C# 2.0 Generics and collections

[Michael Giagnocavo [MVP]]

Re: C# 2.0 Generics and collections

For ValueTypes, a specialized type class is created. For RefTypes, only one
is created and it's reused. This solves the boxing issue (preliminary
results tell me it's at least 5 times faster in a tight loop that woult
otherwise box and unbox).

-mike
MVP

"Jon Skeet [C# MVP]" <skeet@pobox.co m> wrote in message
news:MPG.1a1b18 2f87021faa989a8 0@msnews.micros oft.com...[color=blue]
> Keith Patrick <richard_keith_ patrick@hotmail .com> wrote:[color=green]
> > A generic wouldn't solve the problem of the speed of boxing, would it?[/color][/color]
I[color=blue][color=green]
> > haven't read about how they are implemented, but I would think that[/color][/color]
integers[color=blue][color=green]
> > would still have to be treated a objects by the generic.[/color]
>
> I don't believe so. I believe each generic class is compiled into a
> type-specific class when it's first used, and that can take advantage
> of anything type-specific such as size. (Note that this *isn't* the way
> generics will be working in Java, AFAIK.)
>
> --
> Jon Skeet - <skeet@pobox.co m>
> http://www.pobox.com/~skeet
> If replying to the group, please do not mail me too[/color]

[Michael Giagnocavo [MVP]]

Re: C# 2.0 Generics and collections

Jon's answer is about Generics in general. The .NET implementation will
eliminate boxing (support at the IL level). The Java implementation doesn't
get around this.
-mike
MVP

"Joe Mayo" <jmayo@ddiieess ppaammeerrssddi iee.com> wrote in message
news:OMBh8bHqDH A.488@tk2msftng p13.phx.gbl...[color=blue]
> "Keith Patrick" <richard_keith_ patrick@hotmail .com> wrote in message
> news:ecZhcwGqDH A.1948@TK2MSFTN GP12.phx.gbl...[color=green]
> > A generic wouldn't solve the problem of the speed of boxing, would it?[/color][/color]
I[color=blue][color=green]
> > haven't read about how they are implemented, but I would think that[/color]
> integers[color=green]
> > would still have to be treated a objects by the generic.[/color]
>
> Hi Keith,
>
> In the September MSDN, Jason Clark writes about Generics. In the
> Performance paragraph he states "This is because boxing of the values is
> avoided completely.", but you should read the whole paragraph for context:
>
> http://msdn.microsoft.com/msdnmag/issues/03/09/NET/
>
> There is also a follow-up article in the October edition that talks more
> about the implementation:
>
> http://msdn.microsoft.com/msdnmag/issues/03/10/NET/
>
> Joe
> --
> http://www.csharp-station.com
>
>[/color]

[Gary van der Merwe]

Re: C# 2.0 Generics and collections

Hi

Dose the runtime casting from object -> string not cause performence
problems?

Would this improve with the use of generics?

Gary

"Michael Giagnocavo [MVP]" <mggUNSPAM@Atre vido.net> wrote in message
news:uWbY5zHqDH A.2964@tk2msftn gp13.phx.gbl...[color=blue]
> StringCollectio n doesn't have any boxing going on, but the other
> object-based ones would suffer.
> -mike
> MVP
>
> "Gary van der Merwe" <garyvdm@hotmai l.com> wrote in message
> news:OS74QoFqDH A.2440@TK2MSFTN GP10.phx.gbl...[color=green]
> > Hi
> >
> > I think the refrence to bugs was regarding the fact that the old[/color]
> collections[color=green]
> > are slow because you have to continualy box objects.
> >
> > Gary
> >
> >
> > "Christoph Nahr" <christoph.nahr @kynosarges.de> wrote in message
> > news:ekp1rvsv0e h59m6r8lm3aann0 vhseml2vt@4ax.c om...[color=darkred]
> > > On Tue, 11 Nov 2003 14:26:57 +0100, "Thomas Tomiczek \(MVP\)"
> > > <t.tomiczek@tho na-consulting.com> wrote:
> > >
> > > >SADLY (so to say) the old non-generic collections are still there.
> > > >
> > > >Which just keeps, porapbly, existing bugs around.
> > >
> > > Well, I haven't noticed any bugs in StringCollectio n so far, and
> > > besides they could re-implement it as a pre-created generic collection
> > > in .NET 2.0, couldn't they?
> > > --
> > > http://www.kynosarges.de[/color]
> >
> >[/color]
>
>
>[/color]

