Title: Word Counter and some secrets.
Question: Making fast word counter.
Answer:
In this article my purpose is not only to show you
how to make word counter because it is a trivial task.
I want to show you that knowing what Delphi is doing
at the backstage is important.
Here is a code I write long, long time ago.
It is the execute method of a word counting thread:
procedure TCounter.Execute;
var
WEnd: Boolean;
Ch: PChar;
A: Integer;
begin
if not WordsInfo.Calculated then
begin
WordsInfo.Words := 0;
WordsInfo.Lett := 0;
WEnd := True;
for A := 1 to Length(UniMain.UniRE.Text) do
begin
if Terminated then Exit;
Ch := PChar(Copy(UniMain.UniRE.Text, A, 1));
if ((Ch 'A') and (Ch 'a') and (Ch begin
Inc(WordsInfo.Lett);
if WEnd = True then
begin
WEnd := False;
Inc(WordsInfo.Words)
end;
end
else
WEnd := True;
end;
WordsInfo.Calculated := True;
end;
end;
It is VERY BADLY WRITTEN. Why? I just sat down
and began writting. The result was extremly slow
program. UniMain is my form's name. UniRE is the
name of the rich edit control. And UniMain.UniRE.Text
is a string representation of the memo data. But
the part I was using it was one of the bottlenecks.
To give you a string Delphi makes some convertions.
so I added this:
var
...
TheText: string;
begin
TheText := UniMain.UniRE.Text;
...
and replaced all UniMain.UniRE.Text with TheText.
Making the conversion just once and not in a cycle
speeded the counting over hundred times! You can try
it if you don't believe me. But that is not the only
optimisation I did. Some people may wonder why I
used the line
Ch := PChar(Copy(UniMain.UniRE.Text, A, 1));
but this is the way I was working with strings in
Turbo Pascal. I Replaced this line with
Ch := UniMain.UniRE.Text[A];
and changed Ch from PChar to Char.
This also speeded the program over hundred times!!!
This was enough. But there is still a place for
optimization. I replaced
if ((Ch 'A') and (Ch 'a') and (Ch
with
if ((Ch in ['a'..'z']) or (Ch in ['A'..'Z'])) then
Using "Ch in ['a'..'z']" is theoretical faster than
the previous. I swapped the places of the lowercase
and uppercase check because it is more likely to
encounter lowercase and the check for uppercase letters
will not be perfomed. This is a small optimization
compared to the previous two but is good to be done.
May be there are others optimization I could make
but this code is quick enough - on my computer it
counts 1MB text file for about a second.
Here is the final code:
procedure TCounter.Execute;
var
WEnd: Boolean;
Ch: Char;
A: Integer;
TheText: string;
begin
TheText := UniMain.UniRE.Text;
if not WordsInfo.Calculated then
begin
WordsInfo.Words := 0;
WordsInfo.Lett := 0;
WEnd := True;
for A := 1 to Length(TheText) do
begin
if Terminated then Exit;
Ch := TheText[A];
if ((Ch in ['a'..'z']) or (Ch in ['A'..'Z'])) then
begin
Inc(WordsInfo.Lett);
if WEnd = True then
begin
WEnd := False;
Inc(WordsInfo.Words)
end;
end
else
WEnd := True;
end;
WordsInfo.Calculated := True;
end;
end;