%% Copyright (c) 2011-2013, Loïc Hoguin <[email protected]>
%% Copyright (c) 2011, Anthony Ramine <[email protected]>
%%
%% Permission to use, copy, modify, and/or distribute this software for any
%% purpose with or without fee is hereby granted, provided that the above
%% copyright notice and this permission notice appear in all copies.
%%
%% THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
%% WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
%% MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
%% ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
%% WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
%% ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
%% OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
%% @doc Core HTTP parsing API.
-module(cowboy_http).
%% Parsing.
-export([list/2]).
-export([nonempty_list/2]).
-export([content_type/1]).
-export([media_range/2]).
-export([conneg/2]).
-export([language_range/2]).
-export([entity_tag_match/1]).
-export([expectation/2]).
-export([params/2]).
-export([http_date/1]).
-export([rfc1123_date/1]).
-export([rfc850_date/1]).
-export([asctime_date/1]).
-export([whitespace/2]).
-export([digits/1]).
-export([token/2]).
-export([token_ci/2]).
-export([quoted_string/2]).
-export([authorization/2]).
-export([range/1]).
-export([parameterized_tokens/1]).
%% Decoding.
-export([te_chunked/2]).
-export([te_identity/2]).
-export([ce_identity/1]).
%% Interpretation.
-export([urldecode/1]).
-export([urldecode/2]).
-export([urlencode/1]).
-export([urlencode/2]).
%% Parsing.
%% @doc Parse a non-empty list of the given type.
-spec nonempty_list(binary(), fun()) -> [any(), ...] | {error, badarg}.
nonempty_list(Data, Fun) ->
case list(Data, Fun, []) of
{error, badarg} -> {error, badarg};
[] -> {error, badarg};
L -> lists:reverse(L)
end.
%% @doc Parse a list of the given type.
-spec list(binary(), fun()) -> list() | {error, badarg}.
list(Data, Fun) ->
case list(Data, Fun, []) of
{error, badarg} -> {error, badarg};
L -> lists:reverse(L)
end.
-spec list(binary(), fun(), [binary()]) -> [any()] | {error, badarg}.
%% From the RFC:
%% <blockquote>Wherever this construct is used, null elements are allowed,
%% but do not contribute to the count of elements present.
%% That is, "(element), , (element) " is permitted, but counts
%% as only two elements. Therefore, where at least one element is required,
%% at least one non-null element MUST be present.</blockquote>
list(Data, Fun, Acc) ->
whitespace(Data,
fun (<<>>) -> Acc;
(<< $,, Rest/binary >>) -> list(Rest, Fun, Acc);
(Rest) -> Fun(Rest,
fun (D, I) -> whitespace(D,
fun (<<>>) -> [I|Acc];
(<< $,, R/binary >>) -> list(R, Fun, [I|Acc]);
(_Any) -> {error, badarg}
end)
end)
end).
%% @doc Parse a content type.
%%
%% We lowercase the charset header as we know it's case insensitive.
-spec content_type(binary()) -> any().
content_type(Data) ->
media_type(Data,
fun (Rest, Type, SubType) ->
params(Rest,
fun (<<>>, Params) ->
case lists:keyfind(<<"charset">>, 1, Params) of
false ->
{Type, SubType, Params};
{_, Charset} ->
Charset2 = cowboy_bstr:to_lower(Charset),
Params2 = lists:keyreplace(<<"charset">>,
1, Params, {<<"charset">>, Charset2}),
{Type, SubType, Params2}
end;
(_Rest2, _) ->
{error, badarg}
end)
end).
%% @doc Parse a media range.
-spec media_range(binary(), fun()) -> any().
media_range(Data, Fun) ->
media_type(Data,
fun (Rest, Type, SubType) ->
media_range_params(Rest, Fun, Type, SubType, [])
end).
-spec media_range_params(binary(), fun(), binary(), binary(),
[{binary(), binary()}]) -> any().
media_range_params(Data, Fun, Type, SubType, Acc) ->
whitespace(Data,
fun (<< $;, Rest/binary >>) ->
whitespace(Rest,
fun (Rest2) ->
media_range_param_attr(Rest2, Fun, Type, SubType, Acc)
end);
(Rest) -> Fun(Rest, {{Type, SubType, lists:reverse(Acc)}, 1000, []})
end).
-spec media_range_param_attr(binary(), fun(), binary(), binary(),
[{binary(), binary()}]) -> any().
media_range_param_attr(Data, Fun, Type, SubType, Acc) ->
token_ci(Data,
fun (_Rest, <<>>) -> {error, badarg};
(<< $=, Rest/binary >>, Attr) ->
media_range_param_value(Rest, Fun, Type, SubType, Acc, Attr)
end).
-spec media_range_param_value(binary(), fun(), binary(), binary(),
[{binary(), binary()}], binary()) -> any().
media_range_param_value(Data, Fun, Type, SubType, Acc, <<"q">>) ->
qvalue(Data,
fun (Rest, Quality) ->
accept_ext(Rest, Fun, Type, SubType, Acc, Quality, [])
end);
media_range_param_value(Data, Fun, Type, SubType, Acc, Attr) ->
word(Data,
fun (Rest, Value) ->
media_range_params(Rest, Fun,
Type, SubType, [{Attr, Value}|Acc])
end).
%% @doc Parse a media type.
-spec media_type(binary(), fun()) -> any().
media_type(Data, Fun) ->
token_ci(Data,
fun (_Rest, <<>>) -> {error, badarg};
(<< $/, Rest/binary >>, Type) ->
token_ci(Rest,
fun (_Rest2, <<>>) -> {error, badarg};
(Rest2, SubType) -> Fun(Rest2, Type, SubType)
end);
%% This is a non-strict parsing clause required by some user agents
%% that use * instead of */* in the list of media types.
(Rest, <<"*">> = Type) ->
token_ci(<<"*", Rest/binary>>,
fun (_Rest2, <<>>) -> {error, badarg};
(Rest2, SubType) -> Fun(Rest2, Type, SubType)
end);
(_Rest, _Type) -> {error, badarg}
end).
-spec accept_ext(binary(), fun(), binary(), binary(),
[{binary(), binary()}], 0..1000,
[{binary(), binary()} | binary()]) -> any().
accept_ext(Data, Fun, Type, SubType, Params, Quality, Acc) ->
whitespace(Data,
fun (<< $;, Rest/binary >>) ->
whitespace(Rest,
fun (Rest2) ->
accept_ext_attr(Rest2, Fun,
Type, SubType, Params, Quality, Acc)
end);
(Rest) ->
Fun(Rest, {{Type, SubType, lists:reverse(Params)},
Quality, lists:reverse(Acc)})
end).
