Let's keep improving the parser... about neslib.multiprecision

[Andreas113]

All such "stupid" inputs containing spaces can be easily solved by removing all spaces and making the program more secure. The following routine is very suitable for this purpose:

{
  Deletes all occurrences of sub-String Needle from String HayStack AND returns
  the resulting String. HayStack is returned unchanged If it does NOT contain
  Needle. The search is Case sensitive.
}
Function DelAllStr(CONST Needle, Haystack: String): String;
// http://delphidabbler.com/url/csdb.
Var
  StartIdx: Integer;
Begin
  Result:= Haystack;
  StartIdx:= SysUtils.AnsiPos(Needle, Result);
  While StartIdx > 0 Do
  Begin
    Delete(Result, StartIdx, Length(Needle));
    StartIdx:= SysUtils.AnsiPos(Needle, Result);
  End;
End;

A call like this solves all these problems:

CONST
  Space = ‘ ’;
. . .
S:= DelAllStr(Space, S);
. . .

Andreas

[neslib]

We are entering a bit of a grey area here. Usually, whitespace is used to separate multiple values. In that case it shouldn't be ignored. Also, entering whitespace in "normal" floating-point literals is not allowed in Delphi.

I will look into this this weekend if I have some time. In general, I want to follow the same conventions as the Delphi RTL. So if the standard StrToFloat function does not accept whitespace, then neither should the DoubleDouble/QuadDouble version.

One way to make stupid users less stupid is to not accept stupid inputs😉

[Andreas113]

OK Erik, agree with you. But then such an input as follows should not be accepted at all, because thereby an apparent result - namely a wrong one - comes out:

Y/X:
----
X            = 1E - 1
Y            = - 2
MPA_Float    = -20 ---> Correct
QuadDouble   = -2.00000000000000000000000000000000000000000000000000000000000000
DoubleDouble = -2.0000000000000000000000000000000

Thanks
Andreas

[neslib]

I made some changes to make the parsing behave more like Delphi's StrToFloat:

• When using the implicit String operator (as in X := '- 2'), it will raise an EConvertError exception now when the input is invalid, so it will not create invalid output.
• Leading and trailing whitespace is accepted (as in StrToFloat).
• Other whitespace is regarded as a separator and stops parsing at that point. So X := '1 2' is parsed successfully as 1, but X := '- 2' results in an error because '-' by itself is not a valid number.

This will probably not account for all possible (parsing) errors, but it is more closely in line with what Delphi's StrToFloat does.

Let me know if there are any other cases where the behavior differs from the StrToFloat behavior.