%% %% %CopyrightBegin% %% %% Copyright Ericsson AB 1996-2013. All Rights Reserved. %% %% The contents of this file are subject to the Erlang Public License, %% Version 1.1, (the "License"); you may not use this file except in %% compliance with the License. You should have received a copy of the %% Erlang Public License along with this software. If not, it can be %% retrieved online at http://www.erlang.org/. %% %% Software distributed under the License is distributed on an "AS IS" %% basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See %% the License for the specific language governing rights and limitations %% under the License. %% %% %CopyrightEnd% %% -module(io_lib_format). %% Formatting functions of io library. -export([fwrite/2,fwrite_g/1,indentation/2]). %% fwrite(Format, ArgList) -> string(). %% Format the arguments in ArgList after string Format. Just generate %% an error if there is an error in the arguments. %% %% To do the printing command correctly we need to calculate the %% current indentation for everything before it. This may be very %% expensive, especially when it is not needed, so we first determine %% if, and for how long, we need to calculate the indentations. We do %% this by first collecting all the control sequences and %% corresponding arguments, then counting the print sequences and %% then building the output. This method has some drawbacks, it does %% two passes over the format string and creates more temporary data, %% and it also splits the handling of the control characters into two %% parts. fwrite(Format, Args) when is_atom(Format) -> fwrite(atom_to_list(Format), Args); fwrite(Format, Args) when is_binary(Format) -> fwrite(binary_to_list(Format), Args); fwrite(Format, Args) -> Cs = collect(Format, Args), Pc = pcount(Cs), build(Cs, Pc, 0). collect([$~|Fmt0], Args0) -> {C,Fmt1,Args1} = collect_cseq(Fmt0, Args0), [C|collect(Fmt1, Args1)]; collect([C|Fmt], Args) -> [C|collect(Fmt, Args)]; collect([], []) -> []. collect_cseq(Fmt0, Args0) -> {F,Ad,Fmt1,Args1} = field_width(Fmt0, Args0), {P,Fmt2,Args2} = precision(Fmt1, Args1), {Pad,Fmt3,Args3} = pad_char(Fmt2, Args2), {Encoding,Fmt4,Args4} = encoding(Fmt3, Args3), {C,As,Fmt5,Args5} = collect_cc(Fmt4, Args4), {{C,As,F,Ad,P,Pad,Encoding},Fmt5,Args5}. encoding([$t|Fmt],Args) -> {unicode,Fmt,Args}; encoding(Fmt,Args) -> {latin1,Fmt,Args}. field_width([$-|Fmt0], Args0) -> {F,Fmt,Args} = field_value(Fmt0, Args0), field_width(-F, Fmt, Args); field_width(Fmt0, Args0) -> {F,Fmt,Args} = field_value(Fmt0, Args0), field_width(F, Fmt, Args). field_width(F, Fmt, Args) when F < 0 -> {-F,left,Fmt,Args}; field_width(F, Fmt, Args) when F >= 0 -> {F,right,Fmt,Args}. precision([$.