Re: ost::String class release 1.1.1

[David Sugar]

Do you have a sample/test case application that you can get to reliably 
segfault?

David

On Monday 05 April 2004 02:11 pm, address@hidden wrote:
> i686 AMD Athlon(tm) Processor AuthenticAMD GNU/Linux
>
> > -----Original Message-----
> > From: David Sugar [mailto:address@hidden
> > Sent: Monday, April 05, 2004 2:10 AM
> > To: Ohrnberger, Erik; address@hidden
> > Subject: Re: ost::String class release 1.1.1
> >
> >
> > The idx[slot] list is a link list sorted by size.  The place
> > to start is first
> > to confirm if the slot[idx] is being correctly initialized to
> > NULL, and, in
> > the separate place where strings are being deleted, if they
> > are being added
> > to the linked list correctly.  This may also be an alignment
> > issue, depending
> > on the architecture.  What cpu architecture is this being built on?
> >
> > On Monday 05 April 2004 01:48 pm, address@hidden wrote:
> > > David,
> > >  635 char *String::getSpace(unsigned size)
> > >  636 {
> > >  637     unsigned        slot;
> > >  638     char*           text = NULL;
> > >  639     char**          next;
> > >  640
> > >  641     if(size > slotlimit)
> > >  642         return new char[size];
> > >  643
> > >  644     slot = size / slotsize;
> > >  645     mutex.enterMutex();
> > >  646     if(!pager)
> > >  647     {
> > >  648         pager = new MemPager(pagesize);
> > >  649         idx = (char **)pager->alloc(sizeof(char *) * slots);
> > >  650         memset(idx, 0, sizeof(char *) * slots);
> > >  651     }
> > >  652     text = idx[slot];
> > >  653     if(text)
> > >  654     {
> > >  655         next = (char **)text;
> > >  656         idx[slot] = *next;
> > >  657     }
> > >  658     else
> > >  659         text = (char *)pager->alloc(++slot * slotsize);
> > >  660     mutex.leaveMutex();
> > >  661     return text;
> > >  662 }
> > >
> > > Right on line 656.
> > > slot = 1
> > > text = 0x32353964 <Address 0x32353964 out of bounds>
> > > size = 1076041732
> > >
> > > I think that size must be way off (duh).  OK, so where'd
> >
> > that come from?
> >
> > >  407 void String::copy(const String &original)
> > >  408 {
> > >  409     char *ptr;
> > >  410     content = original.content;
> > >  411
> > >  412     if(!isBig())
> > >  413         return;
> > >  414
> > >  415     if(getLength() < minsize)
> > >  416     {
> > >  417         content.ministring.length = getLength();
> > >  418         memcpy(content.ministring.text, getText(),
> >
> > getLength() + 1);
> >
> > >  419         content.ministring.big = false;
> > >  420         return;
> > >  421     }
> > >  422
> > >  423     ptr = getText();
> > >  424     content.bigstring.size =
> >
> > setSize(content.bigstring.length + 1);
> >
> > >  425     content.bigstring.text = getSpace(getSize());
> > >  426     memcpy(content.bigstring.text, ptr, getLength() + 1);
> > >  427 }
> > >
> > > at line 425
> > > getLength() = 8
> > > ptr is pointing to a valid string of text.
> > > content.bigstring.size = 32
> > >
> > > Hmm.  Hope that this helps, I'm afraid that I've not
> >
> > studied this String
> >
> > > class sufficiently to diagnose this at the moment.  Perhaps
> >
> > you'l ahve some
> >
> > > insights?
> > >
> > >   Thanks,
> > >           Erik.
> > >
> > > > -----Original Message-----
> > > > From: David Sugar [mailto:address@hidden
> > > > Sent: Monday, April 05, 2004 1:27 AM
> > > > To: Ohrnberger, Erik; address@hidden
> > > > Subject: Re: ost::String class release 1.1.1
> > > >
> > > >
> > > > Hmm...
> > > >
> > > > text is either assigned from a slot list (linked list) or
> > > > from mempager
> > > > alloc'd memory.  If the linked list is messed up, that is one
> > > > possibility.
> > > > However, the other option is if the string is too large, a
> > > > simple new is
> > > > used, and text is never modified.  Text is not initialized
> > > > because it does
> > > > not need a value until it needs to be assigned from the
> >
> > list or from
> >
> > > > mempager.  Hence, it would help to know more about where
> > > > precisely this
> > > > segfault occurs.  One thing you can do is initially set text
> > > > to NULL at the
> > > > start of getSpace, as that would make it easier to determine
> > > > what the real
> > > > state of text is at the time it happens; whether its before
> > > > or after it has
> > > > been assigned by something.
> > > >
> > > > On Monday 05 April 2004 12:15 pm, address@hidden wrote:
> > > > > Everyone,
> > > > >       I'm noticing instances of seg faults in the
> > > >
> > > > ost::String::getSpace()
> > > >
> > > > > method, where the *text variable is assigned what appears
> > > >
> > > > to be random
> > > >
> > > > > garbage, and is yet used as a pointer into some data structure.
> > > > >
> > > > >       Has anyone else experienced a similar failure?
> > > > >
> > > > >       Is there a patch to the ost::String class that needs to
> > > >
> > > > be applied?
> > > >
> > > > >       Thanks,
> > > > >               Erik.
> > > > >
> > > > >
> > > > > _______________________________________________
> > > > > Bug-commoncpp mailing list
> > > > > address@hidden
> > > > > http://mail.gnu.org/mailman/listinfo/bug-commoncpp
> > >
> > > _______________________________________________
> > > Bug-commoncpp mailing list
> > > address@hidden
> > > http://mail.gnu.org/mailman/listinfo/bug-commoncpp
>
> _______________________________________________
> Bug-commoncpp mailing list
> address@hidden
> http://mail.gnu.org/mailman/listinfo/bug-commoncpp