Nicolas Cellier uploaded a new version of Kernel to project The Trunk: http://source.squeak.org/trunk/Kernel-nice.1492.mcz
==================== Summary ====================
Name: Kernel-nice.1492 Author: nice Time: 24 November 2022, 1:03:20.758687 pm UUID: cec2ead5-112d-5e4f-9ffe-969e3a7f575a Ancestors: Kernel-ct.1491
Fix a failure to round to nearest Float for large negative exponent when no fraction part is provided.
For example, this should be true: 25e-108 = 25.0e-108.
Because (10 raisedToInteger: exp) isAnExactFloat is false for any exp abs > 22, we shall not recompose the float as integerPart * (base asFloat raisedTo: exponent), that's potentially two inexact operations in a row, and a chance of incorrect rounding.
=============== Diff against Kernel-ct.1491 ===============
Item was changed: ----- Method: SqNumberParser>>makeIntegerOrScaledInteger (in category 'parsing-private') ----- makeIntegerOrScaledInteger "At this point, there is no digit, nor fractionPart. Maybe it can be a scaled decimal with fraction omitted... Maybe it can be a Float with a negative exponent" neg ifTrue: [integerPart := integerPart negated]. self readExponent ifTrue: ["Check that the result is an integer, otherwise answer a Float. Fractions are /not/ valid literals." + | temp | + temp := integerPart * (base raisedToInteger: exponent). + temp isInteger ifFalse: [^temp asFloat]. + ^temp]. - (exponent >= 0 or: [(integerPart * (base raisedToInteger: exponent)) isInteger]) ifFalse: - [base := base asFloat]. - ^integerPart * (base raisedToInteger: exponent)]. (self readScaleWithDefaultNumberOfDigits: 0) ifTrue: [^integerPart asScaledDecimal: scale]. ^ integerPart!