|Fmt], Args) -> field_value(Fmt, Args); precision(Fmt, Args) -> {none,Fmt,Args}. field_value([$*|Fmt], [A|Args]) when is_integer(A) -> {A,Fmt,Args}; field_value([C|Fmt], Args) when is_integer(C), C >= $0, C =< $9 -> field_value([C|Fmt], Args, 0); field_value(Fmt, Args) -> {none,Fmt,Args}. field_value([C|Fmt], Args, F) when is_integer(C), C >= $0, C =< $9 -> field_value(Fmt, Args, 10*F + (C - $0)); field_value(Fmt, Args, F) -> %Default case {F,Fmt,Args}. pad_char([$.,$*|Fmt], [Pad|Args]) -> {Pad,Fmt,Args}; pad_char([$.,Pad|Fmt], Args) -> {Pad,Fmt,Args}; pad_char(Fmt, Args) -> {$\s,Fmt,Args}. %% collect_cc([FormatChar], [Argument]) -> %% {Control,[ControlArg],[FormatChar],[Arg]}. %% Here we collect the argments for each control character. %% Be explicit to cause failure early. collect_cc([$w|Fmt], [A|Args]) -> {$w,[A],Fmt,Args}; collect_cc([$p|Fmt], [A|Args]) -> {$p,[A],Fmt,Args}; collect_cc([$W|Fmt], [A,Depth|Args]) -> {$W,[A,Depth],Fmt,Args}; collect_cc([$P|Fmt], [A,Depth|Args]) -> {$P,[A,Depth],Fmt,Args}; collect_cc([$s|Fmt], [A|Args]) -> {$s,[A],Fmt,Args}; collect_cc([$e|Fmt], [A|Args]) -> {$e,[A],Fmt,Args}; collect_cc([$f|Fmt], [A|Args]) -> {$f,[A],Fmt,Args}; collect_cc([$g|Fmt], [A|Args]) -> {$g,[A],Fmt,Args}; collect_cc([$b|Fmt], [A|Args]) -> {$b,[A],Fmt,Args}; collect_cc([$B|Fmt], [A|Args]) -> {$B,[A],Fmt,Args}; collect_cc([$x|Fmt], [A,Prefix|Args]) -> {$x,[A,Prefix],Fmt,Args}; collect_cc([$X|Fmt], [A,Prefix|Args]) -> {$X,[A,Prefix],Fmt,Args}; collect_cc([$+|Fmt], [A|Args]) -> {$+,[A],Fmt,Args}; collect_cc([$#|Fmt], [A|Args]) -> {$#,[A],Fmt,Args}; collect_cc([$c|Fmt], [A|Args]) -> {$c,[A],Fmt,Args}; collect_cc([$~|Fmt], Args) when is_list(Args) -> {$~,[],Fmt,Args}; collect_cc([$n|Fmt], Args) when is_list(Args) -> {$n,[],Fmt,Args}; collect_cc([$i|Fmt], [A|Args]) -> {$i,[A],Fmt,Args}. %% pcount([ControlC]) -> Count. %% Count the number of print requests. pcount(Cs) -> pcount(Cs, 0). pcount([{$p,_As,_F,_Ad,_P,_Pad,_Enc}|Cs], Acc) -> pcount(Cs, Acc+1); pcount([{$P,_As,_F,_Ad,_P,_Pad,_Enc}|Cs], Acc) -> pcount(Cs, Acc+1); pcount([_|Cs], Acc) -> pcount(Cs, Acc); pcount([], Acc) -> Acc. %% build([Control], Pc, Indentation) -> string(). %% Interpret the control structures. Count the number of print %% remaining and only calculate indentation when necessary. Must also %% be smart when calculating indentation for characters in format. build([{C,As,F,Ad,P,Pad,Enc}|Cs], Pc0, I) -> S = control(C, As, F, Ad, P, Pad, Enc, I), Pc1 = decr_pc(C, Pc0), if Pc1 > 0 -> [S|build(Cs, Pc1, indentation(S, I))]; true -> [S|build(Cs, Pc1, I)] end; build([$\n|Cs], Pc, _I) -> [$\n|build(Cs, Pc, 0)]; build([$\t|Cs], Pc, I) -> [$\t|build(Cs, Pc, ((I + 8) div 8) * 8)]; build([C|Cs], Pc, I) -> [C|build(Cs, Pc, I+1)]; build([], _Pc, _I) -> []. decr_pc($p, Pc) -> Pc - 1; decr_pc($P, Pc) -> Pc - 1; decr_pc(_, Pc) -> Pc. %% indentation(String, Indentation) -> Indentation. %% Calculate the indentation of the end of a string given its start %% indentation. We assume tabs at 8 cols. indentation([$\n|Cs], _I) -> indentation(Cs, 0); indentation([$\t|Cs], I) -> indentation(Cs, ((I + 8) div 8) * 8); indentation([C|Cs], I) when is_integer(C) -> indentation(Cs, I+1); indentation([C|Cs], I) -> indentation(Cs, indentation(C, I)); indentation([], I) -> I. %% control(FormatChar, [Argument], FieldWidth, Adjust, Precision, PadChar, %% Encoding, Indentation) -> String %% This is the main dispatch function for the various formatting commands. %% Field widths and precisions have already been calculated. control($w, [A], F, Adj, P, Pad, _Enc,_I) -> term(io_lib:write(A, -1), F, Adj, P, Pad); control($p, [A], F, Adj, P, Pad, Enc, I) -> print(A, -1, F, Adj, P, Pad, Enc, I); control($W, [A,Depth], F, Adj, P, Pad, _Enc, _I) when is_integer(Depth) -> term(io_lib:write(A, Depth), F, Adj, P, Pad); control($P, [A,Depth], F, Adj, P, Pad, Enc, I) when is_integer(Depth) -> print(A, Depth, F, Adj, P, Pad, Enc, I); control($s, [A], F, Adj, P, Pad, _Enc, _I) when is_atom(A) -> string(atom_to_list(A), F, Adj, P, Pad); control($s, [L0], F, Adj, P, Pad, latin1, _I) -> L = iolist_to_chars(L0), string(L, F, Adj, P, Pad); control($s, [L0], F, Adj, P, Pad, unicode, _I) -> L = cdata_to_chars(L0), uniconv(string(L, F, Adj, P, Pad)); control($e, [A], F, Adj, P, Pad, _Enc, _I) when is_float(A) -> fwrite_e(A, F, Adj, P, Pad); control($f, [A], F, Adj, P, Pad, _Enc, _I) when is_float(A) -> fwrite_f(A, F, Adj, P, Pad); control($g, [A], F, Adj, P, Pad, _Enc, _I) when is_float(A) -> fwrite_g(A, F, Adj, P, Pad); control($b, [A], F, Adj, P, Pad, _Enc, _I) when is_integer(A) -> unprefixed_integer(A, F, Adj, base(P), Pad, true); control($B, [A], F, Adj, P, Pad, _Enc, _I) when is_integer(A) -> unprefixed_integer(A, F, Adj, base(P), Pad, false); control($x, [A,Prefix], F, Adj, P, Pad, _Enc, _I) when is_integer(A), is_atom(Prefix) -> prefixed_integer(A, F, Adj, base(P), Pad, atom_to_list(Prefix), true); control($x, [A,Prefix], F, Adj, P, Pad, _Enc, _I) when is_integer(A) -> true = io_lib:deep_char_list(Prefix), %Check if Prefix a character list prefixed_integer(A, F, Adj, base(P), Pad, Prefix, true); control($X, [A,Prefix], F, Adj, P, Pad, _Enc, _I) when is_integer(A), is_atom(Prefix) -> prefixed_integer(A, F, Adj, base(P), Pad, atom_to_list(Prefix), false); control($X, [A,Prefix], F, Adj, P, Pad, _Enc, _I) when is_integer(A) -> true = io_lib:deep_char_list(Prefix), %Check if Prefix a character list prefixed_integer(A, F, Adj, base(P), Pad, Prefix, false); control($+, [A], F, Adj, P, Pad, _Enc, _I) when is_integer(A) -> Base = base(P), Prefix = [integer_to_list(Base), $#], prefixed_integer(A, F, Adj, Base, Pad, Prefix, true); control($#, [A], F, Adj, P, Pad, _Enc, _I) when is_integer(A) -> Base = base(P), Prefix = [integer_to_list(Base), $#], prefixed_integer(A, F, Adj, Base, Pad, Prefix, false); control($c, [A], F, Adj, P, Pad, unicode, _I) when is_integer(A) -> char(A, F, Adj, P, Pad); control($c, [A], F, Adj, P, Pad, _Enc, _I) when is_integer(A) -> char(A band 255, F, Adj, P, Pad); control($~, [], F, Adj, P, Pad, _Enc, _I) -> char($~, F, Adj, P, Pad); control($n, [], F, Adj, P, Pad, _Enc, _I) -> newline(F, Adj, P, Pad); control($i, [_A], _F, _Adj, _P, _Pad, _Enc, _I) -> []. -ifdef(UNICODE_AS_BINARIES). uniconv(C) -> unicode:characters_to_binary(C,unicode). -else. uniconv(C) -> C. -endif. %% Default