(format nil "~,1e" 99.999) ; => "10.0e+1"

[Alexander Klimov]

on SBCL 1.0.52.1-805f1dd

 (format nil "~,1e" 99.999) ; => "10.0e+1"

but it should produce the same result as

 (format nil "~,1e" 100.001) ; => "1.0e+2"

The attached patch contains more tests.

I guess the culprit is format-exp-aux that uses scale-exponent on the non-rounded number, but applies (thru flonum-to-string) the produced exponent to the rounded number.

[Nikodemus Siivola (nikodemus)]

I'm not sure they are actually required to produce the same string.

My reasoning:

 "k is a scale factor that defaults to one (not zero)" -

So k is 1.

"If k is positive, then it must be strictly less than d+2; k significant digits are printed before the decimal point, and d-k+1 digits are printed after the decimal point."

Both "10.0e+1" and "1.0e+2" have 1 /significant/ digit before the decimal point.

d-k+1 = 1

Both have 1 digit after the decimal point, so unless there are other constraints which I missed, they both seem legal to me.

Can you lay out your reasoning as to why they should produce the same output?

[Alexander Klimov (alserkli)]

Since <HyperSpec/Body/26_glo_s.htm> does not contain a definition of
"significant digit", I guess we can apply the usual definition applied
to exponential (aka. scientific) notation
<http://en.wikipedia.org/wiki/Significant_figures#Scientific_notation>:

  [I]n the normalized form of that notation, placeholder leading and
  trailing digits do not occur, so all digits are significant. For
  example, 0.00012 (two significant digits) becomes 1.2×10^−4, and
  0.00122300 (six significant digits) becomes 1.22300×10^−3. In
  particular, the potential ambiguity about the significance of
  trailing zeros is eliminated. For example, 1300 to four significant
  digits is written as 1.300×10^3, while 1300 to two significant digits
  is written as 1.3×10^3.

That is 10.0 has two significant digits before the point and one after
it.

Note that if "10.[...]" were understood to have 1 significant digit
because the trailing zeros are considered insignificant, then

 (format nil "~,3,,4e" 123) ; -> "1230.e-1"

would be an error since there would be only 3 significant digits
before point while k=4.

[Nikodemus Siivola (nikodemus)]

Fair point.