On Wed, 1 Mar 2006 21:35:54 +0100 (Romance Standard Time)
Michael Van Canneyt <[EMAIL PROTECTED]> wrote:
>
>
> On Wed, 1 Mar 2006, L505 wrote:
>
> > >
> > > On Wed, 1 Mar 2006, L505 wrote:
> > >
> > > > > > Widestrings need to be like that for compatibility with c and
> > > > > > c++
> > > >
> > > > > Not exactly, they need to be like this because the dang MS COM.
> > > > > However I can't see why is this different than the AnsiString vs
> > > > > Windows PChar-based API...
> > > >
> > > >
> > > > I'm off topic here but:
> > > > I think there should be a standard reference counting mechanism
> > > > (standards comittee?) between language. We have the standard ansi
> > > > string header
> > embedded in
> > > > the ansistring but no standard reference counting mechanism. If this
> > > > was the case we could share our ansi strings in DLL's with C++ and
> > > > other languages.
> > >
> > > Ansistrings are unique to Object Pascal. The implementation of strings
> > > in C++ is completely different.
> > >
> > > > However There are good reasons for using pchars especially during
> > > > string concatenations - ansistrings are dead slow when you are
> > > > making tons of small concatenations.
> > >
> > > Sorry, but that is total nonsense. Not more than 'naive' pchars.
> > >
> >
> > My strloadfile benchmark proves that the standard programmer using
> > simple concatenations with ansistrings will have a serious speed problem
> > in his application.
>
> Exactly my point. The C programmer just already allocates enough
> memory to start with. If he would do a getmem/freemem/strcat each time,
> his application would be just as slow as a Pascal program.
>
> Only as a C programmer he is used to this kind of thinking...
>
> Pascal has the best of both worlds: you can use the 'simple' way,
> for fast coding, but you can also use the 'complicated' but faster, way.
>
> Usually I start with a 'simple' straightforward implementation
> for fast coding. and when speed really becomes an issue, I change
> it to the complicated way...
>
> >
> > Unless you have in depth knowledge about how every bit of ansistring
> > memory management works, using getmem and freemem and pchars are
> > simpler, no? Because you are not battling the automatic memory
> > management. Now, if you are an expert in battling the automatic memory
> > management maybe you can battle ansistrings enough to be almost or just
> > as fast as pchars. Is this what we are doing in the fpc sources? I will
> > have to look, and see.
>
> Yes.
For example: The codetools use ansistrings. Less than 1% use PChars.
Just avoid the + operator as you would avoid ReAllocMem and StrCopy.
> > Does it require some simple setlength calls or is it more complex than
> > that?
>
> Simple setlength will do.
>
> > Does it require less or more lines of code than simply using getmem and
> > freemem?
>
> Less. The setlength takes care of it for you. That is the whole point.
>
> > I bet you end up inserting lots more lines of code to battle the memory
> > manager instead of just manually taking control of it yourself.
>
> No.
> The setlength also provides automatic correct terminating and so on,
> plus you always have the length stored correctly. Not so with
> getmem/freemem.
Mattias
_________________________________________________________________
To unsubscribe: mail [EMAIL PROTECTED] with
"unsubscribe" as the Subject
archives at http://www.lazarus.freepascal.org/mailarchives
