| #! /usr/bin/env perl |
| # Copyright 2018-2021 The OpenSSL Project Authors. All Rights Reserved. |
| # |
| # Licensed under the Apache License 2.0 (the "License"). You may not use |
| # this file except in compliance with the License. You can obtain a copy |
| # in the file LICENSE in the source distribution or at |
| # https://www.openssl.org/source/license.html |
| |
| package OpenSSL::ParseC; |
| |
| use strict; |
| use warnings; |
| |
| use Exporter; |
| use vars qw($VERSION @ISA @EXPORT @EXPORT_OK %EXPORT_TAGS); |
| $VERSION = "0.9"; |
| @ISA = qw(Exporter); |
| @EXPORT = qw(parse); |
| |
| # Global handler data |
| my @preprocessor_conds; # A list of simple preprocessor conditions, |
| # each item being a list of macros defined |
| # or not defined. |
| |
| # Handler helpers |
| sub all_conds { |
| return map { ( @$_ ) } @preprocessor_conds; |
| } |
| |
| # A list of handlers that will look at a "complete" string and try to |
| # figure out what to make of it. |
| # Each handler is a hash with the following keys: |
| # |
| # regexp a regexp to compare the "complete" string with. |
| # checker a function that does a more complex comparison. |
| # Use this instead of regexp if that isn't enough. |
| # massager massages the "complete" string into an array with |
| # the following elements: |
| # |
| # [0] String that needs further processing (this |
| # applies to typedefs of structs), or empty. |
| # [1] The name of what was found. |
| # [2] A character that denotes what type of thing |
| # this is: 'F' for function, 'S' for struct, |
| # 'T' for typedef, 'M' for macro, 'V' for |
| # variable. |
| # [3] Return type (only for type 'F' and 'V') |
| # [4] Value (for type 'M') or signature (for type 'F', |
| # 'V', 'T' or 'S') |
| # [5...] The list of preprocessor conditions this is |
| # found in, as in checks for macro definitions |
| # (stored as the macro's name) or the absence |
| # of definition (stored as the macro's name |
| # prefixed with a '!' |
| # |
| # If the massager returns an empty list, it means the |
| # "complete" string has side effects but should otherwise |
| # be ignored. |
| # If the massager is undefined, the "complete" string |
| # should be ignored. |
| my @opensslcpphandlers = ( |
| ################################################################## |
| # OpenSSL CPP specials |
| # |
| # These are used to convert certain pre-precessor expressions into |
| # others that @cpphandlers have a better chance to understand. |
| |
| # This changes any OPENSSL_NO_DEPRECATED_x_y[_z] check to a check of |
| # OPENSSL_NO_DEPRECATEDIN_x_y[_z]. That's due to <openssl/macros.h> |
| # creating OPENSSL_NO_DEPRECATED_x_y[_z], but the ordinals files using |
| # DEPRECATEDIN_x_y[_z]. |
| { regexp => qr/#if(def|ndef) OPENSSL_NO_DEPRECATED_(\d+_\d+(?:_\d+)?)$/, |
| massager => sub { |
| return (<<"EOF"); |
| #if$1 OPENSSL_NO_DEPRECATEDIN_$2 |
| EOF |
| } |
| } |
| ); |
| my @cpphandlers = ( |
| ################################################################## |
| # CPP stuff |
| |
| { regexp => qr/#ifdef ?(.*)/, |
| massager => sub { |
| my %opts; |
| if (ref($_[$#_]) eq "HASH") { |
| %opts = %{$_[$#_]}; |
| pop @_; |
| } |
| push @preprocessor_conds, [ $1 ]; |
| print STDERR "DEBUG[",$opts{debug_type},"]: preprocessor level: ", scalar(@preprocessor_conds), "\n" |
| if $opts{debug}; |
| return (); |
| }, |
| }, |
| { regexp => qr/#ifndef ?(.*)/, |
| massager => sub { |