integer base base(none) -> 10; base(B) when is_integer(B) -> B. %% term(TermList, Field, Adjust, Precision, PadChar) %% Output the characters in a term. %% Adjust the characters within the field if length less than Max padding %% with PadChar. term(T, none, _Adj, none, _Pad) -> T; term(T, none, Adj, P, Pad) -> term(T, P, Adj, P, Pad); term(T, F, Adj, P0, Pad) -> L = lists:flatlength(T), P = case P0 of none -> erlang:min(L, F); _ -> P0 end, if L > P -> adjust(chars($*, P), chars(Pad, F-P), Adj); F >= P -> adjust(T, chars(Pad, F-L), Adj) end. %% print(Term, Depth, Field, Adjust, Precision, PadChar, Encoding, %% Indentation) %% Print a term. print(T, D, none, Adj, P, Pad, E, I) -> print(T, D, 80, Adj, P, Pad, E, I); print(T, D, F, Adj, none, Pad, E, I) -> print(T, D, F, Adj, I+1, Pad, E, I); print(T, D, F, right, P, _Pad, latin1, _I) -> io_lib_pretty:print(T, P, F, D); print(T, D, F, right, P, _Pad, Enc, _I) -> Options = [{column, P}, {line_length, F}, {depth, D}, {encoding, Enc}], io_lib_pretty:print(T, Options). %% fwrite_e(Float, Field, Adjust, Precision, PadChar) fwrite_e(Fl, none, Adj, none, Pad) -> %Default values fwrite_e(Fl, none, Adj, 6, Pad); fwrite_e(Fl, none, _Adj, P, _Pad) when P >= 2 -> float_e(Fl, float_data(Fl), P); fwrite_e(Fl, F, Adj, none, Pad) -> fwrite_e(Fl, F, Adj, 6, Pad); fwrite_e(Fl, F, Adj, P, Pad) when P >= 2 -> term(float_e(Fl, float_data(Fl), P), F, Adj, F, Pad). float_e(Fl, Fd, P) when Fl < 0.0 -> %Negative numbers [$-|float_e(-Fl, Fd, P)]; float_e(_Fl, {Ds,E}, P) -> case float_man(Ds, 1, P-1) of {[$0|Fs],true} -> [[$1|Fs]|float_exp(E)]; {Fs,false} -> [Fs|float_exp(E-1)] end. %% float_man([Digit], Icount, Dcount) -> {[Chars],CarryFlag}. %% Generate the characters in the mantissa from the digits with Icount %% characters before the '.' and Dcount decimals. Handle carry and let %% caller decide what to do at top. float_man(Ds, 0, Dc) -> {Cs,C} = float_man(Ds, Dc), {[$.|Cs],C}; float_man([D|Ds], I, Dc) -> case float_man(Ds, I-1, Dc) of {Cs,true} when D =:= $9 -> {[$0|Cs],true}; {Cs,true} -> {[D+1|Cs],false}; {Cs,false} -> {[D|Cs],false} end; float_man([], I, Dc) -> %Pad with 0's {string:chars($0, I, [$.|string:chars($0, Dc)]),false}. float_man([D|_], 0) when D >= $5 -> {[],true}; float_man([_|_], 0) -> {[],false}; float_man([D|Ds], Dc) -> case float_man(Ds, Dc-1) of {Cs,true} when D =:= $9 -> {[$0|Cs],true}; {Cs,true} -> {[D+1|Cs],false}; {Cs,false} -> {[D|Cs],false} end; float_man([], Dc) -> {string:chars($0, Dc),false}. %Pad with 0's %% float_exp(Exponent) -> [Char]. %% Generate the exponent of a floating point number. Always include sign. float_exp(E) when E >= 0 -> [$e,$+|integer_to_list(E)]; float_exp(E) -> [$e|integer_to_list(E)]. %% fwrite_f(FloatData, Field, Adjust, Precision, PadChar) fwrite_f(Fl, none, Adj, none, Pad) -> %Default values fwrite_f(Fl, none, Adj, 6, Pad); fwrite_f(Fl, none, _Adj, P, _Pad) when