[Christoph Nahr]

Re: C# 2.0 Generics and collections

On Tue, 11 Nov 2003 20:16:28 +0200, "Gary van der Merwe"
<garyvdm@hotmai l.com> wrote:
[color=blue]
>Dose the runtime casting from object -> string not cause performence
>problems?[/color]

No. String is a reference type, so there's no boxing/unboxing going
on, so the performance penalty will be negligible. (At least compared
to what you'd have to pay for value types!)
--

http://www.kynosarges.de

[Michael Giagnocavo [MVP]]

Re: C# 2.0 Generics and collections

Right, storing a reftype in an object costs nothing (whereas a boxing costs
quite a bit). A downcast costs about 4 times less than a box.
-mike
MVP

"Christoph Nahr" <christoph.nahr @kynosarges.de> wrote in message
news:1pl2rv0blq cir7p48mvf7d0qo bohr1892o@4ax.c om...[color=blue]
> On Tue, 11 Nov 2003 20:16:28 +0200, "Gary van der Merwe"
> <garyvdm@hotmai l.com> wrote:
>[color=green]
> >Dose the runtime casting from object -> string not cause performence
> >problems?[/color]
>
> No. String is a reference type, so there's no boxing/unboxing going
> on, so the performance penalty will be negligible. (At least compared
> to what you'd have to pay for value types!)
> --
> http://www.kynosarges.de[/color]

[Eric Gunnerson [MS]]

Re: C# 2.0 Generics and collections

A bit of a clarification.

In the Java implementation, the casting will be removed from the user code,
but it will still be present at the bytecode level.
--
Eric Gunnerson

Visit the C# product team at http://www.csharp.net
Eric's blog is at http://blogs.gotdotnet.com/ericgu/

This posting is provided "AS IS" with no warranties, and confers no rights.
"Jon Skeet [C# MVP]" <skeet@pobox.co m> wrote in message
news:MPG.1a1b26 0360d9b930989a8 2@msnews.micros oft.com...[color=blue]
> Gary van der Merwe <garyvdm@hotmai l.com> wrote:[color=green]
> > No. Thats the whole point of them. With Generics, the type checking[/color][/color]
happens[color=blue][color=green]
> > when you complile, not at runtime.[/color]
>
> That doesn't *necessarily* get rid of boxing though. For instance, in
> the Java implementation of generics, the casting will be (largely)
> removed, but boxing won't be.
>
> --
> Jon Skeet - <skeet@pobox.co m>
> http://www.pobox.com/~skeet
> If replying to the group, please do not mail me too[/color]

[Eric Gunnerson [MS]]

Re: C# 2.0 Generics and collections

Anders did some timings on this for his PDC presentation (which we're
working to get up on the C# dev center on MSDN).

IIRC, you save around 10% of the loop overhead in the reference type case
(by avoiding the cast), and something like 50% on the value type case. This
was for essentially empty loops.

