks passed

This comment has been minimized.

Sign in to view

leofang linked an issue May 12, 2026 that may be closed by this pull request

[BUG]: Tab completion with DeviceMemoryResource is broken #2053

Closed

1 task

LeSingh1 mentioned this pull request Aug 9, 2026

fix(core): don't crash import cuda.core when CUDA_CORE_DONT_FIX_TAB_COMPLETION is not an integer #2535

Merged

Navigation Menu

Uh oh!

Fix tab completion - #2055

Fix tab completion#2055
leofang merged 7 commits into
NVIDIA:mainfrom
mdboom:fix-autocompletion

mdboom commented May 8, 2026 •

edited

Loading

Uh oh!

mdboom May 8, 2026 •

edited

Loading

Uh oh!

This comment has been minimized.

mdboom commented May 8, 2026

Uh oh!

kkraus14 May 8, 2026

Uh oh!

mdboom May 8, 2026

Uh oh!

kkraus14 May 8, 2026

Uh oh!

mdboom May 9, 2026

Uh oh!

Uh oh!

leofang May 11, 2026

Uh oh!

mdboom commented May 11, 2026 •

edited

Loading

Uh oh!

leofang commented May 11, 2026

Uh oh!

Uh oh!

This comment has been minimized.

github-actions Bot commented May 12, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Uh oh!

Conversation

mdboom commented May 8, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

mdboom May 8, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

This comment has been minimized.

mdboom commented May 8, 2026

Uh oh!

kkraus14 May 8, 2026

Choose a reason for hiding this comment

Uh oh!

mdboom May 8, 2026

Choose a reason for hiding this comment

Uh oh!

kkraus14 May 8, 2026

Choose a reason for hiding this comment

Uh oh!

mdboom May 9, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

leofang May 11, 2026

Choose a reason for hiding this comment

Uh oh!

mdboom commented May 11, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

leofang commented May 11, 2026

Uh oh!

Uh oh!

This comment has been minimized.

github-actions Bot commented May 12, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

mdboom commented May 8, 2026 •

edited

Loading

mdboom May 8, 2026 •

edited

Loading

mdboom commented May 11, 2026 •

edited

Loading