Segfault/aborts calling readline.set_completer_delims in threads in a free-threaded build
#126895
Comments
devdanzin
added
the
type-crash
picnixz
added
stdlib
|
The def _test():
import doctest, difflib
return doctest.testmod(difflib)I think the issue is either with doctest or the imports, not with difflib itself. |
|
I believe this is due to import pdb
import sys
from threading import Thread
stdin = open("/dev/null")
sys.stdin = stdin
def f():
for x in range(100):
p = pdb.Pdb()
p.reset()
try: # Comment
p.set_trace() # these
except: # lines
pass # out.
p.reset()
p.set_continue()
for x in range(100):
Thread(target=f, args=()).start()So maybe the issue is just that Edit: here's a collection of errors this causes: |
|
@gaogaotiantian Is there a plan to make pdb free-threaded friendly or not? if so, how can we help, if not what should we do for this issue? |
|
I'll wait for Tian's input, but I'm happy to spend a few hours covering pdb in locks if we need to. |
|
From the code above, the conclusion I can get is Also, |
|
Does it use some weird frames APIs (e.g. |
|
The culprit seems to be from threading import Thread
import readline
def f():
for x in range(100):
readline.get_completer_delims()
readline.set_completer_delims(' \t\n`@#%^&*()=+[{]}\\|;:\'",<>?')
readline.set_completer_delims(' \t\n`@#%^&*()=+[{]}\\|;:\'",<>?')
readline.get_completer_delims()
for x in range(100):
Thread(target=f, args=()).start()Seems to fail with the same errors as the original code. |
|
It's probably the system that's not thread safe in that case. |
|
The error appears to come from CPython code, line 593 seems to be affecting module global data, right? Lines 578 to 594 in 2313f84 |
|
I won't be surprised if |
|
That issue looks a little out of date, but yeah, we're still working on the thread safety. Thanks for your input, Tian! @devdanzin, would you like to author a PR for fixing the global state in |
|
I wouldn't know how, please take it if you can :) |
|
Ok, I'm busy with a few other issues right now, so I'll leave this open to someone else for a bit. If nobody decides they want to do it, I'll get to it :) |
|
I would be a little bit more careful when dealing with |
|
If it has global state, I doubt the functions themself are thread safe. In that case, we probably should just mark it as needing the GIL. |
devdanzin
changed the title
difflib._test in threads in a free-threaded buildreadline.set_completer_delims in threads in a free-threaded build
Or add |
|
That wouldn't work for subinterpreters, would it? |
Crash report
What happened?
Calling
difflib._testin threads in a free-threaded build (withPYTHON_GIL=0) will result in aborts or segfaults, apparently related to memory issues:Segfault backtrace:
Abort 1 backtrace:
Abort 2 backtrace:
Found using fusil by @vstinner.
CPython versions tested on:
3.13, CPython main branch
Operating systems tested on:
Linux
Output from running 'python -VV' on the command line:
Python 3.14.0a1+ experimental free-threading build (heads/main-dirty:612ac283b81, Nov 16 2024, 01:37:56) [GCC 11.4.0]
The text was updated successfully, but these errors were encountered: