[Mono-list] Why is 0.1f + 0.2f - 0.3f produces such a large error number?

[MarLOne]

Thanks for the reply. The following disclosure may not have anything to do
with explaining this but it highlights some inconsistency in .Net:

I know Mono is not exactly VS (well very close as Xamarin is now part of
Microsoft) but the following comparison is still valid and interesting:
VS2003, 2005, 2010, 2012 all returning a delta value of -7.450581E-09

Anyone with a VS2013 can help by reporting what it produces.

*VS2015 (regardless Update 1 or 2) returning 0*.

Also regardless on what machine and processor type (even on mac), Swift,
Java, and C all returning 0 as expected. It seems from day one the C#
compiler is producing this same number until it comes to VS2015.

I also had a look at the IL code generated and the compiler compiles the
constant expression so the IL code does not show any arithmetic expression.
I guess we can blame the compiler. But I rewrite the code in such a way the
compiler cannot compile away the value at compile time. Yet I am getting the
same non-zero delta. I guess Microsoft and Mono can claim that is literally
zero because float according to ECMA C# standard is only good to 7 digits.
But that does not explain why the other compiler/languages have so much more
accuracy. 

Does this mean one should not use C# when performing scientific
calculations? The above information is interesting.

If you construct a DLL or COM component using C/C++, the return value (by
P/Invoke or COM call) will be zero. If you then enhance the DLL by rewriting
it in a .Net component, the return value to a .net client is -7.450581E-09!!
The same client but component developed with different languages. Is this a
good language to deal with integration?

May be that's why VS2015 took the decision to fix it. When will a mono
compiler matching VS2015's?

Leon



--
View this message in context: http://mono.1490590.n4.nabble.com/Why-is-0-1f-0-2f-0-3f-produces-such-a-large-error-number-tp4667837p4667848.html
Sent from the Mono - General mailing list archive at Nabble.com.