• Matthias Becker (53/53) Nov 17 2003 Here are some ideas, I'm not realy sure about:
• Walter (28/78) Nov 18 2003 types
• Helmut Leitner (5/8) Nov 18 2003 Shared.
• Hauke Duden (3/7) Nov 19 2003 Is it just me, or are the discussions in this newsgroup going downhill
• Charles Sanders (4/11) Nov 19 2003 A couple of strippers is just what this newsgroup needs ;).

Matthias Becker <Matthias_member pathlink.com> writes:

Here are some ideas, I'm not realy sure about:

Everybody starts the name of a type with an uppercase letter. Only D's types
start with a lowercase letter. Perhaps they should start with an upper case
letter as well: Int, UInt, Char, ... .


------------------------------------------------------------------

An modifier no_side_effect micht be helpfull for optimisations. Eg, if you write
the following:

result = (x + y) * (x + y);

where everything is a suerdefined type, the operator + has to be executed twice,
as it could have side effects. With this modifier, the compiler knows that the
function (in this case the operator +) has no side effect and can use the result
of the operation twice. This is only true if the function is deterministic (it
has the same reult if the arguments are the same). So you would eigther need
another attribut "deterministic" or "no_side_effect" would guarantie that the
function is deterministc as well.

If you have the code, the Compiler might know on it's own, whether a function
has side effects or not, but if it's already compiled code, e.g. part of an
external library, the compiler has no chance to find it out. And it might be
(too?) hard to find out whether a function has side effects for the compiler.


------------------------------------------------------------------

An then I want to ask a question to Walter: Youintroduced a new Cast to D that
is realy cool, but you only use C-style casts on your own. (I look in the fildes
that came with the D-compiler). Why?

And I have another question:
As I mentioned I looked at some code of the standard-modules. In conv.d I fould
the following (from toInt):

{
if (cast(uint)v > 0x80000000)
goto Loverflow;
v = -v;
}

There are many constructs like this.

if the condition is true you jump to Loverflow (cool name :) ). There
conv_overflow() is called. And if you look at this function you see the only
thing it does is throwing an exception.

I don't get it. Instead of this strange goto you could directly call the
function or even throw directly. It seems not to have any sense.

{
if (cast(uint)v > 0x80000000)
conv_overflow(s);
v = -v;
}

This code would be as informative and wouldn't use an additional goto. goto
should only be used if there is a real reason for it not just for fun. But
perhaps I missed something.

It's the same with  goto Lerr. There only conv_error() is called which directly
throws.

I would replace 

goto Loverflow;

with

throw new ConvOverflowError(s);

or if it's too long in your opinion with

conv_overflow(s);

"Walter" <walter digitalmars.com> writes:

"Matthias Becker" <Matthias_member pathlink.com> wrote in message
news:bpb0i3$24nn$1 digitaldaemon.com...
 Everybody starts the name of a type with an uppercase letter. Only D's

types
 start with a lowercase letter. Perhaps they should start with an upper

case
 letter as well: Int, UInt, Char, ... .

D follows the C/C++ tradition of all keywords being lower case.

 An modifier no_side_effect micht be helpfull for optimisations. Eg, if you

write
 the following:

 result = (x + y) * (x + y);

 where everything is a suerdefined type, the operator + has to be executed

twice,
 as it could have side effects. With this modifier, the compiler knows that

the
 function (in this case the operator +) has no side effect and can use the

result
 of the operation twice. This is only true if the function is deterministic

(it
 has the same reult if the arguments are the same). So you would eigther

need
 another attribut "deterministic" or "no_side_effect" would guarantie that

the
 function is deterministc as well.

 If you have the code, the Compiler might know on it's own, whether a

function
 has side effects or not, but if it's already compiled code, e.g. part of

an
 external library, the compiler has no chance to find it out. And it might

be
 (too?) hard to find out whether a function has side effects for the

compiler.

A D compiler usually has far more information about functions in different
files than C/C++ compilers do. However, you are right, and some way to say
that a function has no side effects is a useful property for optimization.


 An then I want to ask a question to Walter: Youintroduced a new Cast to D

that
 is realy cool, but you only use C-style casts on your own. (I look in the

fildes
 that came with the D-compiler). Why?

Sloth. <g>

 And I have another question:
 As I mentioned I looked at some code of the standard-modules. In conv.d I

fould
 the following (from toInt):

 {
 if (cast(uint)v > 0x80000000)
 goto Loverflow;
 v = -v;
 }

 There are many constructs like this.

 if the condition is true you jump to Loverflow (cool name :) ). There
 conv_overflow() is called. And if you look at this function you see the

only
 thing it does is throwing an exception.

 I don't get it. Instead of this strange goto you could directly call the
 function or even throw directly. It seems not to have any sense.

 {
 if (cast(uint)v > 0x80000000)
 conv_overflow(s);
 v = -v;
 }

 This code would be as informative and wouldn't use an additional goto.

goto
 should only be used if there is a real reason for it not just for fun. But
 perhaps I missed something.

 It's the same with  goto Lerr. There only conv_error() is called which

directly
 throws.

 I would replace

 goto Loverflow;

 with

 throw new ConvOverflowError(s);

 or if it's too long in your opinion with

 conv_overflow(s);

It's a habit of mine to collect all the errors into one spot using a goto.
It's not right or wrong, just a personal style.

Helmut Leitner <helmut.leitner wikiservice.at> writes:

Walter wrote:
 
 It's a habit of mine to collect all the errors into one spot using a goto.
 It's not right or wrong, just a personal style.

Shared.

-- 
Helmut Leitner    leitner hls.via.at
Graz, Austria   www.hls-software.com

Hauke Duden <H.NS.Duden gmx.net> writes:

I would replace

goto Loverflow;


 I agree with you - a stripper is the way to go.

Is it just me, or are the discussions in this newsgroup going downhill 
somewhat? ;)


Hauke

"Charles Sanders" <sanders-consulting comcast.net> writes:

A couple of strippers is just what this newsgroup needs ;).

C

"Hauke Duden" <H.NS.Duden gmx.net> wrote in message
news:bpfdbf$2ihe$1 digitaldaemon.com...
I would replace

goto Loverflow;


  > I agree with you - a stripper is the way to go.

 Is it just me, or are the discussions in this newsgroup going downhill
 somewhat? ;)


 Hauke