| my %opts; |
| if (ref($_[$#_]) eq "HASH") { |
| %opts = %{$_[$#_]}; |
| pop @_; |
| } |
| push @preprocessor_conds, [ '!'.$1 ]; |
| print STDERR "DEBUG[",$opts{debug_type},"]: preprocessor level: ", scalar(@preprocessor_conds), "\n" |
| if $opts{debug}; |
| return (); |
| }, |
| }, |
| { regexp => qr/#if (0|1)/, |
| massager => sub { |
| my %opts; |
| if (ref($_[$#_]) eq "HASH") { |
| %opts = %{$_[$#_]}; |
| pop @_; |
| } |
| if ($1 eq "1") { |
| push @preprocessor_conds, [ "TRUE" ]; |
| } else { |
| push @preprocessor_conds, [ "!TRUE" ]; |
| } |
| print STDERR "DEBUG[",$opts{debug_type},"]: preprocessor level: ", scalar(@preprocessor_conds), "\n" |
| if $opts{debug}; |
| return (); |
| }, |
| }, |
| { regexp => qr/#if ?(.*)/, |
| massager => sub { |
| my %opts; |
| if (ref($_[$#_]) eq "HASH") { |
| %opts = %{$_[$#_]}; |
| pop @_; |
| } |
| my @results = (); |
| my $conds = $1; |
| if ($conds =~ m|^defined<<<\(([^\)]*)\)>>>(.*)$|) { |
| push @results, $1; # Handle the simple case |
| my $rest = $2; |
| my $re = qr/^(?:\|\|defined<<<\([^\)]*\)>>>)*$/; |
| print STDERR "DEBUG[",$opts{debug_type},"]: Matching '$rest' with '$re'\n" |
| if $opts{debug}; |
| if ($rest =~ m/$re/) { |
| my @rest = split /\|\|/, $rest; |
| shift @rest; |
| foreach (@rest) { |
| m|^defined<<<\(([^\)]*)\)>>>$|; |
| die "Something wrong...$opts{PLACE}" if $1 eq ""; |
| push @results, $1; |
| } |
| } else { |
| $conds =~ s/<<<|>>>//g; |
| warn "Warning: complicated #if expression(1): $conds$opts{PLACE}" |
| if $opts{warnings}; |
| } |
| } elsif ($conds =~ m|^!defined<<<\(([^\)]*)\)>>>(.*)$|) { |
| push @results, '!'.$1; # Handle the simple case |
| my $rest = $2; |
| my $re = qr/^(?:\&\&!defined<<<\([^\)]*\)>>>)*$/; |
| print STDERR "DEBUG[",$opts{debug_type},"]: Matching '$rest' with '$re'\n" |
| if $opts{debug}; |
| if ($rest =~ m/$re/) { |
| my @rest = split /\&\&/, $rest; |
| shift @rest; |
| foreach (@rest) { |
| m|^!defined<<<\(([^\)]*)\)>>>$|; |
| die "Something wrong...$opts{PLACE}" if $1 eq ""; |
| push @results, '!'.$1; |
| } |
| } else { |
| $conds =~ s/<<<|>>>//g; |
| warn "Warning: complicated #if expression(2): $conds$opts{PLACE}" |
| if $opts{warnings}; |
| } |
| } else { |
| $conds =~ s/<<<|>>>//g; |
| warn "Warning: complicated #if expression(3): $conds$opts{PLACE}" |
| if $opts{warnings}; |
| } |
| print STDERR "DEBUG[",$opts{debug_type},"]: Added preprocessor conds: '", join("', '", @results), "'\n" |
| if $opts{debug}; |
| push @preprocessor_conds, [ @results ]; |
| print STDERR "DEBUG[",$opts{debug_type},"]: preprocessor level: ", scalar(@preprocessor_conds), "\n" |
| if $opts{debug}; |
| return (); |
| }, |
| }, |
| { regexp => qr/#elif (.*)/, |
| massager => sub { |
| my %opts; |
| if (ref($_[$#_]) eq "HASH") { |
| %opts = %{$_[$#_]}; |
| pop @_; |
| } |
| die "An #elif without corresponding condition$opts{PLACE}" |
| if !@preprocessor_conds; |
| pop @preprocessor_conds; |
| print STDERR "DEBUG[",$opts{debug_type},"]: preprocessor level: ", scalar(@preprocessor_conds), "\n" |
| if $opts{debug}; |
| return (<<"EOF"); |
| #if $1 |
| EOF |
| }, |
| }, |
| { regexp => qr/#else/, |
| massager => sub { |
| my %opts; |
| if (ref($_[$#_]) eq "HASH") { |
| %opts = %{$_[$#_]}; |
| pop @_; |
| } |
| die "An #else without corresponding condition$opts{PLACE}" |
| if !@preprocessor_conds; |
| # Invert all conditions on the last level |
| my $stuff = pop @preprocessor_conds; |
| push @preprocessor_conds, [ |