-spec accept_ext_attr(binary(), fun(), binary(), binary(),
[{binary(), binary()}], 0..1000,
[{binary(), binary()} | binary()]) -> any().
accept_ext_attr(Data, Fun, Type, SubType, Params, Quality, Acc) ->
token_ci(Data,
fun (_Rest, <<>>) -> {error, badarg};
(<< $=, Rest/binary >>, Attr) ->
accept_ext_value(Rest, Fun, Type, SubType, Params,
Quality, Acc, Attr);
(Rest, Attr) ->
accept_ext(Rest, Fun, Type, SubType, Params,
Quality, [Attr|Acc])
end).
-spec accept_ext_value(binary(), fun(), binary(), binary(),
[{binary(), binary()}], 0..1000,
[{binary(), binary()} | binary()], binary()) -> any().
accept_ext_value(Data, Fun, Type, SubType, Params, Quality, Acc, Attr) ->
word(Data,
fun (Rest, Value) ->
accept_ext(Rest, Fun,
Type, SubType, Params, Quality, [{Attr, Value}|Acc])
end).
%% @doc Parse a conneg header (Accept-Charset, Accept-Encoding),
%% followed by an optional quality value.
-spec conneg(binary(), fun()) -> any().
conneg(Data, Fun) ->
token_ci(Data,
fun (_Rest, <<>>) -> {error, badarg};
(Rest, Conneg) ->
maybe_qparam(Rest,
fun (Rest2, Quality) ->
Fun(Rest2, {Conneg, Quality})
end)
end).
%% @doc Parse a language range, followed by an optional quality value.
-spec language_range(binary(), fun()) -> any().
language_range(<< $*, Rest/binary >>, Fun) ->
language_range_ret(Rest, Fun, '*');
language_range(Data, Fun) ->
language_tag(Data,
fun (Rest, LanguageTag) ->
language_range_ret(Rest, Fun, LanguageTag)
end).
-spec language_range_ret(binary(), fun(), '*' | {binary(), [binary()]}) -> any().
language_range_ret(Data, Fun, LanguageTag) ->
maybe_qparam(Data,
fun (Rest, Quality) ->
Fun(Rest, {LanguageTag, Quality})
end).
-spec language_tag(binary(), fun()) -> any().
language_tag(Data, Fun) ->
alpha(Data,
fun (_Rest, Tag) when byte_size(Tag) =:= 0; byte_size(Tag) > 8 ->
{error, badarg};
(<< $-, Rest/binary >>, Tag) ->
language_subtag(Rest, Fun, Tag, []);
(Rest, Tag) ->
Fun(Rest, Tag)
end).
-spec language_subtag(binary(), fun(), binary(), [binary()]) -> any().
language_subtag(Data, Fun, Tag, Acc) ->
alpha(Data,
fun (_Rest, SubTag) when byte_size(SubTag) =:= 0;
byte_size(SubTag) > 8 -> {error, badarg};
(<< $-, Rest/binary >>, SubTag) ->
language_subtag(Rest, Fun, Tag, [SubTag|Acc]);
(Rest, SubTag) ->
%% Rebuild the full tag now that we know it's correct
Sub = << << $-, S/binary >> || S <- lists:reverse([SubTag|Acc]) >>,
Fun(Rest, << Tag/binary, Sub/binary >>)
end).
-spec maybe_qparam(binary(), fun()) -> any().
maybe_qparam(Data, Fun) ->
whitespace(Data,
fun (<< $;, Rest/binary >>) ->
whitespace(Rest,
fun (Rest2) ->
%% This is a non-strict parsing clause required by some user agents
%% that use the wrong delimiter putting a charset where a qparam is
%% expected.
try qparam(Rest2, Fun) of
Result -> Result
catch
error:function_clause ->
Fun(<<",", Rest2/binary>>, 1000)
end
end);
(Rest) ->
Fun(Rest, 1000)
end).
%% @doc Parse a quality parameter string (for example q=0.500).
-spec qparam(binary(), fun()) -> any().
qparam(<< Q, $=, Data/binary >>, Fun) when Q =:= $q; Q =:= $Q ->
qvalue(Data, Fun).
%% @doc Parse either a list of entity tags or a "*".
-spec entity_tag_match(binary()) -> any().
entity_tag_match(<< $*, Rest/binary >>) ->
whitespace(Rest,
fun (<<>>) -> '*';
(_Any) -> {error, badarg}
end);
entity_tag_match(Data) ->
nonempty_list(Data, fun entity_tag/2).
%% @doc Parse an entity-tag.
-spec entity_tag(binary(), fun()) -> any().
entity_tag(<< "W/", Rest/binary >>, Fun) ->
opaque_tag(Rest, Fun, weak);
entity_tag(Data, Fun) ->
opaque_tag(Data, Fun, strong).
-spec opaque_tag(binary(), fun(), weak | strong) -> any().
opaque_tag(Data, Fun, Strength) ->
quoted_string(Data,
fun (_Rest, <<>>) -> {error, badarg};
(Rest, OpaqueTag) -> Fun(Rest, {Strength, OpaqueTag})
end).
%% @doc Parse an expectation.
-spec expectation(binary(), fun()) -> any().
expectation(Data, Fun) ->
token_ci(Data,
fun (_Rest, <<>>) -> {error, badarg};
(<< $=, Rest/binary >>, Expectation) ->
word(Rest,
fun (Rest2, ExtValue) ->
params(Rest2, fun (Rest3, ExtParams) ->
Fun(Rest3, {Expectation, ExtValue, ExtParams})
end)
end);
(Rest, Expectation) ->
Fun(Rest, Expectation)
end).
%% @doc Parse a list of parameters (a=b;c=d).
-spec params(binary(), fun()) -> any().
params(Data, Fun) ->
params(Data, Fun, []).
-spec params(binary(), fun(), [{binary(), binary()}]) -> any().
params(Data, Fun, Acc) ->
whitespace(Data,
fun (<< $;, Rest/binary >>) ->
param(Rest,
fun (Rest2, Attr, Value) ->
params(Rest2, Fun, [{Attr, Value}|Acc])
end);
(Rest) ->
Fun(Rest, lists:reverse(Acc))
end).
-spec param(binary(), fun()) -> any().
param(Data, Fun) ->
whitespace(Data,
fun (Rest) ->
token_ci(Rest,
fun (_Rest2, <<>>) -> {error, badarg};
(<< $=, Rest2/binary >>, Attr) ->
word(Rest2,
fun (Rest3, Value) ->
Fun(Rest3, Attr, Value)
end);
(_Rest2, _Attr) -> {error, badarg}
end)
end).
