Re: Some new pf_frag methods...?

On Fri, 2011-07-01 at 13:03 +1000, Martin Sevior wrote:
> Hi Ben,
>
> I'm still working my way through these. Have you updated your GIT
> repository the svn trunk? We made some important fixes to the
> Piecetable a few weeks ago.

My git is basically still the same subversion it was when I svn checked
out to make it. I thought there was some magic with git merging that
made things easier to merge into trunk again. Perhaps Marc has some
insight into the simplest way to roll things back in.

Many of my piecetable things are new methods and should work OK with an
updated PT implementation.

>
> Cheers
>
> Martin
>
>
> On Fri, Jul 1, 2011 at 10:16 AM, Ben Martin
> <monkeyiq@users.sourceforge.net> wrote:
> > Hi,
> > While hacking away on some odf change tracking updates [1] I added a
> > few methods to the core piecetable and pf_frag tree in my git repo. I
> > was doing some things over and over and they seemed good candidates for
> > abstraction, but there is also the possibility that I just missed the
> > current good way to do the same thing :/ I might have missed one or two
> > changes here, the patch will tell all ;)
> >
> > I added
> > pf_Frag_Strux*
> > pt_PieceTable::_getBlockFromPosition(PT_DocPosition pos) const;
> >
> > Because using _getStruxOfTypeFromPosition() if pos == a PTX_Block then
> > you get the previous block instead of the block at pos. Using the new
> > _getBlockFromPosition() will return the PTX_Block right at pos if one
> > exists.
> >
> > I added the below method before the above one, it is a simpler
> > implementation now. Basically if startpos and endpos are in the same
> > PTX_Block then a pointer to its Frag_Strux is returned or NULL
> > otherwise. So it can be used as a bool test to see if these positions
> > share a block, and also you can use the return value to access that same
> > block if desired.
> >
> > pf_Frag_Strux*
> > pt_PieceTable::inSameBlock(
> > PT_DocPosition startpos,
> > PT_DocPosition endpos );
> >
> > The below will get the first strux that marks the end of the block
> > containing currentpos or null. The end of block strux has to be
> > positioned before endpos or null is returned. Note that if currentpos is
> > itself a block this method will move over that PTX_Block before
> > searching for the end of block to allow simpler iteration of the
> > document. endpos could be optional, but it is handy for selections which
> > is the context I created it for. ie, you want to get the end of block
> > only if it is not beyond the user's active text selection.
> >
> > pf_Frag*
> > pt_PieceTable::getEndOfBlock(
> > PT_DocPosition currentpos,
> > PT_DocPosition endpos );
> >
> > I extended _findLastStruxOfType to take a stopCondition as follows. Note
> > that the old method is still there too and just delegates to this new
> > one. IIRC the return value used to be pf_Frag, but as we are explicitly
> > finding a frag_strux I changed that to the later to give the best type
> > information to the caller. The stopConditions is an array terminated by
> > Dummy. The method comment is;
> >
> > /**
> > * MIQ11: Extends the old _findLastStruxOfType adding a stopCondition
> > * for failure and returning a Strux* directly in the case of success.
> > * This is like a findBackwards() from a fragment.
> > *
> > * stopConditions must be terminated with a PTX_StruxDummy entry like:
> > * PTStruxType stopCondition[] = { PTX_SectionTable, PTX_StruxDummy };
> > *
> > * Find a fragment of strux type pst looking backwards from pfStart.
> > * If a strux fragment matching the stopCondition is found first then
> > * the function stops and returns 0. If no fragment with pst is found
> > * then 0 is returned.
> > *
> > * MAYBE: extend this again to take yes() and no() functors so a
> > * function can call _findLastStruxOfType() and decide what is ok
> > * and what is not using those.
> > * boost::lambda would be handy to simplify the functors?
> > */
> > pf_Frag_Strux* _findLastStruxOfType(
> > pf_Frag * pfStart,
> > PTStruxType pst,
> > PTStruxType* stopConditions,
> > bool bSkipEmbededSections );
> >
> > ------------
> >
> > In pf_frag I added the following:
> >
> > The getNextStrux() is like getNext() but will skip over fragments until
> > a strux of this type is found. Return 0 otherwise
> >
> > pf_Frag_Strux* getNextStrux(PTStruxType t) const;
> >
> > In the below, if this fragment is a strux, and it is of the type passed
> > in then return this fragment downcast to the frag_strux type. It seems
> > to be quite common to test for the strux and then subtype, and then
> > static_cast<> down to the subtype.
> >
> > pf_Frag_Strux* tryDownCastStrux(PTStruxType t) const;
> >
> >
> > It might also be an idea to add templates for downcasts like
> >
> > template <class Frag>
> > Frag* tryDownCastStrux() const;
> >
> > To be used like
> > pf_Frag_Text* pft = pf->tryDownCastStrux<pf_Frag_Text>();
> >
> > Of course the template parameter can be omitted if the pft is passed as
> > an argument instead of as the return value, but then you have to declare
> > the pft before the call. This way the whole thing can be put into a line
> > like the following. But if you pass the class or strux enum in this
> > process doesn't seem to make much difference.
> >
> > if( pf_Frag_Text* pft = pf->tryDownCastStrux<pf_Frag_Text>() )
> > {
> > // use pft
> > }
> >
> > Mainly for debugging I added the following to pf_frag_text. This avoids
> > using getBufIndex() and playing with char pointers directly, but does
> > require a string copy for the result. Not a huge problem in #if DEBUG
> > code though.
> >
> > std::string toString() const;
> >
> >
> > ------------
> >
> > I also have some additions to pp_revision
> >
> > To merge another RevisionAttr
> >
> > void mergeAll( const PP_RevisionAttr& ra );
> >
> > To add a attr=value at a given revisionid and type
> >
> > void mergeAttr( UT_uint32 iId, PP_RevisionType t,
> > const gchar* pzName, const gchar* pzValue );
> >
> > The same but do not replace an existing value
> >
> > void mergeAttrIfNotAlreadyThere(
> > UT_uint32 iId, PP_RevisionType t,
> > const gchar* pzName, const gchar* pzValue );
> >
> > Get the lowest deletion revision for content that might be deleted more
> > than once.
> >
> > const PP_Revision* getLowestDeletionRevision() const;
> >
> > I might also add a more STL like interface for PP_Revision to the
> > PP_RevisionAttr class.
> >
> > Something like:
> >
> > iterator begin();
> > iterator end();
> > iterator find( PP_Revision* );
> > iterator find( UT_uint32 iId,
> > PP_RevisionType t = PP_REVISION_NONE );
> >
> >
> >
> > [1]
> > http://lists.oasis-open.org/archives/office-collab/201106/msg00024.html
> >
> >
>