Can 'is' followed by cast by optimized away

[Thorsten Ottosen]

Hi,

Can the following ugly type-switching code by optimized
by the compiler/jitter?:

object o = ...;
if( o is Foo )
return (Foo)o;

Arguably, the cast is guranteed to succeed, but do the
compiler take advantage of that?

If yes, would it be possible to each that conclusion
by looking at the assembler output?

Thanks

-Thorsten

[Hans Kesting]

Thorsten Ottosen wrote:

Hi,

Can the following ugly type-switching code by optimized
by the compiler/jitter?:

object o = ...;
if( o is Foo )
return (Foo)o;

Arguably, the cast is guranteed to succeed, but do the
compiler take advantage of that?

If yes, would it be possible to each that conclusion
by looking at the assembler output?

Thanks

-Thorsten

see the "as" operator

return o as Foo;

if o is not a Foo, a null is returned
(works only on reference types, not on value types, as they can't be null)

Hans Kesting

[Thorsten Ottosen]

"Hans Kesting" <ne***********@spamgourmet.com> wrote in message
news:Oe**************@TK2MSFTNGP12.phx.gbl...

Thorsten Ottosen wrote:

Hi,

Can the following ugly type-switching code by optimized
by the compiler/jitter?:

object o = ...;
if( o is Foo )
return (Foo)o;

see the "as" operator

return o as Foo;

if o is not a Foo, a null is returned
(works only on reference types, not on value types, as they can't be null)

Thanks, but it does not quite answer my question. Since as is equivalent to

expression is type ? (type)expression : (type)null

my question remains unanswered: does this lead to essentially two casts (at
runtime) if the expression is really
of the type 'type'? (one cast to see if it is the right type, then a cast to
convert).

-Thorsten

[Christof Nordiek]

"Hans Kesting" <ne***********@spamgourmet.com> schrieb im Newsbeitrag
news:Oe**************@TK2MSFTNGP12.phx.gbl...

Thorsten Ottosen wrote:

Hi,

Can the following ugly type-switching code by optimized
by the compiler/jitter?:

object o = ...;
if( o is Foo )
return (Foo)o;

Arguably, the cast is guranteed to succeed, but do the
compiler take advantage of that?

If yes, would it be possible to each that conclusion
by looking at the assembler output?

Thanks

-Thorsten

see the "as" operator

return o as Foo;

if o is not a Foo, a null is returned
(works only on reference types, not on value types, as they can't be null)

Hans Kesting

No, it's not really the same.
Thorsten didn't say, what he want's to do in the else part.

Christof

[Daniel Jin]

> my question remains unanswered: does this lead to essentially two casts (at

runtime) if the expression is really
of the type 'type'? (one cast to see if it is the right type, then a cast to
convert).
you are correct that your original example requires essentially two casts.
and I don't believe there is optimization of any kind to eliminate the extra
cast.
Thanks, but it does not quite answer my question. Since as is equivalent to

expression is type ? (type)expression : (type)null
no it's not. 'is' operator and 'as' operator performs the exact same IL
instruction, with the only difference being, 'as' will generate an extra
instruction to save the result of the cast. which is precisely the
optimization, so to speak, that you seek.

-Thorsten

[Thorsten Ottosen]

"Daniel Jin" <Da*******@discussions.microsoft.com> wrote in message
news:C2**********************************@microsof t.com...

Thanks, but it does not quite answer my question. Since as is equivalent
to

expression is type ? (type)expression : (type)null
no it's not.

ok, but I copied this from the .Net reference documentation :-)
'is' operator and 'as' operator performs the exact same IL
instruction, with the only difference being, 'as' will generate an extra
instruction to save the result of the cast. which is precisely the
optimization, so to speak, that you seek.

ok, thanks.

-Thorsten