[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
bug#46796: Cuirass & pointer finalization.
From: |
zimoun |
Subject: |
bug#46796: Cuirass & pointer finalization. |
Date: |
Fri, 26 Feb 2021 21:12:56 +0100 |
Hi Mathieu,
I know nothing about the topic and I probably out-of-scope.
On Fri, 26 Feb 2021 at 15:14, Mathieu Othacehe <othacehe@gnu.org> wrote:
> I'm trying to fix a memory corruption in the remote-server process of
> Cuirass since a few days. Even though I don't have a usable core dump
> file yet, I'm pretty sure the error comes from the "zmq-msg-init"
> procedure of Guile-Simple-ZMQ.
>
> This procedure creates a bytevector, call the C function zmq_msg_init to
> initialize it, adds zmq_msg_close as pointer finalizer and returns a
> wrapped pointer.
>
> My understanding is that the wrapped pointer that is passed around in
> Cuirass ensures that the underlying bytevector is not garbage collected
> until the pointer goes out of scope. However, some assertions failures
> such as this one:
>
> --8<---------------cut here---------------start------------->8---
> Assertion failed: check () (src/msg.cpp:394)
> --8<---------------cut here---------------end--------------->8---
>
> let me think that the bytevector is garbage collected, while ZMQ is
> still using it. Some help would be much appreciated here :).
>From ’zmq-msg-init’ defined here:
<https://github.com/jerry40/guile-simple-zmq/blob/master/simple-zmq.scm.in#L543>
and why is ’zmq-message-content’ used for? Since ’message’ is
initialized with zero, I guess. Well, I am confused by:
--8<---------------cut here---------------start------------->8---
(let ((content-ptr (zmq_msg_data (message->pointer message)))
[...]
(pointer->bytevector content-ptr size))))
…
(let ((msg (pointer->message! msg-pointer)))
(when content-bv
(let ((target (zmq-message-content msg)))
(bytevector-copy! content-bv 0 target 0 len)))
msg))))
--8<---------------cut here---------------end--------------->8---
Is ’target’ at the same address than ’msg’? Maybe ’target’ creates
somehow a dangling pointer.
Cheers,
simon