Adds conformance tests for floats #134
Conversation
conformance/data_model/float.ion
Outdated
| (then "a subnormal" | ||
| (then "f32 value" | ||
| (binary "44 00 00 00 01") | ||
| (denotes (Float "1.401298464324817E-45"))) |
There was a problem hiding this comment.
| (denotes (Float "1.401298464324817E-45"))) | |
| (denotes (Float "1e-45"))) |
conformance/data_model/float.ion
Outdated
| (denotes (Float "5.9604645e-8"))) | ||
| (then "f32 value" | ||
| (binary "6C 01 00 00 00") | ||
| (denotes (Float "1.401298464324817E-45"))) |
There was a problem hiding this comment.
| (denotes (Float "1.401298464324817E-45"))) | |
| (denotes (Float "1e-45"))) |
conformance/data_model/float.ion
Outdated
| (denotes (Float "-5.9604645e-8"))) | ||
| (then "f32 value" | ||
| (binary "6C 01 00 00 80") | ||
| (denotes (Float "-1.401298464324817E-45"))) |
There was a problem hiding this comment.
| (denotes (Float "-1.401298464324817E-45"))) | |
| (denotes (Float "-1e-45"))) |
conformance/grammar.isl
Outdated
| // optional '-', one or more digits, '.', zero or more digits, optional group of 'e' and one or more digits OR | ||
| // optional '-', one or more digits, 'e', one or more digits OR | ||
| // one of the non numeric float values | ||
| regex: "^-?\\d+[.]\\d*(e\\d+)?|-?\\d+e\\d+|[+-]inf|nan$" |
There was a problem hiding this comment.
I wondered about this earlier, but why allow the conformance tests to express floats using the same syntax as Ion decimals? Why is the e optional?
There was a problem hiding this comment.
There was a bit of a TBD here, I guess, that didn't really get resolved. TLDR; model-float relies on the IEEE-754 string representation of floating point numbers.
We had a bit of a bootstrapping problem when it comes to testing floats. Unless we want to have model-float define values in terms of a sign, exponent, and mantissa, we need to have some alternate representation. I took the lazy/simple way and said that it should be a string that can be parsed by a floating point library (usually in the standard library of a language). Then, I ran into the problem that these libraries don't always agree on the serialization of non-numeric/finite values, so I had to specify that in a way that excluded things like +Infinity, +Inf, etc. The result is that I had to specify something for the finite float values. I could have done [0-9.e-]+, but that seemed too sloppy, so I approximated what is allowed in IEEE-754.
Anyway, although it might make it look like there's a bit of a circular definition, I don't mind changing it to Ion float syntax.
Edit—I was wrong. |
Co-authored-by: Tyler Gregg <[email protected]>
Issue #, if available:
#88
Description of changes:
Adds test cases for Ion floats.
By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.