NetNews Offline 2/NetNews Offline Volume 2.iso/news/comp/sys/amiga/programmer/824

Internet Message Format | 1996-08-05 | 2.9 KB

Path: newsserver.rrzn.uni-hannover.de!tubsibr!duening
From: [e-mail] (Lars Duening)
Newsgroups: comp.sys.amiga.programmer
Subject: Re: PPC compilers
Date: 11 Jan 1996 22:36:09 GMT
Organization: TU Braunschweig, Informatik (Bueltenweg), Germany
Distribution: world
Message-ID: <4d43cq$[e-mail]>
References: <[e-mail]> <4b77tq$[e-mail]> <MQAQx*[e-mail]> <4bqhnf$[e-mail]> <[e-mail]> <4c9i2l$[e-mail]> <4cf0ep$233@ra.i <4ck37m$[e-mail]> <4cs8ce$[e-mail]> <4cujve$[e-mail]>
Reply-To: [e-mail] (Lars Duening)
NNTP-Posting-Host: kastor.ibr.cs.tu-bs.de

In article <4cujve$[e-mail]> [e-mail] (Stephan Schaem) writes:
>Lars Duening ([e-mail]) wrote:
>: I need to know if it's a scalar type (further subdivided into integer
>: and floating types), an array, a structure, a pointer or else, but I
>: do not need to know it's exakt size in byte when using it.
>
> You need to know the type range... if you choose to ignore this, you
> will have to select the largest type avail for real/interger.

I still don't need to know or care about the _implementation_ and the
_size_ - and that's what this discussion was about.
Nobody denies that knowing the other type properties is still
necessary (more or less).

> But the point really is you need to know and keep track what your
> variable are. basicly in asm you write what you are thinking,
> but with a limited range of type and no typedef offered.
>
>: Sometimes I don't even need to know the exact type 'cause the compiler
>: will create conversion code plus overflow checking (I'm not talking
>: just C here).
>
> I think thats a bad practice... just for the fact that you let the
> compiler tell you what you assumed wrong.

I usually try to outsmart the compiler, but should I fail it's nice
to know that the program won't go gaga.
And look at this:

  a := SHORT(b);

The compiler is instructed here to create conversion code to assign b
to a. I still don't need to know if it's a longword->word, word->byte or
even quadword->longword conversion. I _do_ know that I convert a
scalar type to a smaller value range.

>: > Otherwise you need to know that color0 is a word holding 12bit
>: > of valid data grouped in 4bit nibles,
>
>: typedef struct {
>:   int r : 4;
>:   int g : 4;
>:   int b : 4;
>: } color0_t;
>
>: color0_t color0;
>
> Now somewhere one should read the range of r,g,b... or you must
> also provide a set of routine that will work on this structure/type
> (Example: only value from 0 to 12 are valid. But you will make your
> user function prototype use float and define the valid range)

This is a completely different problem, and even that could be solved:

TYPE COLOR0 = PACKED RECORD
  r, g, b : INTEGER [0..12];
END;

VAR color0 : COLOR0;
--
Lars Duening; [e-mail]