aboutsummaryrefslogtreecommitdiffstats
path: root/lib/asn1/src/asn1ct_tok.erl
diff options
context:
space:
mode:
authorErlang/OTP <[email protected]>2009-11-20 14:54:40 +0000
committerErlang/OTP <[email protected]>2009-11-20 14:54:40 +0000
commit84adefa331c4159d432d22840663c38f155cd4c1 (patch)
treebff9a9c66adda4df2106dfd0e5c053ab182a12bd /lib/asn1/src/asn1ct_tok.erl
downloadotp-84adefa331c4159d432d22840663c38f155cd4c1.tar.gz
otp-84adefa331c4159d432d22840663c38f155cd4c1.tar.bz2
otp-84adefa331c4159d432d22840663c38f155cd4c1.zip
The R13B03 release.OTP_R13B03
Diffstat (limited to 'lib/asn1/src/asn1ct_tok.erl')
-rw-r--r--lib/asn1/src/asn1ct_tok.erl391
1 files changed, 391 insertions, 0 deletions
diff --git a/lib/asn1/src/asn1ct_tok.erl b/lib/asn1/src/asn1ct_tok.erl
new file mode 100644
index 0000000000..27116c46c5
--- /dev/null
+++ b/lib/asn1/src/asn1ct_tok.erl
@@ -0,0 +1,391 @@
+%%
+%% %CopyrightBegin%
+%%
+%% Copyright Ericsson AB 1997-2009. 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(asn1ct_tok).
+
+%% Tokenize ASN.1 code (input to parser generated with yecc)
+
+-export([get_name/2,tokenise/2, file/1]).
+
+
+file(File) ->
+ case file:open(File, [read]) of
+ {error, Reason} ->
+ {error,{File,file:format_error(Reason)}};
+ {ok,Stream} ->
+ process0(Stream)
+ end.
+
+process0(Stream) ->
+ process(Stream,0,[]).
+
+process(Stream,Lno,R) ->
+ process(io:get_line(Stream, ''), Stream,Lno+1,R).
+
+process(eof, Stream,Lno,R) ->
+ file:close(Stream),
+ lists:flatten(lists:reverse([{'$end',Lno}|R]));
+
+
+process(L, Stream,Lno,R) when is_list(L) ->
+ %%io:format('read:~s',[L]),
+ case catch tokenise(L,Lno) of
+ {'ERR',Reason} ->
+ io:format("Tokeniser error on line: ~w ~w~n",[Lno,Reason]),
+ exit(0);
+ {multiline_comment,NestingLevel} ->
+ {RestL,Lno2} = process_skip_multiline_comment(Stream,Lno,NestingLevel),
+ process(RestL,Stream,Lno2,R);
+ T ->
+ %%io:format('toks:~w~n',[T]),
+ process(Stream,Lno,[T|R])
+ end.
+
+process_skip_multiline_comment(Stream,Lno,NestingLevel) ->
+ process_skip_multiline_comment(io:get_line(Stream, ''),
+ Stream, Lno + 1, NestingLevel).
+process_skip_multiline_comment(eof,_Stream,Lno,_NestingLevel) ->
+ io:format("Tokeniser error on line: ~w, premature end of multiline comment~n",[Lno]),
+ exit(0);
+process_skip_multiline_comment(Line,Stream,Lno,NestingLevel) ->
+ case catch skip_multiline_comment(Line,NestingLevel) of
+ {multiline_comment,NestingLevel2} ->
+ process_skip_multiline_comment(Stream,Lno,NestingLevel2);
+ T ->
+ {T,Lno}
+ end.