P >= 1 -> float_f(Fl, float_data(Fl), P); fwrite_f(Fl, F, Adj, none, Pad) -> fwrite_f(Fl, F, Adj, 6, Pad); fwrite_f(Fl, F, Adj, P, Pad) when P >= 1 -> term(float_f(Fl, float_data(Fl), P), F, Adj, F, Pad). float_f(Fl, Fd, P) when Fl < 0.0 -> [$-|float_f(-Fl, Fd, P)]; float_f(Fl, {Ds,E}, P) when E =< 0 -> float_f(Fl, {string:chars($0, -E+1, Ds),1}, P); %Prepend enough 0's float_f(_Fl, {Ds,E}, P) -> case float_man(Ds, E, P) of {Fs,true} -> "1" ++ Fs; %Handle carry {Fs,false} -> Fs end. %% float_data([FloatChar]) -> {[Digit],Exponent} float_data(Fl) -> float_data(float_to_list(Fl), []). float_data([$e|E], Ds) -> {lists:reverse(Ds),list_to_integer(E)+1}; float_data([D|Cs], Ds) when D >= $0, D =< $9 -> float_data(Cs, [D|Ds]); float_data([_|Cs], Ds) -> float_data(Cs, Ds). %% fwrite_g(Float) %% Writes the shortest, correctly rounded string that converts %% to Float when read back with list_to_float/1. %% %% See also "Printing Floating-Point Numbers Quickly and Accurately" %% in Proceedings of the SIGPLAN '96 Conference on Programming %% Language Design and Implementation. fwrite_g(0.0) -> "0.0"; fwrite_g(Float) when is_float(Float) -> {Frac, Exp} = mantissa_exponent(Float), {Place, Digits} = fwrite_g_1(Float, Exp, Frac), R = insert_decimal(Place, [$0 + D || D <- Digits]), [$- || true <- [Float < 0.0]] ++ R. -define(BIG_POW, (1 bsl 52)). -define(MIN_EXP, (-1074)). mantissa_exponent(F) -> case <> of <<_S:1, 0:11, M:52>> -> % denormalized E = log2floor(M), {M bsl (53 - E), E - 52 - 1075}; <<_S:1, BE:11, M:52>> when BE < 2047 -> {M + ?BIG_POW, BE - 1075} end. fwrite_g_1(Float, Exp, Frac) -> Round = (Frac band 1) =:= 0, if Exp >= 0 -> BExp = 1 bsl Exp, if Frac =:= ?BIG_POW -> scale(Frac * BExp * 4, 4, BExp * 2, BExp, Round, Round, Float); true -> scale(Frac * BExp * 2, 2, BExp, BExp, Round, Round, Float) end; Exp < ?MIN_EXP -> BExp = 1 bsl (?MIN_EXP - Exp), scale(Frac * 2, 1 bsl (1 - Exp), BExp, BExp, Round, Round, Float); Exp > ?MIN_EXP, Frac =:= ?BIG_POW -> scale(Frac * 4, 1 bsl (2 - Exp), 2, 1, Round, Round, Float); true -> scale(Frac * 2, 1 bsl (1 - Exp), 1, 1, Round, Round, Float) end. scale(R, S, MPlus, MMinus, LowOk, HighOk, Float) -> Est = int_ceil(math:log10(abs(Float)) - 1.0e-10), %% Note that the scheme implementation uses a 326 element look-up %% table for int_pow(10, N) where we do not. if Est >= 0 -> fixup(R, S * int_pow(10, Est), MPlus, MMinus, Est, LowOk, HighOk); true -> Scale = int_pow(10, -Est), fixup(R * Scale, S, MPlus * Scale, MMinus * Scale, Est, LowOk, HighOk) end. fixup(R, S, MPlus, MMinus, K, LowOk, HighOk) -> TooLow = if HighOk -> R + MPlus >= S; true -> R + MPlus > S end, case TooLow of true -> {K + 1, generate(R, S, MPlus, MMinus, LowOk, HighOk)}; false -> {K, generate(R * 10, S, MPlus * 10, MMinus * 10, LowOk, HighOk)} end. generate(R0, S, MPlus, MMinus, LowOk, HighOk) -> D = R0 div S, R = R0 rem S, TC1 = if LowOk -> R =< MMinus; true -> R < MMinus end, TC2 = if HighOk -> R + MPlus >= S; true -> R + MPlus > S end, case {TC1, TC2} of {false, false} -> [D | generate(R * 10, S, MPlus * 10, MMinus * 10, LowOk, HighOk)]; {false, true} -> [D + 1]; {true, false} -> [D]; {true, true} when R * 2 < S -> [D]; {true, true} -> [D + 1] end. insert_decimal(0, S) -> "0." ++ S; insert_decimal(Place, S) -> L = length(S), if Place < 0; Place >= L -> ExpL = integer_to_list(Place - 1), ExpDot = if L =:= 1 -> 2; true -> 1 end, ExpCost = length(ExpL) + 1 + ExpDot, if Place < 0 -> if 2 - Place =< ExpCost -> "0." ++ lists:duplicate(-Place, $0) ++ S; true -> insert_exp(ExpL, S) end; true -> if Place - L + 2 =< ExpCost -> S ++ lists:duplicate(Place - L, $0) ++ ".0"; true -> insert_exp(ExpL, S) end end; true -> {S0, S1} = lists:split(Place, S), S0 ++ "." ++ S1 end. insert_exp(ExpL, [C]) -> [C] ++ ".0e" ++ ExpL; insert_exp(ExpL, [C | S]) -> [C] ++ "." ++ S ++ "e" ++ ExpL. int_ceil(X) when is_float(X) -> T = trunc(X), case (X - T) of Neg when Neg < 0 -> T; Pos when Pos > 0 -> T + 1; _ -> T end. int_pow(X, 0) when is_integer(X) -> 1; int_pow(X, N) when is_integer(X), is_integer(N), N > 0 -> int_pow(X, N, 1). int_pow(X, N, R) when N < 2 -> R * X; int_pow(X, N, R) -> int_pow(X * X, N bsr 1, case N band 1 of 1 -> R * X; 0 -> R end). log2floor(Int) when is_integer(Int), Int > 0 -> log2floor(Int, 0). log2floor(0, N) -> N; log2floor(Int, N) -> log2floor(Int bsr 1, 1 + N). %% fwrite_g(Float, Field, Adjust, Precision, PadChar) %% Use the f form if Float is >= 0.1 and < 1.0e4, %% and the prints correctly in the f form, else the e form. %% Precision always means the # of significant digits. fwrite_g(Fl, F, Adj, none, Pad) -> fwrite_g(Fl, F, Adj, 6, Pad); fwrite_g(Fl, F, Adj, P, Pad) when P >= 1 -> A = abs(Fl), E = if A < 1.0e-1 -> -2; A < 1.0e0 -> -1; A < 1.0e1 -> 0; A < 1.0e2 -> 1; A < 1.0e3 -> 2; A < 1.0e4 -> 3; true -> fwrite_f end, if P =< 1, E =:= -1; P-1 > E, E >= -1 -> fwrite_f(Fl, F, Adj, P-1-E, Pad); P =< 1 -> fwrite_e(Fl, F, Adj, 2, Pad); true -> fwrite_e(Fl, F, Adj, P, Pad) end. %% iolist_to_chars(iolist()) -> deep_char_list() iolist_to_chars([C|Cs]) when is_integer(C), C >= $\000, C =< $\377 -> [C | iolist_to_chars(Cs)]; iolist_to_chars([I|Cs]) -> [iolist_to_chars(I) | iolist_to_chars(Cs)]; iolist_to_chars([]) -> []; iolist_to_chars(B) when is_binary(B) -> binary_to_list(B). %% cdata() :: clist() | cbinary() %% clist() :: maybe_improper_list(char() | cbinary() | clist(), %% cbinary() | nil()) %% cbinary() :: unicode:unicode_binary() | unicode:latin1_binary() %% cdata_to_chars(cdata()) -> io_lib:deep_char_list() cdata_to_chars([C|Cs]) when is_integer(C), C >= $\000 -> [C | cdata_to_chars(Cs)]; cdata_to_chars([I|Cs]) -> [cdata_to_chars(I) | cdata_to_chars(Cs)]; cdata_to_chars([]) -> []; cdata_to_chars(B) when is_binary(B) -> case catch unicode:characters_to_list(B) of L when is_list(L) -> L; _ -> binary_to_list(B) end. %% string(String, Field, Adjust, Precision, PadChar) string(S, none, _Adj, none, _Pad) -> S; string(S, F, Adj, none, Pad) -> string_field(S, F, Adj, lists:flatlength(S), Pad); string(S, none, _Adj, P, Pad) -> string_field(S, P, left, lists:flatlength(S), Pad); string(S, F, Adj, P, Pad) when