Re: [aspell-devel] Aspell core dump

[Kevin Atkinson]

On Wed, 15 Dec 2004, Gary Setter wrote:

> <snip>
> > Hi,
> > I tried to look into this. The win32 port does not dump core, but
> > it does reprompt for when webserver is replaced by web-server or
> > web server. I have had it stuck in an infinite loop, but I'm
> > having difficulty recreating the problem.

So does Aspell hang because it is in an infinite loop or is it just keep on 
repromting?

> >
> > At the center of the infinite loop problem is this function in
> > writeable.
> >
> > static void soundslike_next(WordEntry * w)
> > {
> >   const Str * i   = (const Str *)(w->intr[0]);
> >   const Str * end = (const Str *)(w->intr[1]);
> >   set_word(*w, *i);
> >   ++i;
> >   if (i == end) w->adv_ = 0;
> > }
> > Nothing is changing w->intr[0] or w-inter[1] so w->adv is never
> > set to zero and the loop never terminates.
> >
> > Also the WordEntry::intr array of void pointers is a bad idea in
> > my opinion. Is there any support for eliminating them?
> 
> I'm still studying writable.cpp. I'm finding what are in my
> opinion bad ideas. This is the first:
> static inline StrVector * get_vector(Str s)
> {
>   return (StrVector *)(s - sizeof(StrVector) - 2);
> }
> It seems a Vector<Str> object is allocated, but only the address
> of first string is saved. Then, when we want to use the Vector,
> we just subtract and cast.
> I think the
>  WritableReplDict::add_repl()
> and
>  WritableDict::add()
> functions are along the same lines.
> Does anyone else have problems with these functions as they are?
> 
> I would like to straighten this out as part of fixing the
> infinite loop problem.

There is nothing wrong with them.  I do it to save space.  Yes it 
contains some ugly stuff but it is completely contained within the single 
source file so I don't have to worry about it anywhere else.  I will not 
unless it causes a problem.

> Also, I had a problem saving replacement pairs that included
> hyphenated words. I fixed it in
>   PosibErr<void> SpellerImpl::store_replacement().
> The fix was to retrieve the separator characters from config and
> use them, (not just space) to break up the replacement string
> into its parts. I wasn't going to submit a patch until I fixed
> the infinite loop bug, but I would be interested in what the list
> things before I submit a patch.

I believe that is the correct thing to do but I will have to see the patch 
to be sure.

-- 
http://kevin.atkinson.dhs.org