Search code examples
cfloating-pointprintflanguage-lawyerformat-specifiers

Interpreting the format specifier in printf("%# 01.1g",9.8)

Consider the following printf instruction:

printf("%# 01.1g", 9.8);

what should it print?

I'm reading the description of the g specifier on cppreference.com, which says (text for G removed):

converts floating-point number to decimal or decimal exponent notation depending on the value and the precision.

For the g conversion style conversion with style e or f will be performed.
Let P equal the precision if nonzero, 6 if the precision is not specified, or 1 if the precision is ​0​. Then, if a conversion with style e would have an exponent of X:

  • if P > X ≥ −4, the conversion is with style f and precision P − 1 − X.
  • otherwise, the conversion is with style e and precision P − 1.

Unless alternative representation is requested the trailing zeros are removed, also the decimal point character is removed if no fractional part is left.

In our case,

  • P = 1 , specified explicitly.
  • X = 0, since a "conversion with style e", i.e. "%# 01.1e", yields 9.8e+00 (adjust the GodBolt program and you'll see).
  • 1 > 0 >= -4 holds.

consequently, the conversion should be with style f and precision P - 1 - X = 1 - 1 - 0 = 0, i.e. "%# 01.0f", which yields 10..

... but that is not what glibc produces: I get 1.e01, as can also be seen on GodBolt.

So,

  • Am I mis-reading the quoted text?
  • Is cppreference.com wrong?
  • Is this - perish the tought - a glib 2.36 bug?
Solution

  • This is a lack of clarity about “a conversion with style E” and a discrepancy about what the precision means for E and g.

    With E, the precision is the number of digits to appear after the decimal point, per C 2018 7.21.6.1 4. With g, the precision is the maximum number of significant digits. Those differ; E has an additional digit before the decimal point, giving a total of one more digit than its nominal “precision.”

    Thus, in considering how to format 9.8 for %.1g, we first consider how it would be formatted for %.0E, not %.1E, as both %.1g and %.0E request one digit, whereas %.1E requests two digits. For %.0E, “1e+01” would be produced. So the X, the exponent, is 1, not 0.