Original bug ID: 4345
Reporter: brunodefraine
Assigned to: @xclerc
Status: closed (set by @xavierleroy on 2013-08-31T10:44:32Z)
Resolution: fixed
Priority: normal
Severity: minor
Version: 3.10.0
Fixed in version: 3.12.1+dev
Category: -for Camlp4 use https://github.com/ocaml/camlp4/issues
Related to: #3352 #4767
Monitored by: @Chris00

Bug description

First, consider the top-level printing of values:

$ ocaml

let pi = 3.14159265358979312 ;;

val pi : float = 3.14159265358979312

float_of_string "3.14159265358979312" = pi ;;

• : bool = true

While:

$ ocaml camlp4rf.cma

value pi = 3.14159265358979312 ;

value pi : float = 3.14159265359

float_of_string "3.14159265359" = pi ;

• : bool = False

Similarly, consider `flo antiquotations:

$ ocaml camlp4rf.cma

open Camlp4.PreCast;

value _loc = Loc.ghost;

value pi = 3.14159265358979312;

<:expr< $`flo:pi$ >> ;

• : Camlp4.PreCast.Ast.expr =
  Camlp4.PreCast.Ast.ExFlo "3.14159265359"

While it might not be possible to produce string versions whose float_of_string value always exactly equals the original float value, the precision of the currently produced versions is far below the precision of Ocaml's float type.

Additional information

This is probably related to the usage of string_of_float to produce the string versions. Consider the following alternative, located inside typing/oprint.ml, and probably used by OCaml's standard toplevel:

let float_repres f =
match classify_float f with
FP_nan -> "nan"
| FP_infinite ->
if f < 0.0 then "neg_infinity" else "infinity"
| _ ->
let s1 = Printf.sprintf "%.12g" f in
if f = float_of_string s1 then valid_float_lexeme s1 else
let s2 = Printf.sprintf "%.15g" f in
if f = float_of_string s2 then valid_float_lexeme s2 else
Printf.sprintf "%.18g" f