%% @doc Parse an HTTP date (RFC1123, RFC850 or asctime date).
%% @end
%%
%% While this may not be the most efficient date parsing we can do,
%% it should work fine for our purposes because all HTTP dates should
%% be sent as RFC1123 dates in HTTP/1.1.
-spec http_date(binary()) -> any().
http_date(Data) ->
case rfc1123_date(Data) of
{error, badarg} ->
case rfc850_date(Data) of
{error, badarg} ->
case asctime_date(Data) of
{error, badarg} ->
{error, badarg};
HTTPDate ->
HTTPDate
end;
HTTPDate ->
HTTPDate
end;
HTTPDate ->
HTTPDate
end.
%% @doc Parse an RFC1123 date.
-spec rfc1123_date(binary()) -> any().
rfc1123_date(Data) ->
wkday(Data,
fun (<< ", ", Rest/binary >>, _WkDay) ->
date1(Rest,
fun (<< " ", Rest2/binary >>, Date) ->
time(Rest2,
fun (<< " GMT", Rest3/binary >>, Time) ->
http_date_ret(Rest3, {Date, Time});
(_Any, _Time) ->
{error, badarg}
end);
(_Any, _Date) ->
{error, badarg}
end);
(_Any, _WkDay) ->
{error, badarg}
end).
%% @doc Parse an RFC850 date.
-spec rfc850_date(binary()) -> any().
%% From the RFC:
%% HTTP/1.1 clients and caches SHOULD assume that an RFC-850 date
%% which appears to be more than 50 years in the future is in fact
%% in the past (this helps solve the "year 2000" problem).
rfc850_date(Data) ->
weekday(Data,
fun (<< ", ", Rest/binary >>, _WeekDay) ->
date2(Rest,
fun (<< " ", Rest2/binary >>, Date) ->
time(Rest2,
fun (<< " GMT", Rest3/binary >>, Time) ->
http_date_ret(Rest3, {Date, Time});
(_Any, _Time) ->
{error, badarg}
end);
(_Any, _Date) ->
{error, badarg}
end);
(_Any, _WeekDay) ->
{error, badarg}
end).
%% @doc Parse an asctime date.
-spec asctime_date(binary()) -> any().
asctime_date(Data) ->
wkday(Data,
fun (<< " ", Rest/binary >>, _WkDay) ->
date3(Rest,
fun (<< " ", Rest2/binary >>, PartialDate) ->
time(Rest2,
fun (<< " ", Rest3/binary >>, Time) ->
asctime_year(Rest3,
PartialDate, Time);
(_Any, _Time) ->
{error, badarg}
end);
(_Any, _PartialDate) ->
{error, badarg}
end);
(_Any, _WkDay) ->
{error, badarg1}
end).
-spec asctime_year(binary(), tuple(), tuple()) -> any().
asctime_year(<< Y1, Y2, Y3, Y4, Rest/binary >>, {Month, Day}, Time)
when Y1 >= $0, Y1 =< $9, Y2 >= $0, Y2 =< $9,
Y3 >= $0, Y3 =< $9, Y4 >= $0, Y4 =< $9 ->
Year = (Y1 - $0) * 1000 + (Y2 - $0) * 100 + (Y3 - $0) * 10 + (Y4 - $0),
http_date_ret(Rest, {{Year, Month, Day}, Time}).
-spec http_date_ret(binary(), tuple()) -> any().
http_date_ret(Data, DateTime = {Date, _Time}) ->
whitespace(Data,
fun (<<>>) ->
case calendar:valid_date(Date) of
true -> DateTime;
false -> {error, badarg}
end;
(_Any) ->
{error, badarg}
end).
%% We never use it, pretty much just checks the wkday is right.
-spec wkday(binary(), fun()) -> any().
wkday(<< WkDay:3/binary, Rest/binary >>, Fun)
when WkDay =:= <<"Mon">>; WkDay =:= <<"Tue">>; WkDay =:= <<"Wed">>;
WkDay =:= <<"Thu">>; WkDay =:= <<"Fri">>; WkDay =:= <<"Sat">>;
WkDay =:= <<"Sun">> ->
Fun(Rest, WkDay);
wkday(_Any, _Fun) ->
{error, badarg}.
%% We never use it, pretty much just checks the weekday is right.
-spec weekday(binary(), fun()) -> any().
weekday(<< "Monday", Rest/binary >>, Fun) ->
Fun(Rest, <<"Monday">>);
weekday(<< "Tuesday", Rest/binary >>, Fun) ->
Fun(Rest, <<"Tuesday">>);
weekday(<< "Wednesday", Rest/binary >>, Fun) ->
Fun(Rest, <<"Wednesday">>);
weekday(<< "Thursday", Rest/binary >>, Fun) ->
Fun(Rest, <<"Thursday">>);
weekday(<< "Friday", Rest/binary >>, Fun) ->
Fun(Rest, <<"Friday">>);
weekday(<< "Saturday", Rest/binary >>, Fun) ->
Fun(Rest, <<"Saturday">>);
weekday(<< "Sunday", Rest/binary >>, Fun) ->
Fun(Rest, <<"Sunday">>);
weekday(_Any, _Fun) ->
{error, badarg}.
-spec date1(binary(), fun()) -> any().
date1(<< D1, D2, " ", M:3/binary, " ", Y1, Y2, Y3, Y4, Rest/binary >>, Fun)
when D1 >= $0, D1 =< $9, D2 >= $0, D2 =< $9,
Y1 >= $0, Y1 =< $9, Y2 >= $0, Y2 =< $9,
Y3 >= $0, Y3 =< $9, Y4 >= $0, Y4 =< $9 ->
case month(M) of
{error, badarg} ->
{error, badarg};
Month ->
Fun(Rest, {
(Y1 - $0) * 1000 + (Y2 - $0) * 100 + (Y3 - $0) * 10 + (Y4 - $0),
Month,
(D1 - $0) * 10 + (D2 - $0)
})
end;
date1(_Data, _Fun) ->
{error, badarg}.
