#73 TryStr2Float(Single/Double): a value outside the type's range raises EOverflow or returns True with INF

[Thomas Mueller]

The Single and Double overloads of TryStr2Float in u_dzConvertUtils convert the string to Extended and then assign that to the target type without a range check. A value that is a valid Extended but too large for Single / Double therefore does not return False:

	Delphi 7, 2007, 11 (Win32)	Delphi 12 Win32	Delphi 12 Win64
Single 1e300	raises EOverflow	True, INF	True, INF
Double 1e400	raises EOverflow	True, INF	False (correct, Extended is Double there)

Up to Delphi 11 the assignment raises because overflow exceptions are unmasked; Delphi 12 masks floating point exceptions, so the result is INF and the function returns True.

This also affects everything built on these overloads, e.g. the Single and Double overloads of TIniFile_TryReadFloat in u_dzClassUtils, which with ehNoException are supposed to return False for an invalid value.

Expected: False if the value is outside the range of the target type.

Measured with a console program calling TryStr2Float('1e300', SingleVar) and TryStr2Float('1e400', DoubleVar) and printing the result or the exception class.

[Thomas Mueller]

Fixed in revision #2473.

The Single and Double overloads of TryStr2Float now return False for a value that would round to INF in the target type, on every version, instead of raising EOverflow (up to Delphi 11) or returning True with INF (Delphi 12). The check compares with the type's largest value plus half a unit in the last place, so a value that rounds down to MaxSingle / MaxDouble, e.g. 3.40282346638529E38, is still accepted. INF and NAN, which Delphi 12 accepts, stay valid. A value too small for the type still becomes a denormal or 0. TIniFile_TryReadFloat with Single or Double benefits from this. New tests in ConvertUtils.