Re: [Gtk2hs-users] reliable segfault

[<wag...@se...>]

Quoting wag...@se...:

> In the latest stable release of gtk, the following minimal program  
> segfaults as soon as I whack any key:
>
> <actual program snipped>

Okay, I investigated this problem. It turns out that the TextBuffer  
class's "insert-text" signal sends its callback a (char *) and a  
(gint) representing its length -- and is *the only signal* in the  
entire Gtk library to do this rather than using a null-terminated  
string. Brilliant.

I've enclosed a patch that fixes the problem (at least on my machine),  
but I'd love a code review for it. The description of the patch tells  
why -- my solution is somewhat unclean, in that it copies and modifies  
some generated code. I'd love to hear some feedback -- is this  
acceptable? If not, how should it be done instead?

I've also enclosed a few other patches for Entry and EntryBuffer that  
I did while trying to understand what was going wrong with TextView  
and TextBuffer -- which are very close brothers.

Thanks!
~d

Thu Jun  9 19:55:56 EDT 2011  Daniel Wagner <da...@wa...>
   * bind to the gtk_entry_new_with_buffer function

Thu Jun  9 19:56:25 EDT 2011  Daniel Wagner <da...@wa...>
   * add the "entryBuffer" attribute to bind to the  
"gtk_entry_set_buffer" and "gtk_entry_get_buffer" functions

Thu Jun  9 20:54:56 EDT 2011  Daniel Wagner <da...@wa...>
   * instead of recording the bound [0,65535] in the documentation,  
record it in the types

   This changes the types of a few attributes and functions in a
   backwards-incompatible way. At least in the case of the EntryBuffer  
attributes,
   this should be no problem, since there was no way to actually use  
EntryBuffer's
   until recently, anyway.

Thu Jun  9 21:16:47 EDT 2011  Daniel Wagner <da...@wa...>
   * fix a segfaulting signal

   It turns out that the "insert-text" signal attached to the  
GtkTextBuffer class
   passes a string and its byte length, rather than a null-terminated  
string like
   every other string-based signal in the entire Gtk library. Since the Gtk
   library doesn't distinguish between null-terminated and length-terminated
   strings, this means we can't use the built-in list of possible  
signal types to
   automatically generate a good connection function. This patch copies the
   definition of connect_BOXED_STRING__NONE into the TextBuffer  
module, modifying
   it as necessary to handle length-terminated strings.

   This may not be the cleanest solution, but it is fairly simple and  
seems to fix
   the segfault in my test case.