Draft: floating point literals

eeps/eep-0075.md

@@ -0,0 +1,119 @@
+    Author: Richard Carlsson <carlsson.richard(at)gmail(dot)com>
+    Status: Draft
+    Type: Standards Track
+    Created: 7-Jan-2025
+    Erlang-Version: OTP-28.0
+    Post-History:
+****
+EEP 75: Based Floating Point Literals
+----
+
+Abstract
+========
+
+This EEP adds floating point literals using other bases than ten, such
+as 16 or 2, for exact text representation, as also found in Ada and
+C99/C++17.
+
+Rationale
+=========
+
+Computers represent floating point numbers in binary, but such numbers
+are typically printed using base ten, for example `0.314159265e1`. In
+order to preserve the exact bit level precision when writing a printed
+number as text and later reading it back again, it is better to use a
+base that matches the internally used base, such as 16 for a compact
+but still exact representation, or 2 for visualizing or writing down
+the exact internal format. One particular case where such exact
+representations are useful is in code generating tools.
+
+Some other languages have support for floating point literals in other
+bases. Notably, [C99/C++17 floating-point literals][C99] can be
+written in hexadecimal, as e.g. `0xf.ffp8`, where the `p` indicates
+that the exponent is a power of 2, and in the [Ada][Ada] programming
+language, the corresponding syntax is `16#F.FF#E8`. The latter should
+look familiar to Erlang users, because it is from Ada that the
+`<Base>#<Numeral>` syntax was borrowed. (Ada however requires a final
+`#` even for integers, e.g. `16#fffe#`.) Ada also allows base 2 in a
+floating point number, e.g. `2#0.1111_1111#E8`, using underscores as
+separators just like Erlang does.
+
+Where Erlang differs from Ada is that Ada does not allow the base to
+be larger than 16. Hence, in Ada, `2#111#`, `7#10#`, and `16#7#` are
+all the same number, but `17#7#` is illegal, while Erlang allows any
+base up to 36 in its integer literals, e.g. `36#z`. It should also be
+noted that in the C99 hexadecimal literals, the letter `p` for the
+exponent is a valid digit in bases above 25 in Erlang, whereas C99
+only allows digits up to `f` (and could not use `e` for exponents in
+hex). Because the Ada notation requires a `#` character before the
+exponent, it has no ambiguity between digits and exponent indicator.
+
+Staying with the Ada notation then seems to be the wise choice, both
+for consistency and because it makes it trivial to keep allowing any
+base up to 36 also for floating point literals in Erlang.
+
+Examples:
+
+    2#0.111
+    2#0.10101#e8
+    16#ff.ff
+    16#fefe.fefe#e16
+    32#vrv.vrv#e15
+
+It should be noted that both the base and the exponent are always
+interpreted in base ten. Only the digits between the two `#`
+characters are interpreted using the given base. Because Erlang uses
+the `#` characters at the start of constructs such as maps `#{...}`,
+we do not want to allow a final trailing `#` in a number, like Ada
+does. If there is a second `#`, it must be followed by the exponent.
+
+Specification
+========================
+
+In addition to the current based notation:
+
+    base # based_numeral
+
+(borrowing Ada's terminology), where `base` is a decimal number and
+`based_numeral` is a sequence of digits in `0-9` and `a-z` or `A-Z`,
+optionally separated with `_`, we extend the parser to also allow:
+
+    base # based_numeral.based_numeral [ # exponent ]
+
+where `exponent` is the letter `e` or `E` followed by an optionally
+signed decimal number, exactly as in ordinary decimal floating point
+literals.
+
+Reference Implementation
+------------------------
+
+A [reference implementation][GitHub branch] exists in the
+`hexbinfloat` branch of the author's GitHub account, together with a
+[GitHub pull request][GitHub PR] to the Erlang/OTP repository.
+
+[C99]: https://en.cppreference.com/w/cpp/language/floating_literal
+    "C99 and C++17 Floating-point Literal"
+
+[Ada]: https://ada-lang.io/docs/arm/AA-2/AA-2.4#242--based-literals
+   "Ada Based Literals"
+
+[GitHub branch]: https://github.com/richcarl/otp/tree/hexbinfloat
+    "Reference implementation branch on GitHub"
+
+[GitHub PR]: https://github.com/erlang/otp/pull/9106
+    "GitHub Pull Request"
+
+Copyright
+=========
+
+This document is placed in the public domain or under the CC0-1.0-Universal
+license, whichever is more permissive.
+
+[EmacsVar]: <> "Local Variables:"
+[EmacsVar]: <> "mode: indented-text"
+[EmacsVar]: <> "indent-tabs-mode: nil"
+[EmacsVar]: <> "sentence-end-double-space: t"
+[EmacsVar]: <> "fill-column: 70"
+[EmacsVar]: <> "coding: utf-8"
+[EmacsVar]: <> "End:"
+[VimVar]: <> " vim: set fileencoding=utf-8 expandtab shiftwidth=4 softtabstop=4: "