FW: Couple of things

[Daniel Leuck]

Hi Jesse,

 

The Mono issue appears to be a bug with the Mono compiler (those two
interface methods are not ambiguous), but I will look into what can be done.

 

Re: I would think that a 32-bit integer and a 64-bit floating point would be
enough.

 

The two options I would consider are:

1.	Make the 128 bit decimal type an optional feature.
2.	Redefine decimal as the most accurate decimal type available on the
platform.

 

I am leaning strongly towards one because I don't like the notion of
something having significantly different meaning on different platforms.
Any objections?

 

Dan

 

  _____  

From: Jesse Weaver [mailto:pianohacker at gmail.com] 
Sent: Sunday, November 12, 2006 12:35 PM
To: sdl-developers at ikayzo.org
Subject: Re: FW: Couple of things

 

 

I am not an expert on numerical analysis, and have not worked on SDL. I
wrote some code which used and compared GMP and OpenSSL BigNum libraries,
which is my only hook into this thread...

29 decimal digits of precision is equivalent to 96 binary digits of
precision. Therefore it seems reasonable to use GMP (or appropriate
equivalent) with precision set to 0x60.

However, the exponent in GMP is always a word--32 or 64 bits, where the
"decimal" type is claiming 32 bits (128 - 96). So GMP in this case would
only be equivalent on 32 bit systems.

GMP does not claim to conform to any specific floating point standard (much
less THE gold standard  IEEE754). Nor does GMP even claim that the same
operations will give the same results on architectures with different native
word sizes.

This might be a minor point if the implementation of SDL allows platforms to
do calculations to the best of their abilities, instead of with exact
equivalency. 

Do the Java and C# implementations give the same results when operating on
128 bit fp quantities?

One route could be to restrict SDL to floats and doubles, and make this 128
bit (C#) fp operation an extension. That way the C++ port gets off the
ground without having to worry about it :)

Eric

 

> Also, there is currently no portable way to implement the 128 bit decimal
type in SDL C++, that I

> know of, short of bundling the GNU MP library.

 

That's a tough one.  Do any other list members have suggestions?  Eric?  A
description of the C# decimal type is here:
http://msdn2.microsoft.com/en-US/library/364x0z75.aspx 




I tried to compile the SDL C++ port on mono, and it got rather angry:
Ikayzo/SDL/Tag.cs(345,7):   warning CS0278:
`System.Collections.Generic.IList<Ikayzo.SDL.Tag>' contains ambiguous
implementation of `enumerable' pattern. Method
`System.Collections.IEnumerable.GetEnumerator()' is ambiguous with method
`System.Collections.Generic.IEnumerable< Ikayzo.SDL.Tag>.GetEnumerator()'
Ikayzo/SDL/Tag.cs(431,13):   warning CS0278:
`System.Collections.Generic.IList<Ikayzo.SDL.Tag>' contains ambiguous
implementation of `enumerable' pattern. Method
`System.Collections.IEnumerable.GetEnumerator()' is ambiguous with method
`System.Collections.Generic.IEnumerable< Ikayzo.SDL.Tag>.GetEnumerator()'
Ikayzo/SDL/Tag.cs(554,7):   warning CS0278:
`System.Collections.Generic.IList<Ikayzo.SDL.Tag>' contains ambiguous
implementation of `enumerable' pattern. Method
`System.Collections.IEnumerable.GetEnumerator()' is ambiguous with method
`System.Collections.Generic.IEnumerable< Ikayzo.SDL.Tag>.GetEnumerator()'
Ikayzo/SDL/Tag.cs(168,17):   warning CS0278:
`System.Collections.Generic.IList<object>' contains ambiguous implementation
of `enumerable' pattern. Method
`System.Collections.IEnumerable.GetEnumerator()' is ambiguous with method
`System.Collections.Generic.IEnumerable<object>.GetEnumerator()' 
Ikayzo/SDL/Tag.cs(195,17):   warning CS0278:
`System.Collections.Generic.ICollection<string>' contains ambiguous
implementation of `enumerable' pattern. Method
`System.Collections.IEnumerable.GetEnumerator()' is ambiguous with method
`System.Collections.Generic.IEnumerable<string>.GetEnumerator()' 

Also, the openSSL bignum library doesn't seem to be distributed separately,
making that option difficult. Does anyone know if we really need as many
number types as we have? I would think that a 32-bit integer and a 64-bit
floating point would be enough. 

-- 
Jesse

-----------
There is a theory which states that if anybody ever discovers exactly what
the Universe is for and why it is here, it will instantly disappear and be
replaced by something even more bizarre and inexplicable. There is another
theory which states that this has already happened. - Douglas Adams, author
of Hitchhiker's Guide to the Galaxy 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://ml.ikayzo.org/pipermail/sdl-developers/attachments/20061113/00defab9/attachment.htm>