Re: [Bug-readline] SIGSEGV in _rl_dispatch_callback()

[Anatol Pomozov]

Hi

On Fri, Mar 7, 2014 at 7:21 PM, Chet Ramey <address@hidden> wrote:
> On 3/7/14, 6:46 PM, Anatol Pomozov wrote:
>
>>
>> So it is the exact place where python calls the readline function
>> http://hg.python.org/cpython/file/7ff62415e426/Modules/readline.c#l1115
>>  (the file in url points to 3.3.4 tag - it is where we see the crash).
>>
>>
>> Here is output of rl_readline_state and _rl_kscxt variables.
>
> It would be great if you could set watchpoints on these variables and see
> when they change.  Every time _rl_kscxt goes to 0, the readline state
> should not have RL_STATE_MULTIKEY (or it should be turned off right after
> the change to _rl_kscxt).  RL_STATE_MULTIKEY is 0x0200000.

Here is the debug session http://pastebin.com/pWGDrt0P

The interesting thing happens in _rl_dispatch_callback(). ctx parameter is

{flags = 1, subseq_arg = 0, subseq_retval = 0, dmap = 0x7f1740c95d20
<emacs_meta_keymap>,
  oldmap = 0x7f1740c96d40 <emacs_standard_keymap>, okey = 27, ocxt =
0x0, childval = 1}

r is set to "1" so RL_STATE_MULTIKEY is not cleared. A few lines below
it sets _rl_kscxt into cxt->ocxt that has 0 value.

I suspect that ocxt should not be 0 here.