-spec date2(binary(), fun()) -> any().
date2(<< D1, D2, "-", M:3/binary, "-", Y1, Y2, Rest/binary >>, Fun)
when D1 >= $0, D1 =< $9, D2 >= $0, D2 =< $9,
Y1 >= $0, Y1 =< $9, Y2 >= $0, Y2 =< $9 ->
case month(M) of
{error, badarg} ->
{error, badarg};
Month ->
Year = (Y1 - $0) * 10 + (Y2 - $0),
Year2 = case Year > 50 of
true -> Year + 1900;
false -> Year + 2000
end,
Fun(Rest, {
Year2,
Month,
(D1 - $0) * 10 + (D2 - $0)
})
end;
date2(_Data, _Fun) ->
{error, badarg}.
-spec date3(binary(), fun()) -> any().
date3(<< M:3/binary, " ", D1, D2, Rest/binary >>, Fun)
when (D1 >= $0 andalso D1 =< $3) orelse D1 =:= $\s,
D2 >= $0, D2 =< $9 ->
case month(M) of
{error, badarg} ->
{error, badarg};
Month ->
Day = case D1 of
$\s -> D2 - $0;
D1 -> (D1 - $0) * 10 + (D2 - $0)
end,
Fun(Rest, {Month, Day})
end;
date3(_Data, _Fun) ->
{error, badarg}.
-spec month(<< _:24 >>) -> 1..12 | {error, badarg}.
month(<<"Jan">>) -> 1;
month(<<"Feb">>) -> 2;
month(<<"Mar">>) -> 3;
month(<<"Apr">>) -> 4;
month(<<"May">>) -> 5;
month(<<"Jun">>) -> 6;
month(<<"Jul">>) -> 7;
month(<<"Aug">>) -> 8;
month(<<"Sep">>) -> 9;
month(<<"Oct">>) -> 10;
month(<<"Nov">>) -> 11;
month(<<"Dec">>) -> 12;
month(_Any) -> {error, badarg}.
-spec time(binary(), fun()) -> any().
time(<< H1, H2, ":", M1, M2, ":", S1, S2, Rest/binary >>, Fun)
when H1 >= $0, H1 =< $2, H2 >= $0, H2 =< $9,
M1 >= $0, M1 =< $5, M2 >= $0, M2 =< $9,
S1 >= $0, S1 =< $5, S2 >= $0, S2 =< $9 ->
Hour = (H1 - $0) * 10 + (H2 - $0),
case Hour < 24 of
true ->
Time = {
Hour,
(M1 - $0) * 10 + (M2 - $0),
(S1 - $0) * 10 + (S2 - $0)
},
Fun(Rest, Time);
false ->
{error, badarg}
end.
%% @doc Skip whitespace.
-spec whitespace(binary(), fun()) -> any().
whitespace(<< C, Rest/binary >>, Fun)
when C =:= $\s; C =:= $\t ->
whitespace(Rest, Fun);
whitespace(Data, Fun) ->
Fun(Data).
%% @doc Parse a list of digits as a non negative integer.
-spec digits(binary()) -> non_neg_integer() | {error, badarg}.
digits(Data) ->
digits(Data,
fun (Rest, I) ->
whitespace(Rest,
fun (<<>>) ->
I;
(_Rest2) ->
{error, badarg}
end)
end).
-spec digits(binary(), fun()) -> any().
digits(<< C, Rest/binary >>, Fun)
when C >= $0, C =< $9 ->
digits(Rest, Fun, C - $0);
digits(_Data, _Fun) ->
{error, badarg}.
-spec digits(binary(), fun(), non_neg_integer()) -> any().
digits(<< C, Rest/binary >>, Fun, Acc)
when C >= $0, C =< $9 ->
digits(Rest, Fun, Acc * 10 + (C - $0));
digits(Data, Fun, Acc) ->
Fun(Data, Acc).
%% @doc Parse a list of case-insensitive alpha characters.
%%
%% Changes all characters to lowercase.
-spec alpha(binary(), fun()) -> any().
alpha(Data, Fun) ->
alpha(Data, Fun, <<>>).
-spec alpha(binary(), fun(), binary()) -> any().
alpha(<<>>, Fun, Acc) ->
Fun(<<>>, Acc);
alpha(<< C, Rest/binary >>, Fun, Acc)
when C >= $a andalso C =< $z;
C >= $A andalso C =< $Z ->
C2 = cowboy_bstr:char_to_lower(C),
alpha(Rest, Fun, << Acc/binary, C2 >>);
alpha(Data, Fun, Acc) ->
Fun(Data, Acc).
%% @doc Parse either a token or a quoted string.
-spec word(binary(), fun()) -> any().
word(Data = << $", _/binary >>, Fun) ->
quoted_string(Data, Fun);
word(Data, Fun) ->
token(Data,
fun (_Rest, <<>>) -> {error, badarg};
(Rest, Token) -> Fun(Rest, Token)
end).
%% @doc Parse a case-insensitive token.
%%
%% Changes all characters to lowercase.
-spec token_ci(binary(), fun()) -> any().
token_ci(Data, Fun) ->
token(Data, Fun, ci, <<>>).
%% @doc Parse a token.
-spec token(binary(), fun()) -> any().
token(Data, Fun) ->
token(Data, Fun, cs, <<>>).
-spec token(binary(), fun(), ci | cs, binary()) -> any().
token(<<>>, Fun, _Case, Acc) ->
Fun(<<>>, Acc);
token(Data = << C, _Rest/binary >>, Fun, _Case, Acc)
when C =:= $(; C =:= $); C =:= $<; C =:= $>; C =:= $@;
C =:= $,; C =:= $;; C =:= $:; C =:= $\\; C =:= $";
C =:= $/; C =:= $[; C =:= $]; C =:= $?; C =:= $=;
C =:= ${; C =:= $}; C =:= $\s; C =:= $\t;
C < 32; C =:= 127 ->
Fun(Data, Acc);
token(<< C, Rest/binary >>, Fun, Case = ci, Acc) ->
C2 = cowboy_bstr:char_to_lower(C),
token(Rest, Fun, Case, << Acc/binary, C2 >>);
token(<< C, Rest/binary >>, Fun, Case, Acc) ->
token(Rest, Fun, Case, << Acc/binary, C >>).
%% @doc Parse a quoted string.
-spec quoted_string(binary(), fun()) -> any().
quoted_string(<< $", Rest/binary >>, Fun) ->
quoted_string(Rest, Fun, <<>>).
-spec quoted_string(binary(), fun(), binary()) -> any().
quoted_string(<<>>, _Fun, _Acc) ->
{error, badarg};
quoted_string(<< $", Rest/binary >>, Fun, Acc) ->
Fun(Rest, Acc);
quoted_string(<< $\\, C, Rest/binary >>, Fun, Acc) ->
quoted_string(Rest, Fun, << Acc/binary, C >>);
quoted_string(<< C, Rest/binary >>, Fun, Acc) ->
quoted_string(Rest, Fun, << Acc/binary, C >>).