--
Eric Gunnerson

Visit the C# product team at http://www.csharp.net
Eric's blog is at http://blogs.gotdotnet.com/ericgu/

This posting is provided "AS IS" with no warranties, and confers no rights.
"Michael Giagnocavo [MVP]" <mggUNSPAM@Atre vido.net> wrote in message
news:%23$rPGXKq DHA.1488@TK2MSF TNGP12.phx.gbl. ..[color=blue]
> Right, storing a reftype in an object costs nothing (whereas a boxing[/color]
costs[color=blue]
> quite a bit). A downcast costs about 4 times less than a box.
> -mike
> MVP
>
> "Christoph Nahr" <christoph.nahr @kynosarges.de> wrote in message
> news:1pl2rv0blq cir7p48mvf7d0qo bohr1892o@4ax.c om...[color=green]
> > On Tue, 11 Nov 2003 20:16:28 +0200, "Gary van der Merwe"
> > <garyvdm@hotmai l.com> wrote:
> >[color=darkred]
> > >Dose the runtime casting from object -> string not cause performence
> > >problems?[/color]
> >
> > No. String is a reference type, so there's no boxing/unboxing going
> > on, so the performance penalty will be negligible. (At least compared
> > to what you'd have to pay for value types!)
> > --
> > http://www.kynosarges.de[/color]
>
>[/color]

[Eric Gunnerson [MS]]

Re: C# 2.0 Generics and collections

We certainly will not get rid of the existing types, as lots of programs
depend upon them.

We had talked about auto-migration, but it's somewhat unpalatable as we've
taken the opportunity to fix some problems with the ArrayList and Hashtable
classes, but the fixes might break existing code in some rare cases. In the
generics world, you'll use List<T> and Dictionary<K, V> instead.

--
Eric Gunnerson

Visit the C# product team at http://www.csharp.net
Eric's blog is at http://blogs.gotdotnet.com/ericgu/

This posting is provided "AS IS" with no warranties, and confers no rights.
"Gary van der Merwe" <garyvdm@hotmai l.com> wrote in message
news:%23cYB0YFq DHA.1496@TK2MSF TNGP11.phx.gbl. ..[color=blue]
> Hi
>
> When C# 2.0 arrives System.Collecti ons.Specialized .StringCollecti on will
> become "obsolete". Will the framework still contain this class, or will
> there be a C# 1.X to 2.0 conversion utility that will change
> System.Collecti ons.Specialized .StringCollecti on to
> System.Collecti ons.ArrayList <string> and ArrayList to ArrayList<objec t>
> ??
>
> Gary
>
>[/color]

[Michael Giagnocavo [MVP]]

Re: C# 2.0 Generics and collections

I'd love to see his benchmarks. I did mine by adding an item on each
iteration (requiring a box in, or for a reftype, just passing the ref) and
then taking it back out.

The generics version for the reftypes was barely noticable (2**24
iterations, something like 0.02 seconds speed up). Of course, this was
actually doing some real work. I timed the actual cast down to about 3ns
(on a P4c 3GHz), so I'm saving that each time.

For valuetypes, it was a whole different story, the memory requirements got
so high they slowed it down immensely (2**24 objects, since each int was
boxed).

So for "real world" apps, I'd imagine we'll see great (runtime performance)
benefits for valuetypes and an almost unnoticeable improvement for
reftypes...

-mike
MVP

"Eric Gunnerson [MS]" <ericgu@online. microsoft.com> wrote in message
news:%23lFMqQLq DHA.2808@TK2MSF TNGP10.phx.gbl. ..[color=blue]
> Anders did some timings on this for his PDC presentation (which we're
> working to get up on the C# dev center on MSDN).
>
> IIRC, you save around 10% of the loop overhead in the reference type case
> (by avoiding the cast), and something like 50% on the value type case.[/color]
This[color=blue]
> was for essentially empty loops.
>
> --
> Eric Gunnerson
>
> Visit the C# product team at http://www.csharp.net
> Eric's blog is at http://blogs.gotdotnet.com/ericgu/
>
> This posting is provided "AS IS" with no warranties, and confers no[/color]
rights.[color=blue]
> "Michael Giagnocavo [MVP]" <mggUNSPAM@Atre vido.net> wrote in message
> news:%23$rPGXKq DHA.1488@TK2MSF TNGP12.phx.gbl. ..[color=green]
> > Right, storing a reftype in an object costs nothing (whereas a boxing[/color]
> costs[color=green]
> > quite a bit). A downcast costs about 4 times less than a box.
> > -mike
> > MVP
> >
> > "Christoph Nahr" <christoph.nahr @kynosarges.de> wrote in message
> > news:1pl2rv0blq cir7p48mvf7d0qo bohr1892o@4ax.c om...[color=darkred]
> > > On Tue, 11 Nov 2003 20:16:28 +0200, "Gary van der Merwe"
> > > <garyvdm@hotmai l.com> wrote:
> > >
> > > >Dose the runtime casting from object -> string not cause performence
> > > >problems?
> > >
> > > No. String is a reference type, so there's no boxing/unboxing going
> > > on, so the performance penalty will be negligible. (At least compared
> > > to what you'd have to pay for value types!)
> > > --
> > > http://www.kynosarges.de[/color]
> >
> >[/color]
>
>[/color]

