Bug report
Bug description:
Basically, when profiling with Tachyon I'm seeing stack traces that are just impossible.
Truth to be told, this is an absolutely pathological script with completely unstable (flutter frames) stack traces. That said, I'd expect them to be rejected (thus increasing the error rate), not corrupted.
As before, I've spotted this issue - albeit not that catastrophically - while observing my vm_remap runs.
Reproduction
def leaf_a(): return sum(range(50))
def leaf_b(): return sum(range(50))
def hot_a(): return leaf_a()
def hot_b(): return leaf_b()
while True:
hot_a(); hot_b();Then:
2026-06-12T20:56:38.335765000+0200 maurycy@gimel /Users/maurycy/work/cpython (main f4f1020?) % sudo ./python.exe -m profiling.sampling run --collapsed -o /tmp/stacks.txt -d 10 repro.py
Captured 10,000 samples in 10.00 seconds
Sample rate: 999.99 samples/sec
Error rate: 16.50
Collapsed stack output written to /tmp/stacks.txt
2026-06-12T20:56:48.704707000+0200 maurycy@gimel /Users/maurycy/work/cpython (main f4f1020?) % cat /tmp/stacks.txt
tid:12100288;<frozen runpy>:_run_module_as_main:201;<frozen runpy>:_run_code:87;repro.py:<module>:6;repro.py:hot_b:4;repro.py:leaf_a:1 1721
tid:12100288;<frozen runpy>:_run_module_as_main:201;<frozen runpy>:_run_code:87;repro.py:<module>:6;repro.py:hot_b:4;repro.py:leaf_b:2 1436
tid:12100288;<frozen runpy>:_run_module_as_main:201;<frozen runpy>:_run_code:87;repro.py:<module>:6;repro.py:hot_a:3;repro.py:leaf_b:2 1430
tid:12100288;<frozen runpy>:_run_module_as_main:201;<frozen runpy>:_run_code:87;repro.py:<module>:6 1170
tid:12100288;<frozen runpy>:_run_module_as_main:201;<frozen runpy>:_run_code:87;repro.py:<module>:6;repro.py:hot_a:3;repro.py:leaf_a:1 1075
tid:12100288;<frozen runpy>:_run_module_as_main:201;<frozen runpy>:_run_code:87;repro.py:<module>:6;repro.py:hot_a:3 413
tid:12100288;<frozen runpy>:_run_module_as_main:201;<frozen runpy>:_run_code:87;repro.py:<module>:6;repro.py:hot_b:4 396
tid:12100288;<frozen runpy>:_run_module_as_main:201;<frozen runpy>:_run_code:87;repro.py:<module>:6;repro.py:leaf_a:1 262
tid:12100288;<frozen runpy>:_run_module_as_main:201;<frozen runpy>:_run_code:87;repro.py:<module>:6;repro.py:leaf_b:2 207
tid:12100288;<frozen runpy>:_run_module_as_main:201;<frozen runpy>:_run_code:87;repro.py:<module>:6;repro.py:hot_a:-1;repro.py:leaf_b:2 75
tid:12100288;<frozen runpy>:_run_module_as_main:201;<frozen runpy>:_run_code:87;repro.py:<module>:6;repro.py:hot_b:-1;repro.py:leaf_a:1 69
tid:12100288;<frozen runpy>:_run_module_as_main:201;<frozen runpy>:_run_code:87;repro.py:<module>:6;repro.py:hot_b:-1;repro.py:leaf_b:2 41
tid:12100288;<frozen runpy>:_run_module_as_main:201;<frozen runpy>:_run_code:87;repro.py:<module>:6;repro.py:hot_a:-1;repro.py:leaf_a:1 31
tid:12100288;<frozen runpy>:_run_module_as_main:201;<frozen runpy>:_run_code:87;repro.py:<module>:6;repro.py:hot_a:3;repro.py:leaf_b:-1 11
tid:12100288;<frozen runpy>:_run_module_as_main:201;<frozen runpy>:_run_code:87;repro.py:<module>:6;repro.py:hot_b:4;repro.py:leaf_a:-1 9
tid:12100288;<frozen runpy>:_run_module_as_main:201;<frozen runpy>:_run_code:87;repro.py:<module>:6;repro.py:leaf_a:-1 2
tid:12100288;<frozen runpy>:_run_module_as_main:201;<frozen runpy>:_run_code:87;repro.py:<module>:5;repro.py:hot_a:-1;repro.py:leaf_b:2 1
tid:12100288;<frozen runpy>:_run_module_as_main:201;<frozen runpy>:_run_code:87;repro.py:<module>:6;repro.py:hot_a:3;repro.py:leaf_a:-1 1
2026-06-12T20:58:42.867874000+0200 maurycy@gimel /Users/maurycy/work/cpython (main f4f1020?) %
More precisely, out of 8476 successful stacks, 4316 (50.9%) are impossible.
Cross:
tid:12252325;...;repro.py:<module>:6;repro.py:hot_b:4;repro.py:leaf_a:1 1736
tid:12252325;...;repro.py:<module>:6;repro.py:hot_a:3;repro.py:leaf_b:2 1584
tid:12252325;...;repro.py:<module>:6;repro.py:hot_b:-1;repro.py:leaf_a:1 23
tid:12252325;...;repro.py:<module>:6;repro.py:hot_a:-1;repro.py:leaf_b:2 14
tid:12252325;...;repro.py:<module>:6;repro.py:hot_a:3;repro.py:leaf_b:-1 7
tid:12252325;...;repro.py:<module>:6;repro.py:hot_b:4;repro.py:leaf_a:-1 4
tid:12252325;...;repro.py:<module>:5;repro.py:hot_a:3;repro.py:leaf_b:2 2
tid:12252325;...;repro.py:<module>:5;repro.py:hot_b:4;repro.py:leaf_a:1 2
Orphans:
tid:12252325;...;repro.py:<module>:6;repro.py:leaf_b:2 464
tid:12252325;...;repro.py:<module>:6;repro.py:leaf_a:1 462
tid:12252325;...;repro.py:<module>:5;repro.py:leaf_b:2 1
tid:12252325;...;repro.py:<module>:6;repro.py:leaf_a:-1 1
Wrong lines:
tid:12252325;...;repro.py:<module>:6;repro.py:hot_a:-1;repro.py:leaf_a:1 7
tid:12252325;...;repro.py:<module>:6;repro.py:hot_b:-1;repro.py:leaf_b:2 4
tid:12252325;...;repro.py:<module>:5;repro.py:hot_b:4;repro.py:leaf_b:2 4
tid:12252325;...;repro.py:<module>:5;repro.py:hot_a:3;repro.py:leaf_a:1 1
Very important caveat here is that it's just 1 level deep, there are no generators etc.,
Isolation to cache_frames=True
diff --git i/Lib/profiling/sampling/sample.py w/Lib/profiling/sampling/sample.py
index 50ccc57566d..16de26b2dc5 100644
--- i/Lib/profiling/sampling/sample.py
+++ w/Lib/profiling/sampling/sample.py
@@ -110,7 +110,7 @@ def _new_unwinder(self, native, gc, opcodes, skip_non_matching_threads):
gc=gc,
opcodes=opcodes,
skip_non_matching_threads=skip_non_matching_threads,
- cache_frames=True,
+ cache_frames=os.environ.get("PROFILING_CACHE_FRAMES", "1") != "0",
stats=self.collect_stats,
**kwargs
)2026-06-13T02:57:07.911742000+0200 maurycy@gimel /Users/maurycy/src/github.com/maurycy/cpython (main f4f1020*?) % sudo PROFILING_CACHE_FRAMES=0
./python.exe -m profiling.sampling run --collapsed -o /tmp/stacks.txt -d 10 repro.py
Captured 10,000 samples in 10.00 seconds
Sample rate: 999.99 samples/sec
Error rate: 15.09
Collapsed stack output written to /tmp/stacks.txt
2026-06-13T02:57:18.021250000+0200 maurycy@gimel /Users/maurycy/src/github.com/maurycy/cpython (main f4f1020*?) % cat /tmp/stacks.txt
tid:13147584;<frozen runpy>:_run_module_as_main:201;<frozen runpy>:_run_code:87;repro.py:<module>:6;repro.py:hot_b:4;repro.py:leaf_b:2 3337
tid:13147584;<frozen runpy>:_run_module_as_main:201;<frozen runpy>:_run_code:87;repro.py:<module>:6;repro.py:hot_a:3;repro.py:leaf_a:1 3257
tid:13147584;<frozen runpy>:_run_module_as_main:201;<frozen runpy>:_run_code:87;repro.py:<module>:6 1264
tid:13147584;<frozen runpy>:_run_module_as_main:201;<frozen runpy>:_run_code:87;repro.py:<module>:6;repro.py:hot_a:3 332
tid:13147584;<frozen runpy>:_run_module_as_main:201;<frozen runpy>:_run_code:87;repro.py:<module>:6;repro.py:hot_b:4 282
tid:13147584;<frozen runpy>:_run_module_as_main:201;<frozen runpy>:_run_code:87;repro.py:<module>:6;repro.py:hot_a:3;repro.py:leaf_a:-1 9
tid:13147584;<frozen runpy>:_run_module_as_main:201;<frozen runpy>:_run_code:87;repro.py:<module>:6;repro.py:hot_b:4;repro.py:leaf_b:-1 5
tid:13147584;<frozen runpy>:_run_module_as_main:201;<frozen runpy>:_run_code:87;repro.py:<module>:5 2
tid:13147584;<frozen runpy>:_run_module_as_main:201;<frozen runpy>:_run_code:87;repro.py:<module>:6;repro.py:leaf_b:2 2
tid:13147584;<frozen runpy>:_run_module_as_main:201;<frozen runpy>:_run_code:87;repro.py:<module>:6;repro.py:hot_b:4;repro.py:leaf_a:1 1
...so now it's only 0,20% impossible stack traces (leaving figuring out which as a trivial excercise to the reader.)
CPython versions tested on:
CPython main branch
Operating systems tested on:
macOS
Linked PRs
Bug report
Bug description:
Basically, when profiling with Tachyon I'm seeing stack traces that are just impossible.
Truth to be told, this is an absolutely pathological script with completely unstable (flutter frames) stack traces. That said, I'd expect them to be rejected (thus increasing the error rate), not corrupted.
As before, I've spotted this issue - albeit not that catastrophically - while observing my vm_remap runs.
Reproduction
Then:
More precisely, out of 8476 successful stacks, 4316 (50.9%) are impossible.
Cross:
Orphans:
Wrong lines:
Very important caveat here is that it's just 1 level deep, there are no generators etc.,
Isolation to
cache_frames=True...so now it's only 0,20% impossible stack traces (leaving figuring out which as a trivial excercise to the reader.)
CPython versions tested on:
CPython main branch
Operating systems tested on:
macOS
Linked PRs
RemoteUnwinder(..., cache_frames=True)by copying chunks on cache miss #151426