%% @doc Parse a quality value.
-spec qvalue(binary(), fun()) -> any().
qvalue(<< $0, $., Rest/binary >>, Fun) ->
qvalue(Rest, Fun, 0, 100);
%% Some user agents use q=.x instead of q=0.x
qvalue(<< $., Rest/binary >>, Fun) ->
qvalue(Rest, Fun, 0, 100);
qvalue(<< $0, Rest/binary >>, Fun) ->
Fun(Rest, 0);
qvalue(<< $1, $., $0, $0, $0, Rest/binary >>, Fun) ->
Fun(Rest, 1000);
qvalue(<< $1, $., $0, $0, Rest/binary >>, Fun) ->
Fun(Rest, 1000);
qvalue(<< $1, $., $0, Rest/binary >>, Fun) ->
Fun(Rest, 1000);
qvalue(<< $1, Rest/binary >>, Fun) ->
Fun(Rest, 1000);
qvalue(_Data, _Fun) ->
{error, badarg}.
-spec qvalue(binary(), fun(), integer(), 1 | 10 | 100) -> any().
qvalue(Data, Fun, Q, 0) ->
Fun(Data, Q);
qvalue(<< C, Rest/binary >>, Fun, Q, M)
when C >= $0, C =< $9 ->
qvalue(Rest, Fun, Q + (C - $0) * M, M div 10);
qvalue(Data, Fun, Q, _M) ->
Fun(Data, Q).
%% @doc Parse authorization value according rfc 2617.
%% Only Basic authorization is supported so far.
-spec authorization(binary(), binary()) -> {binary(), any()} | {error, badarg}.
authorization(UserPass, Type = <<"basic">>) ->
whitespace(UserPass,
fun(D) ->
authorization_basic_userid(base64:mime_decode(D),
fun(Rest, Userid) ->
authorization_basic_password(Rest,
fun(Password) ->
{Type, {Userid, Password}}
end)
end)
end);
authorization(String, Type) ->
whitespace(String, fun(Rest) -> {Type, Rest} end).
%% @doc Parse user credentials.
-spec authorization_basic_userid(binary(), fun()) -> any().
authorization_basic_userid(Data, Fun) ->
authorization_basic_userid(Data, Fun, <<>>).
authorization_basic_userid(<<>>, _Fun, _Acc) ->
{error, badarg};
authorization_basic_userid(<<C, _Rest/binary>>, _Fun, Acc)
when C < 32; C =:= 127; (C =:=$: andalso Acc =:= <<>>) ->
{error, badarg};
authorization_basic_userid(<<$:, Rest/binary>>, Fun, Acc) ->
Fun(Rest, Acc);
authorization_basic_userid(<<C, Rest/binary>>, Fun, Acc) ->
authorization_basic_userid(Rest, Fun, <<Acc/binary, C>>).
-spec authorization_basic_password(binary(), fun()) -> any().
authorization_basic_password(Data, Fun) ->
authorization_basic_password(Data, Fun, <<>>).
authorization_basic_password(<<>>, _Fun, <<>>) ->
{error, badarg};
authorization_basic_password(<<C, _Rest/binary>>, _Fun, _Acc)
when C < 32; C=:= 127 ->
{error, badarg};
authorization_basic_password(<<>>, Fun, Acc) ->
Fun(Acc);
authorization_basic_password(<<C, Rest/binary>>, Fun, Acc) ->
authorization_basic_password(Rest, Fun, <<Acc/binary, C>>).
%% @doc Parse range header according rfc 2616.
-spec range(binary()) -> {Unit, [Range]} | {error, badarg} when
Unit :: binary(),
Range :: {non_neg_integer(), non_neg_integer() | infinity} | neg_integer().
range(Data) ->
token_ci(Data, fun range/2).
range(Data, Token) ->
whitespace(Data,
fun(<<"=", Rest/binary>>) ->
case list(Rest, fun range_beginning/2) of
{error, badarg} ->
{error, badarg};
Ranges ->
{Token, Ranges}
end;
(_) ->
{error, badarg}
end).
range_beginning(Data, Fun) ->
range_digits(Data, suffix,
fun(D, RangeBeginning) ->
range_ending(D, Fun, RangeBeginning)
end).
range_ending(Data, Fun, RangeBeginning) ->
whitespace(Data,
fun(<<"-", R/binary>>) ->
case RangeBeginning of
suffix ->
range_digits(R, fun(D, RangeEnding) -> Fun(D, -RangeEnding) end);
_ ->
range_digits(R, infinity,
fun(D, RangeEnding) ->
Fun(D, {RangeBeginning, RangeEnding})
end)
end;
(_) ->
{error, badarg}
end).
-spec range_digits(binary(), fun()) -> any().
range_digits(Data, Fun) ->
whitespace(Data,
fun(D) ->
digits(D, Fun)
end).
-spec range_digits(binary(), any(), fun()) -> any().
range_digits(Data, Default, Fun) ->
whitespace(Data,
fun(<< C, Rest/binary >>) when C >= $0, C =< $9 ->
digits(Rest, Fun, C - $0);
(_) ->
Fun(Data, Default)
end).
%% @doc Parse a non empty list of tokens followed with optional parameters.
-spec parameterized_tokens(binary()) -> any().
parameterized_tokens(Data) ->
nonempty_list(Data,
fun (D, Fun) ->
token(D,
fun (_Rest, <<>>) -> {error, badarg};
(Rest, Token) ->
parameterized_tokens_params(Rest,
fun (Rest2, Params) ->
Fun(Rest2, {Token, Params})
end, [])
end)
end).
-spec parameterized_tokens_params(binary(), fun(), [binary() | {binary(), binary()}]) -> any().
parameterized_tokens_params(Data, Fun, Acc) ->
whitespace(Data,
fun (<< $;, Rest/binary >>) ->
parameterized_tokens_param(Rest,
fun (Rest2, Param) ->
parameterized_tokens_params(Rest2, Fun, [Param|Acc])
end);
(Rest) ->
Fun(Rest, lists:reverse(Acc))
end).
-spec parameterized_tokens_param(binary(), fun()) -> any().
parameterized_tokens_param(Data, Fun) ->
whitespace(Data,
fun (Rest) ->
token(Rest,
fun (_Rest2, <<>>) -> {error, badarg};
(<< $=, Rest2/binary >>, Attr) ->
word(Rest2,
fun (Rest3, Value) ->
Fun(Rest3, {Attr, Value})
end);
(Rest2, Attr) ->
Fun(Rest2, Attr)
end)
end).