[Willy Denoyette [MVP]]

Re: C# 2.0 Generics and collections

Michael,

Mind to share your code? My observations are on par with Anders timings.

Willy.

"Michael Giagnocavo [MVP]" <mggUNSPAM@Atre vido.net> wrote in message news:uJuy9RNqDH A.1928@TK2MSFTN GP12.phx.gbl...[color=blue]
> I'd love to see his benchmarks. I did mine by adding an item on each
> iteration (requiring a box in, or for a reftype, just passing the ref) and
> then taking it back out.
>
> The generics version for the reftypes was barely noticable (2**24
> iterations, something like 0.02 seconds speed up). Of course, this was
> actually doing some real work. I timed the actual cast down to about 3ns
> (on a P4c 3GHz), so I'm saving that each time.
>
> For valuetypes, it was a whole different story, the memory requirements got
> so high they slowed it down immensely (2**24 objects, since each int was
> boxed).
>
> So for "real world" apps, I'd imagine we'll see great (runtime performance)
> benefits for valuetypes and an almost unnoticeable improvement for
> reftypes...
>
> -mike
> MVP
>
> "Eric Gunnerson [MS]" <ericgu@online. microsoft.com> wrote in message
> news:%23lFMqQLq DHA.2808@TK2MSF TNGP10.phx.gbl. ..[color=green]
> > Anders did some timings on this for his PDC presentation (which we're
> > working to get up on the C# dev center on MSDN).
> >
> > IIRC, you save around 10% of the loop overhead in the reference type case
> > (by avoiding the cast), and something like 50% on the value type case.[/color]
> This[color=green]
> > was for essentially empty loops.
> >
> > --
> > Eric Gunnerson
> >
> > Visit the C# product team at http://www.csharp.net
> > Eric's blog is at http://blogs.gotdotnet.com/ericgu/
> >
> > This posting is provided "AS IS" with no warranties, and confers no[/color]
> rights.[color=green]
> > "Michael Giagnocavo [MVP]" <mggUNSPAM@Atre vido.net> wrote in message
> > news:%23$rPGXKq DHA.1488@TK2MSF TNGP12.phx.gbl. ..[color=darkred]
> > > Right, storing a reftype in an object costs nothing (whereas a boxing[/color]
> > costs[color=darkred]
> > > quite a bit). A downcast costs about 4 times less than a box.
> > > -mike
> > > MVP
> > >
> > > "Christoph Nahr" <christoph.nahr @kynosarges.de> wrote in message
> > > news:1pl2rv0blq cir7p48mvf7d0qo bohr1892o@4ax.c om...
> > > > On Tue, 11 Nov 2003 20:16:28 +0200, "Gary van der Merwe"
> > > > <garyvdm@hotmai l.com> wrote:
> > > >
> > > > >Dose the runtime casting from object -> string not cause performence
> > > > >problems?
> > > >
> > > > No. String is a reference type, so there's no boxing/unboxing going
> > > > on, so the performance penalty will be negligible. (At least compared
> > > > to what you'd have to pay for value types!)
> > > > --
> > > > http://www.kynosarges.de
> > >
> > >[/color]
> >
> >[/color]
>
>[/color]