On Thu, Nov 09, 2000 at 02:26:19PM +1100, Martin Sevior wrote:
>
>
> On Wed, 8 Nov 2000, Sam TH wrote:
> >
> > Ha - I win. The following patch fixes the problem for me.
> > As it turns out, the problem was much worse than it appeared. Just
> > try opening a new document, and hitting the left arrow key. Instant
> > segfault. This fixes that, and the undo of fields at the beginning
> > of the document, and I think a bunch of other problems. Detailed
> > explanation after the patch.
> >
>
> Glad I was starting on this then :-)
>
> > Patch:
> >
> > diff -u -r1.307 fv_View.cpp
> > --- src/text//fmt/xp/fv_View.cpp 2000/11/09 01:00:57 1.307
> > +++ src/text//fmt/xp/fv_View.cpp 2000/11/09 02:59:46
> > @@ -4567,6 +4567,9 @@
> > UT_sint32 iPointHeight;
> >
> > fl_BlockLayout* pBlock = _findBlockAtPosition(pos);
> > + while(!pBlock)
> > + pBlock = _findBlockAtPosition((PT_DocPosition) pos++);
> > +
> > UT_ASSERT(pBlock);
> > fp_Run* pRun = pBlock->findPointCoords(pos, bEOL, xPoint, yPoint,
>iPointHeight);
> >
> > Explanation:
> >
> > The problem was, we were asking for the Block in the document at or before
> > pos. However, if pos == 1, then there was no block that fit these
> > criteria. Then, when you try to use pBlock about at the bottom of the patch
> > there, it segfaults. In order to fix this, I just said that if we don't
> > find the block there, we keep going through the document until we find
> > one.
> >
> > This could have been done deeper in the call chain resulting from
> > _findBlockAtPosition(), but this was where the syntax was easiest.
> >
> > As this isn't really the answer we want in the case that we have to
> > use the while loop, this may introduce bugs. Therefore, I haven't
> > committed it yet. If people think that it's not a good idea, I'll
> > hold off, but this issue has to be fixed before 0.7.12 goes out.
> >
>
> Yes there is a potential problem if pos runs past the end of the document.
> You might want a call to m_pDoc->getbounds() before/in the loop to make
> sure pos is sane. You use the code like this
>
>
> PT_DocPosition posBOD;
> PT_DocPosition posEOD;
> UT_Bool bRes;
>
> bRes = m_pDoc->getBounds(UT_FALSE, posBOD);
> bRes = m_pDoc->getBounds(UT_TRUE, posEOD);
> UT_ASSERT(bRes);
>
> I'll leave the details for you... (handwave, handwave)
Handwaving accepted. Here's the revised patch. If anyone has
objections, please speak now, or else I'll commit this.
Index: src/text/fmt/xp/fv_View.cpp
===================================================================
RCS file: /cvsroot/abi/src/text/fmt/xp/fv_View.cpp,v
retrieving revision 1.307
diff -u -r1.307 fv_View.cpp
--- src/text/fmt/xp/fv_View.cpp 2000/11/09 01:00:57 1.307
+++ src/text/fmt/xp/fv_View.cpp 2000/11/09 07:37:04
@@ -4566,7 +4566,23 @@
UT_sint32 yPoint;
UT_sint32 iPointHeight;
+ PT_DocPosition posEOD;
+ UT_Bool bRes;
+
+
+ // The idea of the following code is thus: we need the previous block in
+ // the document. But at the beginning of the document, sometimes there
+ // isn't a previous block. So we get the next block in the document.
+ // If we don't do this, we end up in big trouble, since we reference
+ // that block in about 8 lines. Sam, 11.9.00
+
+ bRes = m_pDoc->getBounds(UT_TRUE, posEOD);
+ UT_ASSERT(bRes);
fl_BlockLayout* pBlock = _findBlockAtPosition(pos);
+ while(!pBlock && (pos < posEOD))
+ pBlock = _findBlockAtPosition((PT_DocPosition) pos++);
+ pos = (PT_DocPosition) pos;
+
UT_ASSERT(pBlock);
fp_Run* pRun = pBlock->findPointCoords(pos, bEOL, xPoint, yPoint,
iPointHeight);
sam th
[EMAIL PROTECTED]
http://www.abisource.com/~sam/
GnuPG Key:
http://pgp5.ai.mit.edu:11371/pks/lookup?op=get&search=0xCABD33FC
PGP signature