%% Decoding.
%% @doc Decode a stream of chunks.
-spec te_chunked(Bin, TransferState)
-> more | {more, non_neg_integer(), Bin, TransferState}
| {ok, Bin, Bin, TransferState}
| {done, non_neg_integer(), Bin} | {error, badarg}
when Bin::binary(), TransferState::{non_neg_integer(), non_neg_integer()}.
te_chunked(<< "0\r\n\r\n", Rest/binary >>, {0, Streamed}) ->
{done, Streamed, Rest};
te_chunked(Data, {0, Streamed}) ->
%% @todo We are expecting an hex size, not a general token.
token(Data,
fun (<< "\r\n", Rest/binary >>, BinLen) ->
case list_to_integer(binary_to_list(BinLen), 16) of
%% Final chunk is parsed in one go above. Rest would be
%% <<\r\n">> if complete.
0 when byte_size(Rest) < 2 ->
more;
%% Normal chunk. Add 2 to Len for trailing <<"\r\n">>. Note
%% that repeated <<"-2\r\n">> would be streamed, and
%% accumulated, until out of memory if Len could be -2.
Len when Len > 0 ->
te_chunked(Rest, {Len + 2, Streamed})
end;
%% Chunk size shouldn't take too many bytes,
%% don't try to stream forever.
(Rest, _) when byte_size(Rest) < 16 ->
more;
(_, _) ->
{error, badarg}
end);
%% <<"\n">> from trailing <<"\r\n">>.
te_chunked(<< "\n", Rest/binary>>, {1, Streamed}) ->
{ok, <<>>, Rest, {0, Streamed}};
te_chunked(<<>>, State={1, _Streamed}) ->
{more, 1, <<>>, State};
%% Remainder of chunk (if any) and as much of trailing <<"\r\n">> as possible.
te_chunked(Data, {ChunkRem, Streamed}) when byte_size(Data) >= ChunkRem - 2 ->
ChunkSize = ChunkRem - 2,
Streamed2 = Streamed + ChunkSize,
case Data of
<< Chunk:ChunkSize/binary, "\r\n", Rest/binary >> ->
{ok, Chunk, Rest, {0, Streamed2}};
<< Chunk:ChunkSize/binary, "\r" >> ->
{more, 1, Chunk, {1, Streamed2}};
<< Chunk:ChunkSize/binary >> ->
{more, 2, Chunk, {2, Streamed2}}
end;
%% Incomplete chunk.
te_chunked(Data, {ChunkRem, Streamed}) ->
ChunkRem2 = ChunkRem - byte_size(Data),
Streamed2 = Streamed + byte_size(Data),
{more, ChunkRem2, Data, {ChunkRem2, Streamed2}}.
%% @doc Decode an identity stream.
-spec te_identity(Bin, TransferState)
-> {more, non_neg_integer(), Bin, TransferState}
| {done, Bin, non_neg_integer(), Bin}
when Bin::binary(), TransferState::{non_neg_integer(), non_neg_integer()}.
te_identity(Data, {Streamed, Total})
when Streamed + byte_size(Data) < Total ->
Streamed2 = Streamed + byte_size(Data),
{more, Total - Streamed2, Data, {Streamed2, Total}};
te_identity(Data, {Streamed, Total}) ->
Size = Total - Streamed,
<< Data2:Size/binary, Rest/binary >> = Data,
{done, Data2, Total, Rest}.
%% @doc Decode an identity content.
-spec ce_identity(binary()) -> {ok, binary()}.
ce_identity(Data) ->
{ok, Data}.
%% Interpretation.
%% @doc Decode a URL encoded binary.
%% @equiv urldecode(Bin, crash)
-spec urldecode(binary()) -> binary().
urldecode(Bin) when is_binary(Bin) ->
urldecode(Bin, <<>>, crash).
%% @doc Decode a URL encoded binary.
%% The second argument specifies how to handle percent characters that are not
%% followed by two valid hex characters. Use `skip' to ignore such errors,
%% if `crash' is used the function will fail with the reason `badarg'.
-spec urldecode(binary(), crash | skip) -> binary().
urldecode(Bin, OnError) when is_binary(Bin) ->
urldecode(Bin, <<>>, OnError).
-spec urldecode(binary(), binary(), crash | skip) -> binary().
urldecode(<<$%, H, L, Rest/binary>>, Acc, OnError) ->
G = unhex(H),
M = unhex(L),
if G =:= error; M =:= error ->
case OnError of skip -> ok; crash -> erlang:error(badarg) end,
urldecode(<<H, L, Rest/binary>>, <<Acc/binary, $%>>, OnError);
true ->
urldecode(Rest, <<Acc/binary, (G bsl 4 bor M)>>, OnError)
end;
urldecode(<<$%, Rest/binary>>, Acc, OnError) ->
case OnError of skip -> ok; crash -> erlang:error(badarg) end,
urldecode(Rest, <<Acc/binary, $%>>, OnError);
urldecode(<<$+, Rest/binary>>, Acc, OnError) ->
urldecode(Rest, <<Acc/binary, $ >>, OnError);
urldecode(<<C, Rest/binary>>, Acc, OnError) ->
urldecode(Rest, <<Acc/binary, C>>, OnError);
urldecode(<<>>, Acc, _OnError) ->
Acc.
-spec unhex(byte()) -> byte() | error.
unhex(C) when C >= $0, C =< $9 -> C - $0;
unhex(C) when C >= $A, C =< $F -> C - $A + 10;
unhex(C) when C >= $a, C =< $f -> C - $a + 10;
unhex(_) -> error.
%% @doc URL encode a string binary.
%% @equiv urlencode(Bin, [])
-spec urlencode(binary()) -> binary().
urlencode(Bin) ->
urlencode(Bin, []).
%% @doc URL encode a string binary.
%% The `noplus' option disables the default behaviour of quoting space
%% characters, `\s', as `+'. The `upper' option overrides the default behaviour
%% of writing hex numbers using lowecase letters to using uppercase letters
%% instead.
-spec urlencode(binary(), [noplus|upper]) -> binary().
urlencode(Bin, Opts) ->
Plus = not lists:member(noplus, Opts),
Upper = lists:member(upper, Opts),
urlencode(Bin, <<>>, Plus, Upper).
