factor out float formatting into nicedecimal, include as dep

makefile

@@ -5,4 +5,4 @@ clean:
 	./rebar clean
 
 test: clean
-	./rebar --jobs=1 skip_deps=true eunit
+	./rebar --jobs=1 eunit

rebar.config

@@ -21,3 +21,10 @@
 {dialyzer_opts, [{warnings, [unmatched_returns, error_handling, race_conditions, behaviours]}]}.
 {xref_checks, [undefined_function_calls]}.
 {cover_enabled, true}.
+
+{deps, [
+    {'nicedecimal',
+        ".*",
+        {git, "git://github.com/talentdeficit/nicedecimal.git"}
+    }
+]}.

src/jsx_to_json.erl

@@ -139,7 +139,7 @@ encode(literal, Literal, _Opts) ->
 encode(integer, Integer, _Opts) ->
     erlang:integer_to_list(Integer);
 encode(float, Float, _Opts) ->
-    jsx_utils:nice_decimal(Float).
+    nicedecimal:format(Float).
 
 
 space(Opts) ->

src/jsx_utils.erl

@@ -24,7 +24,6 @@
 -module(jsx_utils).
 
 -export([parse_opts/1]).
--export([nice_decimal/1]).
 -export([json_escape/2]).
 
 -include("../include/jsx_opts.hrl").
@@ -49,137 +48,6 @@ parse_opts(_, _) ->
     {error, badarg}.
 
 
-
-%% conversion of floats to 'nice' decimal output. erlang's float implementation 
-%%   is almost but not quite ieee 754. it converts negative zero to plain zero 
-%%   silently, and throws exceptions for any operations that would produce NaN 
-%%   or infinity. as far as I can tell that is. trying to match against NaN or 
-%%   infinity binary patterns produces nomatch exceptions, and arithmetic 
-%%   operations produce badarg exceptions. with that in mind, this function 
-%%   makes no attempt to handle special values (except for zero)
-
-%% algorithm from "Printing Floating-Point Numbers Quickly and Accurately" by 
-%%   Burger & Dybvig
-
--spec nice_decimal(Float::float()) -> string().
-
-nice_decimal(0.0) -> "0.0";
-nice_decimal(Num) ->
-    {F, E} = extract(<<Num:64/float>>),
-    {R, S, MP, MM} = initial_vals(F, E),
-    K = ceiling((math:log(abs(Num)) / math:log(10)) - 1.0e-10),
-    Round = F band 1 =:= 0,
-    {Dpoint, Digits} = scale(R, S, MP, MM, K, 10, Round),
-    if Num >= 0 -> digits_to_list(Dpoint, Digits)
-        ; Num < 0 -> "-" ++ digits_to_list(Dpoint, Digits)
-    end.
-
-
-%% internal functions
-
-extract(<<_:1, 0:11, Frac:52>>) -> {Frac, -1074};
-extract(<<_:1, Exp:11, Frac:52>>) -> {Frac + (1 bsl 52), Exp - 1075}.
-
-
-initial_vals(F, E) when E >= 0, F /= 1 bsl 52 ->
-    BE = 1 bsl E,
-    {F * BE * 2, 2, BE, BE};    
-initial_vals(F, E) when E >= 0 ->
-    BE = 1 bsl E,
-    {F * BE * 4, 4, BE * 2, BE};
-initial_vals(F, E) when E == -1074; F /= 1 bsl 52 ->
-    {F * 2, 1 bsl (-E + 1), 1, 1};
-initial_vals(F, E) ->
-    {F * 4, 1 bsl (-E + 2), 2, 1}.
-
-
-ceiling(X) ->
-    Y = erlang:trunc(X),
-    case X - Y of 
-        Z when Z > 0 -> Y + 1 
-        ; _ -> Y 
-    end.
-
-
-scale(R, S, MP, MM, K, B, Round) ->
-    case K >= 0 of
-        true -> fixup(R, S * pow(B, K), MP, MM, K, B, Round)
-        ; false -> 
-            Scale = pow(B, -1 * K),
-            fixup(R * Scale, S, MP * Scale, MM * Scale, K, B, Round)
-    end.
-
-
-fixup(R, S, MP, MM, K, B, true) ->
-    case (R + MP >= S) of
-        true -> {K + 1, generate(R, S, MP, MM, B, true)}
-        ; false -> {K, generate(R * B, S, MP * B, MM * B, B, true)}
-    end;
-fixup(R, S, MP, MM, K, B, false) ->
-    case (R + MP > S) of
-        true -> {K + 1, generate(R, S, MP, MM, B, true)}
-        ; false -> {K, generate(R * B, S, MP * B, MM * B, B, true)}
-    end.
-
-
-generate(RT, S, MP, MM, B, Round) ->
-    D = RT div S,
-    R = RT rem S,
-    TC1 = case Round of true -> (R =< MM); false -> (R < MM) end,
-    TC2 = case Round of true -> (R + MP >= S); false -> (R + MP > S) end,
-    case TC1 of
-        false -> case TC2 of
-                false -> [D | generate(R * B, S, MP * B, MM * B, B, Round)]
-                ; true -> [D + 1]
-            end
-        ; true -> case TC2 of
-                false -> [D]
-                ; true -> case R * 2 < S of
-                    true -> [D]
-                    ; false -> [D + 1]
-                end
-            end
-    end.
-
-
-%% this is not efficient at all and should be replaced with a lookup table 
-%%   probably
-pow(_B, 0) -> 1;
-pow(B, E) when E > 0 -> pow(B, E, 1).
-
-pow(B, E, Acc) when E < 2 -> B * Acc;
-pow(B, E, Acc) when E band 1 == 1 -> pow(B * B, E bsr 1, B * Acc);
-pow(B, E, Acc) -> pow(B * B, E bsr 1, Acc).
-
-
-digits_to_list(0, Digits) ->
-    digits_to_list(Digits, ignore, ".0");
-digits_to_list(Dpoint, Digits) when Dpoint =< length(Digits), Dpoint > 0 ->
-    digits_to_list(Digits, Dpoint, []);
-digits_to_list(Dpoint, Digits) when Dpoint > 0 ->
-    Pad = Dpoint - length(Digits),
-    case Pad of
-        X when X > 6 -> 
-            digits_to_list(Digits, 1, []) ++ "e" ++ integer_to_list(Dpoint - 1)
-        ; _ -> 
-            digits_to_list(Digits ++ [ 0 || _ <- lists:seq(1, Pad)], Dpoint, [])
-    end;
-digits_to_list(Dpoint, Digits) when Dpoint < 0 ->
-    digits_to_list(Digits, 1, []) ++ "e" ++ integer_to_list(Dpoint - 1).
-
-digits_to_list([], 0, Acc) ->
-    lists:reverse("0." ++ Acc);
-digits_to_list([], ignore, Acc) ->
-    lists:reverse(Acc);
-digits_to_list(Digits, 0, Acc) ->
-    digits_to_list(Digits, ignore, "." ++ Acc); 
-digits_to_list([Digit|Digits], Dpoint, Acc) ->
-    digits_to_list(Digits, 
-        case Dpoint of ignore -> ignore; X -> X - 1 end, [to_hex(Digit)] ++ Acc
-    ).
-
-
-
 %% json string escaping, for utf8 binaries. escape the json control sequences to 
 %%  their json equivalent, escape other control characters to \uXXXX sequences, 
 %%  everything else should be a legal json string component
@@ -254,43 +122,6 @@ to_hex(X) -> X + 48.    %% ascii "1" is [49], "2" is [50], etc...
 -ifdef(TEST).
 -include_lib("eunit/include/eunit.hrl").
 
