Re: dangling pointer crash in write_to_festival (was: witekfl branch status)

On Tue, Mar 06, 2007 at 11:33:44PM +0200, Kalle Olavi Niemitalo wrote:
> Witold Filipczyk <[email protected]> writes:
> 
> > This feature does not collide with screen readers.
> > I just want to listen to ELinks sometimes.
> 
> I tried applying the speech commits to master, but it crashes if,
> during the speech, I close the tab and thereby cause the struct
> document_view to be freed:
> 
> (gdb) backtrace
> #0  0x08115569 in write_to_festival (fest=0x81938b8) at /home/Kalle/src/elinks/src/viewer/text/festival.c:67
> #1  0x08115519 in read_from_festival (fest=0x81938b8) at /home/Kalle/src/elinks/src/viewer/text/festival.c:45
> #2  0x080ca26b in select_loop (init=0x80c8e23 <init>) at /home/Kalle/src/elinks/src/main/select.c:289
> #3  0x080c95af in main (argc=1, argv=0xbffc2444) at /home/Kalle/src/elinks/src/main/main.c:365
> (gdb) frame
> #0  0x08115569 in write_to_festival (fest=0x81938b8) at /home/Kalle/src/elinks/src/viewer/text/festival.c:67
> 67              if (fest->line >= doc->height)
> (gdb) print doc
> $3 = (struct document *) 0x8
> (gdb) list
> 62              int len;
> 63              struct document_view *doc_view = fest->doc_view;
> 64              struct document *doc = doc_view->document;
> 65              struct screen_char *data;
> 66
> 67              if (fest->line >= doc->height)
> 68                      fest->running = 0;
> 69              if (!fest->running)
> 70                      return;
> 71
> 
> This was with the following commits applied on top of
> f2fc4020934621afb9584a468bd87180059ee8c8 (in this order):
> 
> 4e93cbf496c82926f42c0eaf270920f126ace3f8
> 9064e6323b493b5614a9bd02c25729ce2f1650bf
> f260691ac4f58e7ce0e282d7b48bddbae8f00828
> c187df9a0adcf0f9821d9b14b1dfcf43139d9bb3
> e965d07055f5dd3e046469232e4b3986fb60cbaf
> 60fc3bd04fe3f85c66d1dadbc8ba4f56f576f611
> 91be2ea6b89a7514b75fa31dcba6d9a5ef6c978c
> 4d7c491a22c0b9a191df363504f52f8da1c639e1
> 0da23da6b23d25ceb78f0229132d8efb9f3d3781
> 
> I think, before the speech code is pushed to master, one should
> either
> 
> - fix the bug.  A new "document-view-delete" event might be a
>   clean way to do this.

I reset festival.doc_view in really_close_tab. Is it acceptable?

When a link is followed and the previous document was read out, the new document
is read out from the line number of the previous document.
How to handle this?

Witek