+
+tokenise([H|T],Lno) when $a =< H , H =< $z ->
+ {X, T1} = get_name(T, [H]),
+ [{identifier,Lno, list_to_atom(X)}|tokenise(T1,Lno)];
+
+tokenise([$&,H|T],Lno) when $A =< H , H =< $Z ->
+ {Y, T1} = get_name(T, [H]),
+ X = list_to_atom(Y),
+ [{typefieldreference, Lno, X} | tokenise(T1, Lno)];
+
+tokenise([$&,H|T],Lno) when $a =< H , H =< $z ->
+ {Y, T1} = get_name(T, [H]),
+ X = list_to_atom(Y),
+ [{valuefieldreference, Lno, X} | tokenise(T1, Lno)];
+
+tokenise([H|T],Lno) when $A =< H , H =< $Z ->
+ {Y, T1} = get_name(T, [H]),
+ X = list_to_atom(Y),
+ case reserved_word(X) of
+ true ->
+ [{X,Lno}|tokenise(T1,Lno)];
+ false ->
+ [{typereference,Lno,X}|tokenise(T1,Lno)];
+ rstrtype ->
+ [{restrictedcharacterstringtype,Lno,X}|tokenise(T1,Lno)]
+ end;
+
+tokenise([$-,H|T],Lno) when $0 =< H , H =< $9 ->
+ {X, T1} = get_number(T, [H]),
+ [{number,Lno,-1 * list_to_integer(X)}|tokenise(T1,Lno)];
+
+tokenise([H|T],Lno) when $0 =< H , H =< $9 ->
+ {X, T1} = get_number(T, [H]),
+ [{number,Lno,list_to_integer(X)}|tokenise(T1,Lno)];
+
+tokenise([$-,$-|T],Lno) ->
+ tokenise(skip_comment(T),Lno);
+
+tokenise([$/,$*|T],Lno) ->
+ tokenise(skip_multiline_comment(T,0),Lno);
+
+tokenise([$:,$:,$=|T],Lno) ->
+ [{'::=',Lno}|tokenise(T,Lno)];
+
+tokenise([$'|T],Lno) ->
+ case catch collect_quoted(T,Lno,[]) of
+ {'ERR',_} ->
+ throw({'ERR','bad_quote'});
+ {Thing, T1} ->
+ [Thing|tokenise(T1,Lno)]
+ end;
+
+tokenise([$"|T],Lno) ->
+ collect_string(T,Lno);
+
+tokenise([${|T],Lno) ->
+ [{'{',Lno}|tokenise(T,Lno)];
+
+tokenise([$}|T],Lno) ->
+ [{'}',Lno}|tokenise(T,Lno)];
+
+tokenise([$]|T],Lno) ->
+ [{']',Lno}|tokenise(T,Lno)];
+
+tokenise([$[|T],Lno) ->
+ [{'[',Lno}|tokenise(T,Lno)];
+
+tokenise([$,|T],Lno) ->
+ [{',',Lno}|tokenise(T,Lno)];
+
+tokenise([$(|T],Lno) ->
+ [{'(',Lno}|tokenise(T,Lno)];
+tokenise([$)|T],Lno) ->
+ [{')',Lno}|tokenise(T,Lno)];
+
+tokenise([$.,$.,$.|T],Lno) ->
+ [{'...',Lno}|tokenise(T,Lno)];
+
+tokenise([$.,$.|T],Lno) ->
+ [{'..',Lno}|tokenise(T,Lno)];
+
+tokenise([$.|T],Lno) ->
+ [{'.',Lno}|tokenise(T,Lno)];
+tokenise([$^|T],Lno) ->
+ [{'^',Lno}|tokenise(T,Lno)];
+tokenise([$!|T],Lno) ->
+ [{'!',Lno}|tokenise(T,Lno)];
+tokenise([$||T],Lno) ->
+ [{'|',Lno}|tokenise(T,Lno)];
+
+
+tokenise([H|T],Lno) ->
+ case white_space(H) of
+ true ->
+ tokenise(T,Lno);
+ false ->
+ [{list_to_atom([H]),Lno}|tokenise(T,Lno)]
+ end;
+tokenise([],_) ->
+ [].
+
+
+collect_string(L,Lno) ->
+ collect_string(L,Lno,[]).