-spec urlencode(binary(), binary(), boolean(), boolean()) -> binary().
urlencode(<<C, Rest/binary>>, Acc, P=Plus, U=Upper) ->
if C >= $0, C =< $9 -> urlencode(Rest, <<Acc/binary, C>>, P, U);
C >= $A, C =< $Z -> urlencode(Rest, <<Acc/binary, C>>, P, U);
C >= $a, C =< $z -> urlencode(Rest, <<Acc/binary, C>>, P, U);
C =:= $.; C =:= $-; C =:= $~; C =:= $_ ->
urlencode(Rest, <<Acc/binary, C>>, P, U);
C =:= $ , Plus ->
urlencode(Rest, <<Acc/binary, $+>>, P, U);
true ->
H = C band 16#F0 bsr 4, L = C band 16#0F,
H1 = if Upper -> tohexu(H); true -> tohexl(H) end,
L1 = if Upper -> tohexu(L); true -> tohexl(L) end,
urlencode(Rest, <<Acc/binary, $%, H1, L1>>, P, U)
end;
urlencode(<<>>, Acc, _Plus, _Upper) ->
Acc.
-spec tohexu(byte()) -> byte().
tohexu(C) when C < 10 -> $0 + C;
tohexu(C) when C < 17 -> $A + C - 10.
-spec tohexl(byte()) -> byte().
tohexl(C) when C < 10 -> $0 + C;
tohexl(C) when C < 17 -> $a + C - 10.
%% Tests.
-ifdef(TEST).
nonempty_charset_list_test_() ->
%% {Value, Result}
Tests = [
{<<>>, {error, badarg}},
{<<"iso-8859-5, unicode-1-1;q=0.8">>, [
{<<"iso-8859-5">>, 1000},
{<<"unicode-1-1">>, 800}
]},
%% Some user agents send this invalid value for the Accept-Charset header
{<<"ISO-8859-1;utf-8;q=0.7,*;q=0.7">>, [
{<<"iso-8859-1">>, 1000},
{<<"utf-8">>, 700},
{<<"*">>, 700}
]}
],
[{V, fun() -> R = nonempty_list(V, fun conneg/2) end} || {V, R} <- Tests].
nonempty_language_range_list_test_() ->
%% {Value, Result}
Tests = [
{<<"da, en-gb;q=0.8, en;q=0.7">>, [
{<<"da">>, 1000},
{<<"en-gb">>, 800},
{<<"en">>, 700}
]},
{<<"en, en-US, en-cockney, i-cherokee, x-pig-latin">>, [
{<<"en">>, 1000},
{<<"en-us">>, 1000},
{<<"en-cockney">>, 1000},
{<<"i-cherokee">>, 1000},
{<<"x-pig-latin">>, 1000}
]}
],
[{V, fun() -> R = nonempty_list(V, fun language_range/2) end}
|| {V, R} <- Tests].
nonempty_token_list_test_() ->
%% {Value, Result}
Tests = [
{<<>>, {error, badarg}},
{<<" ">>, {error, badarg}},
{<<" , ">>, {error, badarg}},
{<<",,,">>, {error, badarg}},
{<<"a b">>, {error, badarg}},
{<<"a , , , ">>, [<<"a">>]},
{<<" , , , a">>, [<<"a">>]},
{<<"a, , b">>, [<<"a">>, <<"b">>]},
{<<"close">>, [<<"close">>]},
{<<"keep-alive, upgrade">>, [<<"keep-alive">>, <<"upgrade">>]}
],
[{V, fun() -> R = nonempty_list(V, fun token/2) end} || {V, R} <- Tests].
media_range_list_test_() ->
%% {Tokens, Result}
Tests = [
{<<"audio/*; q=0.2, audio/basic">>, [
{{<<"audio">>, <<"*">>, []}, 200, []},
{{<<"audio">>, <<"basic">>, []}, 1000, []}
]},
{<<"text/plain; q=0.5, text/html, "
"text/x-dvi; q=0.8, text/x-c">>, [
{{<<"text">>, <<"plain">>, []}, 500, []},
{{<<"text">>, <<"html">>, []}, 1000, []},
{{<<"text">>, <<"x-dvi">>, []}, 800, []},
{{<<"text">>, <<"x-c">>, []}, 1000, []}
]},
{<<"text/*, text/html, text/html;level=1, */*">>, [
{{<<"text">>, <<"*">>, []}, 1000, []},
{{<<"text">>, <<"html">>, []}, 1000, []},
{{<<"text">>, <<"html">>, [{<<"level">>, <<"1">>}]}, 1000, []},
{{<<"*">>, <<"*">>, []}, 1000, []}
]},
{<<"text/*;q=0.3, text/html;q=0.7, text/html;level=1, "
"text/html;level=2;q=0.4, */*;q=0.5">>, [
{{<<"text">>, <<"*">>, []}, 300, []},
{{<<"text">>, <<"html">>, []}, 700, []},
{{<<"text">>, <<"html">>, [{<<"level">>, <<"1">>}]}, 1000, []},
{{<<"text">>, <<"html">>, [{<<"level">>, <<"2">>}]}, 400, []},
{{<<"*">>, <<"*">>, []}, 500, []}
]},
{<<"text/html;level=1;quoted=\"hi hi hi\";"
"q=0.123;standalone;complex=gits, text/plain">>, [
{{<<"text">>, <<"html">>,
[{<<"level">>, <<"1">>}, {<<"quoted">>, <<"hi hi hi">>}]}, 123,
[<<"standalone">>, {<<"complex">>, <<"gits">>}]},
{{<<"text">>, <<"plain">>, []}, 1000, []}
]},
{<<"text/html, image/gif, image/jpeg, *; q=.2, */*; q=.2">>, [
{{<<"text">>, <<"html">>, []}, 1000, []},
{{<<"image">>, <<"gif">>, []}, 1000, []},
{{<<"image">>, <<"jpeg">>, []}, 1000, []},
{{<<"*">>, <<"*">>, []}, 200, []},
{{<<"*">>, <<"*">>, []}, 200, []}
]}
],
[{V, fun() -> R = list(V, fun media_range/2) end} || {V, R} <- Tests].
entity_tag_match_test_() ->
%% {Tokens, Result}
Tests = [
{<<"\"xyzzy\"">>, [{strong, <<"xyzzy">>}]},
{<<"\"xyzzy\", W/\"r2d2xxxx\", \"c3piozzzz\"">>,
[{strong, <<"xyzzy">>},
{weak, <<"r2d2xxxx">>},
{strong, <<"c3piozzzz">>}]},
{<<"*">>, '*'}
],
[{V, fun() -> R = entity_tag_match(V) end} || {V, R} <- Tests].