-nice_decimal_test_() ->
-    [
-        {"0.0", ?_assert(nice_decimal(0.0) =:= "0.0")},
-        {"1.0", ?_assert(nice_decimal(1.0) =:= "1.0")},
-        {"-1.0", ?_assert(nice_decimal(-1.0) =:= "-1.0")},
-        {"3.1234567890987654321", 
-            ?_assert(
-                nice_decimal(3.1234567890987654321) =:= "3.1234567890987655")
-        },
-        {"1.0e23", ?_assert(nice_decimal(1.0e23) =:= "1.0e23")},
-        {"0.3", ?_assert(nice_decimal(3.0/10.0) =:= "0.3")},
-        {"0.0001", ?_assert(nice_decimal(0.0001) =:= "1.0e-4")},
-        {"0.00000001", ?_assert(nice_decimal(0.00000001) =:= "1.0e-8")},
-        {"1.0e-323", ?_assert(nice_decimal(1.0e-323) =:= "1.0e-323")},
-        {"1.0e308", ?_assert(nice_decimal(1.0e308) =:= "1.0e308")},
-        {"min normalized float", 
-            ?_assert(
-                nice_decimal(math:pow(2, -1022)) =:= "2.2250738585072014e-308"
-            )
-        },
-        {"max normalized float", 
-            ?_assert(
-                nice_decimal((2 - math:pow(2, -52)) * math:pow(2, 1023)) 
-                    =:= "1.7976931348623157e308"
-            )
-        },
-        {"min denormalized float", 
-            ?_assert(nice_decimal(math:pow(2, -1074)) =:= "5.0e-324")
-        },
-        {"max denormalized float", 
-            ?_assert(
-                nice_decimal((1 - math:pow(2, -52)) * math:pow(2, -1022)) 
-                    =:= "2.225073858507201e-308"
-            )
-        }
-    ].
-
 
 binary_escape_test_() ->
     [