F >= P -> N = lists:flatlength(S), if F > P -> if N > P -> adjust(flat_trunc(S, P), chars(Pad, F-P), Adj); N < P -> adjust([S|chars(Pad, P-N)], chars(Pad, F-P), Adj); true -> % N == P adjust(S, chars(Pad, F-P), Adj) end; true -> % F == P string_field(S, F, Adj, N, Pad) end. string_field(S, F, _Adj, N, _Pad) when N > F -> flat_trunc(S, F); string_field(S, F, Adj, N, Pad) when N < F -> adjust(S, chars(Pad, F-N), Adj); string_field(S, _, _, _, _) -> % N == F S. %% unprefixed_integer(Int, Field, Adjust, Base, PadChar, Lowercase) %% -> [Char]. unprefixed_integer(Int, F, Adj, Base, Pad, Lowercase) when Base >= 2, Base =< 1+$Z-$A+10 -> if Int < 0 -> S = cond_lowercase(erlang:integer_to_list(-Int, Base), Lowercase), term([$-|S], F, Adj, none, Pad); true -> S = cond_lowercase(erlang:integer_to_list(Int, Base), Lowercase), term(S, F, Adj, none, Pad) end. %% prefixed_integer(Int, Field, Adjust, Base, PadChar, Prefix, Lowercase) %% -> [Char]. prefixed_integer(Int, F, Adj, Base, Pad, Prefix, Lowercase) when Base >= 2, Base =< 1+$Z-$A+10 -> if Int < 0 -> S = cond_lowercase(erlang:integer_to_list(-Int, Base), Lowercase), term([$-,Prefix|S], F, Adj, none, Pad); true -> S = cond_lowercase(erlang:integer_to_list(Int, Base), Lowercase), term([Prefix|S], F, Adj, none, Pad) end. %% char(Char, Field, Adjust, Precision, PadChar) -> string(). char(C, none, _Adj, none, _Pad) -> [C]; char(C, F, _Adj, none, _Pad) -> chars(C, F); char(C, none, _Adj, P, _Pad) -> chars(C, P); char(C, F, Adj, P, Pad) when F >= P -> adjust(chars(C, P), chars(Pad, F - P), Adj). %% newline(Field, Adjust, Precision, PadChar) -> [Char]. newline(none, _Adj, _P, _Pad) -> "\n"; newline(F, right, _P, _Pad) -> chars($\n, F). %% %% Utilities %% adjust(Data, [], _) -> Data; adjust(Data, Pad, left) -> [Data|Pad]; adjust(Data, Pad, right) -> [Pad|Data]. %% Flatten and truncate a deep list to at most N elements. flat_trunc(List, N) when is_integer(N), N >= 0 -> flat_trunc(List, N, [], []). flat_trunc(L, 0, _, R) when is_list(L) -> lists:reverse(R); flat_trunc([H|T], N, S, R) when is_list(H) -> flat_trunc(H, N, [T|S], R); flat_trunc([H|T], N, S, R) -> flat_trunc(T, N-1, S, [H|R]); flat_trunc([], N, [H|S], R) -> flat_trunc(H, N, S, R); flat_trunc([], _, [], R) -> lists:reverse(R). %% A deep version of string:chars/2,3 chars(_C, 0) -> []; chars(C, 1) -> [C]; chars(C, 2) -> [C,C]; chars(C, 3) -> [C,C,C]; chars(C, N) when is_integer(N), (N band 1) =:= 0 -> S = chars(C, N bsr 1), [S|S]; chars(C, N) when is_integer(N) -> S = chars(C, N bsr 1), [C,S|S]. %chars(C, N, Tail) -> % [chars(C, N)|Tail]. %% Lowercase conversion cond_lowercase(String, true) -> lowercase(String); cond_lowercase(String,false) -> String. lowercase([H|T]) when is_integer(H), H >= $A, H =< $Z -> [(H-$A+$a)|lowercase(T)]; lowercase([H|T]) -> [H|lowercase(T)]; lowercase([]) -> [].