+
+collect_string([],_,_) ->
+ throw({'ERR','bad_quote found eof'});
+
+collect_string([H|T],Lno,Str) ->
+ case H of
+ $" ->
+ [{cstring,1,lists:reverse(Str)}|tokenise(T,Lno)];
+ Ch ->
+ collect_string(T,Lno,[Ch|Str])
+ end.
+
+
+
+% <name> is letters digits hyphens
+% hypen is not the last character. Hypen hyphen is NOT allowed
+%
+% <identifier> ::= <lowercase> <name>
+
+get_name([$-,Char|T], L) ->
+ case isalnum(Char) of
+ true ->
+ get_name(T,[Char,$-|L]);
+ false ->
+ {lists:reverse(L),[$-,Char|T]}
+ end;
+get_name([$-|T], L) ->
+ {lists:reverse(L),[$-|T]};
+get_name([Char|T], L) ->
+ case isalnum(Char) of
+ true ->
+ get_name(T,[Char|L]);
+ false ->
+ {lists:reverse(L),[Char|T]}
+ end;
+get_name([], L) ->
+ {lists:reverse(L), []}.
+
+
+isalnum(H) when $A =< H , H =< $Z ->
+ true;
+isalnum(H) when $a =< H , H =< $z ->
+ true;
+isalnum(H) when $0 =< H , H =< $9 ->
+ true;
+isalnum(_) ->
+ false.
+
+isdigit(H) when $0 =< H , H =< $9 ->
+ true;
+isdigit(_) ->
+ false.
+
+white_space(9) -> true;
+white_space(10) -> true;
+white_space(13) -> true;
+white_space(32) -> true;
+white_space(_) -> false.
+
+
+get_number([H|T], L) ->
+ case isdigit(H) of
+ true ->
+ get_number(T, [H|L]);
+ false ->
+ {lists:reverse(L), [H|T]}
+ end;
+get_number([], L) ->
+ {lists:reverse(L), []}.
+
+skip_comment([]) ->
+ [];
+skip_comment([$-,$-|T]) ->
+ T;
+skip_comment([_|T]) ->
+ skip_comment(T).
+
+
+skip_multiline_comment([],L) ->
+ throw({multiline_comment,L});
+skip_multiline_comment([$*,$/|T],0) ->
+ T;
+skip_multiline_comment([$*,$/|T],Level) ->
+ skip_multiline_comment(T,Level - 1);
+skip_multiline_comment([$/,$*|T],Level) ->
+ skip_multiline_comment(T,Level + 1);
+skip_multiline_comment([_|T],Level) ->
+ skip_multiline_comment(T,Level).
+
+
+collect_quoted([$',$B|T],Lno, L) ->
+ case check_bin(L) of
+ true ->
+ {{bstring,Lno, lists:reverse(L)}, T};
+ false ->
+ throw({'ERR',{invalid_binary_number, lists:reverse(L)}})
+ end;
+collect_quoted([$',$H|T],Lno, L) ->
+ case check_hex(L) of
+ true ->
+ {{hstring,Lno, lists:reverse(L)}, T};
+ false ->
+ throw({'ERR',{invalid_binary_number, lists:reverse(L)}})
+ end;
+collect_quoted([H|T], Lno, L) ->
+ collect_quoted(T, Lno,[H|L]);
+collect_quoted([], _, _) -> % This should be allowed FIX later
+ throw({'ERR',{eol_in_token}}).
+
+check_bin([$0|T]) ->
+ check_bin(T);
+check_bin([$1|T]) ->
+ check_bin(T);
+check_bin([]) ->
+ true;
+check_bin(_) ->
+ false.
+
+check_hex([H|T]) when $0 =< H , H =< $9 ->
+ check_hex(T);
+check_hex([H|T]) when $A =< H , H =< $F ->
+ check_hex(T);
+check_hex([]) ->
+ true;
+check_hex(_) ->
+ false.
