On 2 November 2017 at 15:18, Kazuho Oku <kazuhooku@gmail.com> wrote: > [snip] > > > For floats, first let me state that I do not know if we even need to > support it. To me it seems worth considering to drop it, as PHK > suggested. > > That said, let me point out that you might not need to define the > maximum number of digits permitted in fraction part. You are anyways > expected to have some errors in floating point maths. The reason we > had issues in JSON Header Field Values is because it permitted the use > of exponents (i.e. "eNN") [3]. Without support for exponents, such > issue would not happen in Structured Headers. > > Not all numbers with dots in them are interpreted as IEEE 754 floats; for example Ruby's BigDecimal [1] and Rational [2] classes. If precise fractional numbers are required for a header value, there should definitely be strong statements about the required precision. That's a big "if", though; all the fractional numbers I can think of can quite easily be served by fixed-point numbers/integers, either by changing the units (seconds to microseconds, say), or adding a semantic constraint to the parsed value (e.g. q-value * 1000). So I wouldn't be opposed to it being dropped. Cheers [1]: https://ruby-doc.org/stdlib-2.4.2/libdoc/bigdecimal/rdoc/BigDecimal.html [2]: https://ruby-doc.org/core-2.4.2/Rational.html -- Matthew Kerwin http://matthew.kerwin.net.au/
Received on Thursday, 2 November 2017 05:58:31 UTC