[Libreoffice] [PUSHED] Re: [PATCH] Fix for bug/feature request i#8288

[Mattias Johnsson]

On 2 December 2010 01:31, Lubos Lunak <l.lunak at suse.cz> wrote:
> On Wednesday 01 of December 2010, Jan Holesovsky wrote:
>> >  There appears to be a helper class for this, SwCrsrSaveState. Except
>> > that it
>> > doesn't seem to work, as far as I can judge SwCursor::RestoreState()
>> > doesn't
>> > really restore anything (quite hard to say, so much for "trivial APIs
>> > don't
>> > need docs").
>>
>> Yeah - I tried to use that one instead when testing the patch, but
>> noticed the same thing, so I just pushed the patch verbatim :-)  Do we
>> actually use SwCrsrSaveState somewhere (ie. is it worth fixing/removing
>> for good/anything else)?  [Actually, I wouldn't be surprised if there
>> was a copy somewhere else that works ;-)]
>
>  I've figured it out. SwCrsrSaveState only saves the state on the stack, the
> actual restoring is done by SwCursor::RestoreSavePos(). I'll change the code
> to use it.

Cool, I'll be interested to see how you go with this.

The first problem I ran into was that the place I added my code was
place furthest down the "Search/Replace" calls I could go without the
cursor position information being cleared and lost. At this stage
you're executing code inside the SwView class, and only have easy
access to pointers to SwWrtShell and SwCrsrShell objects, rather than
the SwShellCrsr object (this naming, incidentally, caused me several
hours of head banging until I figured out there were two
almost-identically named classes :-P ).

The reason this matters is that SwShellCrsr::RestoreSavePos() is a
public method in SwShellCrsr, but SwShellCrsr::SaveState () is a
*protected* method. Consequently I didn't see any easy method of
gaining access to the SaveState method to actually save the state. A
few calls further down the Search/Replace call stack you have access
to it, but by then the cursor info has been lost.

So I found that saving the cursor state needed a bit of contorting,
and then making sure later calls in the Search/Replace hierarchy
didn't totally blow away and reset the cursor stack where the position
was saved (which it seems to be really keen to do) required even more
code, and it was starting to look really ugly. Which is why I went
with the two-line solution.

Of course, having written that, I freely admit I'm still (somewhat
blindly) feeling my way around the LO code, and you'll probably find
some really elegant way to do it with a couple of lines of code, and
I'll look silly :-P