+
+
+%% reserved_word(A) -> true|false|rstrtype
+%% A = atom()
+%% returns true if A is a reserved ASN.1 word
+%% returns false if A is not a reserved word
+%% returns rstrtype if A is a reserved word in the group
+%% RestrictedCharacterStringType
+reserved_word('ABSENT') -> true;
+%reserved_word('ABSTRACT-SYNTAX') -> true; % impl as predef item
+reserved_word('ALL') -> true;
+reserved_word('ANY') -> true;
+reserved_word('APPLICATION') -> true;
+reserved_word('AUTOMATIC') -> true;
+reserved_word('BEGIN') -> true;
+reserved_word('BIT') -> true;
+reserved_word('BMPString') -> rstrtype;
+reserved_word('BOOLEAN') -> true;
+reserved_word('BY') -> true;
+reserved_word('CHARACTER') -> true;
+reserved_word('CHOICE') -> true;
+reserved_word('CLASS') -> true;
+reserved_word('COMPONENT') -> true;
+reserved_word('COMPONENTS') -> true;
+reserved_word('CONSTRAINED') -> true;
+reserved_word('CONTAINING') -> true;
+reserved_word('DEFAULT') -> true;
+reserved_word('DEFINED') -> true;
+reserved_word('DEFINITIONS') -> true;
+reserved_word('EMBEDDED') -> true;
+reserved_word('ENCODED') -> true;
+reserved_word('END') -> true;
+reserved_word('ENUMERATED') -> true;
+reserved_word('EXCEPT') -> true;
+reserved_word('EXPLICIT') -> true;
+reserved_word('EXPORTS') -> true;
+reserved_word('EXTERNAL') -> true;
+reserved_word('FALSE') -> true;
+reserved_word('FROM') -> true;
+reserved_word('GeneralizedTime') -> true;
+reserved_word('GeneralString') -> rstrtype;
+reserved_word('GraphicString') -> rstrtype;
+reserved_word('IA5String') -> rstrtype;
+% reserved_word('TYPE-IDENTIFIER') -> true; % impl as predef item
+reserved_word('IDENTIFIER') -> true;
+reserved_word('IMPLICIT') -> true;
+reserved_word('IMPORTS') -> true;
+reserved_word('INCLUDES') -> true;
+reserved_word('INSTANCE') -> true;
+reserved_word('INTEGER') -> true;
+reserved_word('INTERSECTION') -> true;
+reserved_word('ISO646String') -> rstrtype;
+reserved_word('MAX') -> true;
+reserved_word('MIN') -> true;
+reserved_word('MINUS-INFINITY') -> true;
+reserved_word('NULL') -> true;
+reserved_word('NumericString') -> rstrtype;
+reserved_word('OBJECT') -> true;
+reserved_word('ObjectDescriptor') -> true;
+reserved_word('OCTET') -> true;
+reserved_word('OF') -> true;
+reserved_word('OPTIONAL') -> true;
+reserved_word('PATTERN') -> true;
+reserved_word('PDV') -> true;
+reserved_word('PLUS-INFINITY') -> true;
+reserved_word('PRESENT') -> true;
+reserved_word('PrintableString') -> rstrtype;
+reserved_word('PRIVATE') -> true;
+reserved_word('REAL') -> true;
+reserved_word('RELATIVE-OID') -> true;
+reserved_word('SEQUENCE') -> true;
+reserved_word('SET') -> true;
+reserved_word('SIZE') -> true;
+reserved_word('STRING') -> true;
+reserved_word('SYNTAX') -> true;
+reserved_word('T61String') -> rstrtype;
+reserved_word('TAGS') -> true;
+reserved_word('TeletexString') -> rstrtype;
+reserved_word('TRUE') -> true;
+reserved_word('UNION') -> true;
+reserved_word('UNIQUE') -> true;
+reserved_word('UNIVERSAL') -> true;
+reserved_word('UniversalString') -> rstrtype;
+reserved_word('UTCTime') -> true;
+reserved_word('UTF8String') -> rstrtype;
+reserved_word('VideotexString') -> rstrtype;
+reserved_word('VisibleString') -> rstrtype;
+reserved_word('WITH') -> true;
+reserved_word(_) -> false.