`_FREE_THREADED_BUILD` mis-detected when Py_GIL_DISABLED is defined but set to 0
Bug report
Bug description:
I found an issue related to free-threaded build detection in Lib/profiling/sampling/sample.py.
_FREE_THREADED_BUILD = sysconfig.get_config_var("Py_GIL_DISABLED") is not None
I am running a non-free-threaded Windows build, where Py_GIL_DISABLED = 0.
Minimal repro script
Here is a minimal script that demonstrates the issue:
import sysconfig
import profiling.sampling.sample as sp
print("sp._FREE_THREADED_BUILD =", sp._FREE_THREADED_BUILD)
print("sysconfig Py_GIL_DISABLED =", sysconfig.get_config_var("Py_GIL_DISABLED"))
print("expression (is not None) =", sysconfig.get_config_var("Py_GIL_DISABLED") is not None)
print("expression (bool) =", bool(sysconfig.get_config_var("Py_GIL_DISABLED")))
print("expression (==1) =", sysconfig.get_config_var("Py_GIL_DISABLED") == 1)
print("expression (==0) =", sysconfig.get_config_var("Py_GIL_DISABLED") == 0)
print("sample module file =", sp.__file__)
Result and discussion
On my system, this produces:
sp._FREE_THREADED_BUILD = True <------!!! here
sysconfig Py_GIL_DISABLED = 0
expression (is not None) = True
expression (bool) = False
expression (==1) = False
expression (==0) = True
sample module file = d:\MyCode\cpython\Lib\profiling\sampling\sample.py
This appears to mis-detect the build type when Py_GIL_DISABLED is defined but set to 0. In that case, the variable is not None, so _FREE_THREADED_BUILD becomes True, even though the interpreter is not free-threaded.
Proposed Fix
Taking the following constraints into account:
- only_active_thread=False is the default behavior.
- only_active_thread=True is only meaningful when running in GIL profiling mode.
- only_active_thread and all_threads must not be True at the same time.
- Profiling mode must not be used together with all_threads=True.
- the original code took the branch where
_FREE_THREADED_BUILD = False, and since theRemoteUnwinderfunction did not accept anall_threadsparameter,all_threadsdefaulted to False.
I aimed to preserve the original semantics as much as possible while ensuring all test cases pass, so I believe this fix maybe appropriate. However, I find my code changes quite ugly.
diff --git a/Lib/profiling/sampling/sample.py b/Lib/profiling/sampling/sample.py
index e73306ebf2..7b43916b0c 100644
--- a/Lib/profiling/sampling/sample.py
+++ b/Lib/profiling/sampling/sample.py
@@ -41,7 +41,7 @@ def _pause_threads(unwinder, blocking):
except ImportError:
LiveStatsCollector = None
-_FREE_THREADED_BUILD = sysconfig.get_config_var("Py_GIL_DISABLED") is not None
+_FREE_THREADED_BUILD = bool(sysconfig.get_config_var("Py_GIL_DISABLED"))
# Minimum number of samples required before showing the TUI
# If fewer samples are collected, we skip the TUI and just print a message
MIN_SAMPLES_FOR_TUI = 200
@@ -71,11 +71,18 @@ def _new_unwinder(self, native, gc, opcodes, skip_non_matching_threads):
cache_frames=True, stats=self.collect_stats
)
else:
- unwinder = _remote_debugging.RemoteUnwinder(
- self.pid, only_active_thread=bool(self.all_threads), mode=self.mode, native=native, gc=gc,
- opcodes=opcodes, skip_non_matching_threads=skip_non_matching_threads,
- cache_frames=True, stats=self.collect_stats
- )
+ if self.all_threads:
+ unwinder = _remote_debugging.RemoteUnwinder(
+ self.pid, all_threads=self.all_threads, mode=self.mode, native=native, gc=gc,
+ opcodes=opcodes, skip_non_matching_threads=skip_non_matching_threads,
+ cache_frames=True, stats=self.collect_stats
+ )
+ else:
+ unwinder = _remote_debugging.RemoteUnwinder(
+ self.pid, only_active_thread=bool(self.all_threads), mode=self.mode, native=native, gc=gc,
+ opcodes=opcodes, skip_non_matching_threads=skip_non_matching_threads,
+ cache_frames=True, stats=self.collect_stats
+ )
return unwinder
def sample(self, collector, duration_sec=None, *, async_aware=False):
https://github.com/python/cpython/issues/143405#issuecomment-3707774216
CPython versions tested on:
CPython main branch
Operating systems tested on:
Windows
Linked PRs
- gh-143426
I would like to work on this , please assign this to me
Hi, I'd like to work on this issue. Looking at the previous closed PR #143426, I think the fix can be much simpler just changing is not None to bool() on line 41, without modifying the _new_unwinder method. Can I submit a minimal fix?