http_date_test_() ->
%% {Tokens, Result}
Tests = [
{<<"Sun, 06 Nov 1994 08:49:37 GMT">>, {{1994, 11, 6}, {8, 49, 37}}},
{<<"Sunday, 06-Nov-94 08:49:37 GMT">>, {{1994, 11, 6}, {8, 49, 37}}},
{<<"Sun Nov 6 08:49:37 1994">>, {{1994, 11, 6}, {8, 49, 37}}}
],
[{V, fun() -> R = http_date(V) end} || {V, R} <- Tests].
rfc1123_date_test_() ->
%% {Tokens, Result}
Tests = [
{<<"Sun, 06 Nov 1994 08:49:37 GMT">>, {{1994, 11, 6}, {8, 49, 37}}}
],
[{V, fun() -> R = rfc1123_date(V) end} || {V, R} <- Tests].
rfc850_date_test_() ->
%% {Tokens, Result}
Tests = [
{<<"Sunday, 06-Nov-94 08:49:37 GMT">>, {{1994, 11, 6}, {8, 49, 37}}}
],
[{V, fun() -> R = rfc850_date(V) end} || {V, R} <- Tests].
asctime_date_test_() ->
%% {Tokens, Result}
Tests = [
{<<"Sun Nov 6 08:49:37 1994">>, {{1994, 11, 6}, {8, 49, 37}}}
],
[{V, fun() -> R = asctime_date(V) end} || {V, R} <- Tests].
content_type_test_() ->
%% {ContentType, Result}
Tests = [
{<<"text/plain; charset=iso-8859-4">>,
{<<"text">>, <<"plain">>, [{<<"charset">>, <<"iso-8859-4">>}]}},
{<<"multipart/form-data \t;Boundary=\"MultipartIsUgly\"">>,
{<<"multipart">>, <<"form-data">>, [
{<<"boundary">>, <<"MultipartIsUgly">>}
]}},
{<<"foo/bar; one=FirstParam; two=SecondParam">>,
{<<"foo">>, <<"bar">>, [
{<<"one">>, <<"FirstParam">>},
{<<"two">>, <<"SecondParam">>}
]}}
],
[{V, fun () -> R = content_type(V) end} || {V, R} <- Tests].
parameterized_tokens_test_() ->
%% {ParameterizedTokens, Result}
Tests = [
{<<"foo">>, [{<<"foo">>, []}]},
{<<"bar; baz=2">>, [{<<"bar">>, [{<<"baz">>, <<"2">>}]}]},
{<<"bar; baz=2;bat">>, [{<<"bar">>, [{<<"baz">>, <<"2">>}, <<"bat">>]}]},
{<<"bar; baz=2;bat=\"z=1,2;3\"">>, [{<<"bar">>, [{<<"baz">>, <<"2">>}, {<<"bat">>, <<"z=1,2;3">>}]}]},
{<<"foo, bar; baz=2">>, [{<<"foo">>, []}, {<<"bar">>, [{<<"baz">>, <<"2">>}]}]}
],
[{V, fun () -> R = parameterized_tokens(V) end} || {V, R} <- Tests].
digits_test_() ->
%% {Digits, Result}
Tests = [
{<<"42 ">>, 42},
{<<"69\t">>, 69},
{<<"1337">>, 1337}
],
[{V, fun() -> R = digits(V) end} || {V, R} <- Tests].
urldecode_test_() ->
F = fun(Qs, O) ->
try urldecode(Qs, O) of
R ->
{ok, R}
catch _:E ->
{error, E}
end
end,
Tests = [
{<<"%20">>, crash, {ok, <<" ">>}},
{<<"+">>, crash, {ok, <<" ">>}},
{<<"%00">>, crash, {ok, <<0>>}},
{<<"%fF">>, crash, {ok, <<255>>}},
{<<"123">>, crash, {ok, <<"123">>}},
{<<"%i5">>, skip, {ok, <<"%i5">>}},
{<<"%5">>, skip, {ok, <<"%5">>}},
{<<"%i5">>, crash, {error, badarg}},
{<<"%5">>, crash, {error, badarg}}
],
[{Qs, fun() -> R = F(Qs,O) end} || {Qs, O, R} <- Tests].
urlencode_test_() ->
Tests = [
{<<255,0>>, [], <<"%ff%00">>},
{<<255,0>>, [upper], <<"%FF%00">>},
{<<" ">>, [], <<"+">>},
{<<" ">>, [noplus], <<"%20">>},
{<<"aBc">>, [], <<"aBc">>},
{<<".-~_">>, [], <<".-~_">>}
],
Tests2 = [{<<255, " ">>,<<"%ff+">>}],
[{V, fun() -> R = urlencode(V, O) end} || {V, O, R} <- Tests] ++
[{V, fun() -> R = urlencode(V) end} || {V, R} <- Tests2].
http_authorization_test_() ->
Tests = [
{<<"basic">>, <<"QWxsYWRpbjpvcGVuIHNlc2FtZQ==">>,
{<<"basic">>, {<<"Alladin">>, <<"open sesame">>}}},
{<<"basic">>, <<"dXNlcm5hbWUK">>,
{error, badarg}},
{<<"basic">>, <<"_[]@#$%^&*()-AA==">>,
{error, badarg}},
{<<"basic">>, <<"dXNlcjpwYXNzCA==">>,
{error, badarg}},
{<<"bearer">>, <<" some_secret_key">>,
{<<"bearer">>,<<"some_secret_key">>}}
],
[{V, fun() -> R = authorization(V,T) end} || {T, V, R} <- Tests].
http_range_test_() ->
Tests = [
{<<"bytes=1-20">>,
{<<"bytes">>, [{1, 20}]}},
{<<"bytes=-100">>,
{<<"bytes">>, [-100]}},
{<<"bytes=1-">>,
{<<"bytes">>, [{1, infinity}]}},
{<<"bytes=1-20,30-40,50-">>,
{<<"bytes">>, [{1, 20}, {30, 40}, {50, infinity}]}},
{<<"bytes = 1 - 20 , 50 - , - 300 ">>,
{<<"bytes">>, [{1, 20}, {50, infinity}, -300]}},
{<<"bytes=1-20,-500,30-40">>,
{<<"bytes">>, [{1, 20}, -500, {30, 40}]}},
{<<"test=1-20,-500,30-40">>,
{<<"test">>, [{1, 20}, -500, {30, 40}]}},
{<<"bytes=-">>,
{error, badarg}},
{<<"bytes=-30,-">>,
{error, badarg}}
],
[fun() -> R = range(V) end ||{V, R} <- Tests].
-endif.