Strange behavior using TPointSeries.Clear - delphi

Question

(Delphi 6 with TChart, Win XP)
I'm getting erratic behavior trying to clear the points from a point series, and of course, this code used to work.
Basically, this part of my program generates 5 data points and plots them. When I try to clear them using OSC_Series.Clear I get a "List index out of bounds [0]" error.
I checked to make sure there was nothing odd about the values I was plotting. All is good there. Then I tried different things to try to isolate and work around the problem.
Here's some code.
type
TksGraph_DataFrm = class(TForm)
.
.
.
private
OSC_Series: TPointSeries
public
end;
procedure TksGraph_DataFrm.cat7AnalysisInitialize(var P:TTest_Project);
begin
// Do a bunch of stuff.
// Set up the analysis data points series.
OSC_Series:=TPointSeries.Create(self);
AnalysisChart.AddSeries(OSC_Series);
with OSC_Series do
begin
Title:='';
HorizAxis:=aBothHorizAxis;
VertAxis:=aBothVertAxis;
SeriesColor:=clRed;
Pointer.Brush.Color:=clYellow;
Pointer.HorizSize:=4;
Pointer.VertSize:=4;
Pointer.Style:=psRectangle;
Pointer.Visible:=true;
LinePen.Color:=clBlack;
LinePen.Width:=1;
Linepen.Visible:=true;
ShowInLegend:=false;
XValues.Order:=LoNone;
end;
end;
procedure TksGraph_DataFrm.cat7AnalysisRefresh(var P:TTest_Project);
var X,Y:single;
begin
X:= some value
Y:= some value
// Plot the result.
OSC_Series.AddXY(X,Y);
showmessage(
'Count = '+inttostr(OSC_Series.Count)+#13+
'X = '+FloatToStr(X)+#13+
'Y = '+FloatToStr(Y)+#13+
'Plot-X = '+FloatToStr(OSC_Series.XValue[OSC_Series.Count-1])+#13+
'Plot-Y = '+FloatToStr(OSC_Series.YValue[OSC_Series.Count-1]));
end;
Here is the routine I to use to reset the series. I'm including code that does and does not work.
procedure TksGraph_DataFrm.cat7AnalysisClear(var P:TTest_Project);
var i:integer;
begin
// This should work, but it gives me the list out of bounds error
// unless the count is 0.
OSC_Series.Clear;
// This does not work, for obvious reasons. I get a "list out of
// bounds [3] for this one.
for i:=0 to OSC_Series.Count - 1 do OSC_Series.Delete[0];
// It seems this should work, but again I get the out of bounds [0]
// error.
repeat
OSC_Series.Delete(0);
until OSC_Series.Count = 0;
// This works. Don't ask me why.
showmessage('A - '+inttostr(OSC_Series.Count));
OSC_Series.Clear;
showmessage('B - '+inttostr(OSC_Series.Count));
// This also works.
sleep(2000);
OSC_Series.Clear;
// This does not work.
sleep(1000);
OSC_Series.Clear;
end;
I'm stumped, obviously.

Answer 1

This smells like the code is working with an object (OSC_Series) which has been destroyed and the memory then re-used for something else. When you then use the stale reference to that memory you get unexpected and unpredictable results.
Where is OSC_Series free'd ?
I would check all such places and make sure that you do not attempt to use the OSC_Series reference after it has been free'd.
Note also that since the series is owned by the form it could be that the form itself is contriving to executing code in events after it has destroyed its owned components (including this series).

Answer 2

OK, dumb and not dumb.
I experimented with where I put the showmessage statement. I found I could only avoid the error if that statement came after the OSC_Series.Clear statement. I kept moving that statement back until it was after the call to the AnalysisRefresh routine, which is in a button's OnClick event handler. This means that none of the code in the refresh, enable, or update routines was causing this.
Stepping back a bit, if the AnalysisRefresh routine fails the user is shown a message. After that box is closed OSC_Series.Clear is called. If I close the box by pressing SPACE or ENTER on the keyboard... no error. If I use the mouse, error.
The chart behind the mouse has an OnMouseMove event where I display the mouse position on a status bar. I also display a hint if the mouse is near a plotted point. After clicking the message box with a mouse to close it the OnMouseMove event is called and by the time it gets to where it can display the hint, the plotted point is gone, and... error.
So, it seemed like an almost random error, but it wasn't. The trigger was just somewhere else entirely. That's you my try/except block wasn't catching the error. EurekaLog was catching it, but in a different procedure far, far away. (Deltics' answer was pretty close.)
Thanks for your help and suggestions.
Dang it if some days I can push hundreds of lines of code with no problems, then something like this pops up and it costs me near two days.