Crash when setting up a tasklet belonging to a different thread

Originally reported by: Kristján Valur Jónsson (Bitbucket: krisvale, GitHub: kristjanvalur)

---

The attached file demonstrates a crash recently discovered. If a tasklet created on a different thread gets setup, it crashes. This occurs because a cstack is being allocated using the wrong thread state.
This happens because of a call to "slp_ensure_linkage" in the "bind_task_to_frame" call. If this call is omitted, then all runs fine. But then we have failures in the unittests.

I have never understood what the slp_ensure_linkage is supposed to do. I would like to fix this without complicating the cstack allocation by using the task's thread state. What is the slp_ensure_linkage supposed to be doing?

tcrash.py.zip

---

• Bitbucket: https://bitbucket.org/stackless-dev/stackless/issue/60

Original comment by Anselm Kruis (Bitbucket: akruis, GitHub: akruis):

---

I just created pull request #14. It fixes the bug, but the fix is not elegant.
@ctismer and @krisvale : please have a look at the fix.

Once this issue is fixed, we can tag 2.7.9-slp.

Original comment by Kristján Valur Jónsson (Bitbucket: krisvale, GitHub: kristjanvalur):

---

I'm a bit busy at the moment with work, but will try to do this on the weekend.
:)

Original comment by Anselm Kruis (Bitbucket: akruis, GitHub: akruis):

---

I have some time to work on 2.7.9-slp. Unfortunately this patch has not been reviewed up to now.

I tested my patch on Windows 32bit and Linux 64bit. No problems so far.
I'm going to merge it now, because a clumsy patch, that fixes the bug is better than no patch at all.

Original comment by Christian Tismer (Bitbucket: ctismer, GitHub: ctismer):

---

On 30.01.15 15:58, Anselm Kruis wrote:

Hi Anselm,

sorry about this delay. I had a pretty bad phase last year, but I'm
working again, now.

Original comment by Christian Tismer (Bitbucket: ctismer, GitHub: ctismer):

---

I checked a fresh checkout of 2.7.9 on my Yosemite Mac and saw only some skipped tests, no errors:

#!python

$ make teststackless 2>&1 | grep skipped
testUnpickledProfileState (test_tstate.TestTracingState) ... skipped 'Not restorable'
OK (skipped=1)
test_bind_to_other_tid (test_thread.SchedulingBindThreadTests) ... skipped 'test requires softswitching'
test_crash (test_thread.TestRebindCrash) ... skipped 'test requires softswitching'
test_rebind (test_thread.TestRebindCrash) ... skipped 'test requires softswitching'
testUnpickledProfileState (test_tstate.TestTracingState) ... skipped 'Not restorable'
OK (skipped=4)
minimax:stackless tismer$

If that's ok with you, then I can confirm that the patch works on OSX, too.

Original comment by Anselm Kruis (Bitbucket: akruis, GitHub: akruis):

---

Stackless issue #80: fix an assertion failure during shutdown

This commit adds a test case to trigger an assertion failure during shutdown and
fixes the bug. The assertion was added by my first attempt to fix issue #60.
Thanks to Masamitsu Murase for reporting the bug and for providing the fix.

Edit: bitbucket attached this commit message to the wrong bug.

But the important lesson learned is: Just running the test suite isn't enough. It can't replace a careful code review.

Original comment by Anselm Kruis (Bitbucket: akruis, GitHub: akruis):

---

Removing milestone: 2.7.9-slp (automated comment)