GeezerVison

#!/bin/bash
# Find PyMutex definition
echo "=== PyMutex Definition ==="
rg -n 'type PyMutex|struct PyMutex' --type rust -A3 -B1

echo ""
echo "=== source_path method signature and impl ==="
# Search for source_path method definition
rg -n 'fn source_path' --type rust -A8

echo ""
echo "=== Check signal handler context ==="
# Find faulthandler_fatal_error to confirm it's a signal handler
rg -n 'faulthandler_fatal_error|extern "C"' crates/stdlib/src/faulthandler.rs -B2 -A5

#!/bin/bash
# Let's check the specific file to understand the context better
echo "=== Checking faulthandler.rs structure ==="
cat -n crates/stdlib/src/faulthandler.rs | head -300 | tail -100

#!/bin/bash
# Check what the Frame structure looks like and where source_path comes from
echo "=== Frame structure and code field ==="
rg -n 'struct Frame|pub struct Frame' --type rust -A10

echo ""
echo "=== PyCode structure related to source_path ==="
rg -n 'pub struct PyCode|source_path' --type rust -B2 -A8

In `@crates/stdlib/src/faulthandler.rs` around lines 254 - 256,
dump_frame_from_raw is called from the signal handler faulthandler_fatal_error
but calls frame.code.source_path() which acquires a PyMutex and can deadlock;
make this signal-safe by avoiding any mutex acquisition: either add a
signal-safe cached string pointer on PyCode (e.g., an atomic/AtomicPtr holding a
nul-terminated C string updated outside signal handlers and read here) and read
that in dump_frame_from_raw, or replace the call to source_path() with a
non-blocking attempt (call try_lock() on the PyCode/source_path mutex and if it
fails print a fallback like "<unknown>" or the raw pointer/address), updating
references to Frame/ PyCode/ source_path accordingly so dump_frame_from_raw
never blocks in a signal context.