| map { m|^!(.*)$| ? $1 : '!'.$_ } @$stuff |
| ]; |
| print STDERR "DEBUG[",$opts{debug_type},"]: preprocessor level: ", scalar(@preprocessor_conds), "\n" |
| if $opts{debug}; |
| return (); |
| }, |
| }, |
| { regexp => qr/#endif ?/, |
| massager => sub { |
| my %opts; |
| if (ref($_[$#_]) eq "HASH") { |
| %opts = %{$_[$#_]}; |
| pop @_; |
| } |
| die "An #endif without corresponding condition$opts{PLACE}" |
| if !@preprocessor_conds; |
| pop @preprocessor_conds; |
| print STDERR "DEBUG[",$opts{debug_type},"]: preprocessor level: ", scalar(@preprocessor_conds), "\n" |
| if $opts{debug}; |
| return (); |
| }, |
| }, |
| { regexp => qr/#define ([[:alpha:]_]\w*)(<<<\(.*?\)>>>)?( (.*))?/, |
| massager => sub { |
| my $name = $1; |
| my $params = $2; |
| my $spaceval = $3||""; |
| my $val = $4||""; |
| return ("", |
| $1, 'M', "", $params ? "$name$params$spaceval" : $val, |
| all_conds()); } |
| }, |
| { regexp => qr/#.*/, |
| massager => sub { return (); } |
| }, |
| ); |
| |
| my @opensslchandlers = ( |
| ################################################################## |
| # OpenSSL C specials |
| # |
| # They are really preprocessor stuff, but they look like C stuff |
| # to this parser. All of these do replacements, anything else is |
| # an error. |
| |
| ##### |
| # Deprecated stuff, by OpenSSL release. |
| |
| # OSSL_DEPRECATEDIN_x_y[_z] is simply ignored. Such declarations are |
| # supposed to be guarded with an '#ifdef OPENSSL_NO_DEPRECATED_x_y[_z]' |
| { regexp => qr/OSSL_DEPRECATEDIN_\d+_\d+(?:_\d+)?\s+(.*)/, |
| massager => sub { return $1; }, |
| }, |
| { regexp => qr/(.*?)\s+OSSL_DEPRECATEDIN_\d+_\d+(?:_\d+)?\s+(.*)/, |
| massager => sub { return "$1 $2"; }, |
| }, |
| |
| ##### |
| # Core stuff |
| |
| # OSSL_CORE_MAKE_FUNC is a macro to create the necessary data and inline |
| # function the libcrypto<->provider interface |
| { regexp => qr/OSSL_CORE_MAKE_FUNC<<<\((.*?),(.*?),(.*?)\)>>>/, |
| massager => sub { |
| return (<<"EOF"); |
| typedef $1 OSSL_FUNC_$2_fn$3; |
| static ossl_inline OSSL_FUNC_$2_fn *OSSL_FUNC_$2(const OSSL_DISPATCH *opf); |
| EOF |
| }, |
| }, |
| |
| ##### |
| # LHASH stuff |
| |
| # LHASH_OF(foo) is used as a type, but the chandlers won't take it |
| # gracefully, so we expand it here. |
| { regexp => qr/(.*)\bLHASH_OF<<<\((.*?)\)>>>(.*)/, |
| massager => sub { return ("$1struct lhash_st_$2$3"); } |
| }, |
| { regexp => qr/DEFINE_LHASH_OF(?:_INTERNAL)?<<<\((.*)\)>>>/, |
| massager => sub { |
| return (<<"EOF"); |
| static ossl_inline LHASH_OF($1) * lh_$1_new(unsigned long (*hfn)(const $1 *), |
| int (*cfn)(const $1 *, const $1 *)); |
| static ossl_inline void lh_$1_free(LHASH_OF($1) *lh); |
| static ossl_inline $1 *lh_$1_insert(LHASH_OF($1) *lh, $1 *d); |
| static ossl_inline $1 *lh_$1_delete(LHASH_OF($1) *lh, const $1 *d); |
| static ossl_inline $1 *lh_$1_retrieve(LHASH_OF($1) *lh, const $1 *d); |
| static ossl_inline int lh_$1_error(LHASH_OF($1) *lh); |
| static ossl_inline unsigned long lh_$1_num_items(LHASH_OF($1) *lh); |
| static ossl_inline void lh_$1_node_stats_bio(const LHASH_OF($1) *lh, BIO *out); |
| static ossl_inline void lh_$1_node_usage_stats_bio(const LHASH_OF($1) *lh, |
| BIO *out); |
| static ossl_inline void lh_$1_stats_bio(const LHASH_OF($1) *lh, BIO *out); |
| static ossl_inline unsigned long lh_$1_get_down_load(LHASH_OF($1) *lh); |
| static ossl_inline void lh_$1_set_down_load(LHASH_OF($1) *lh, unsigned long dl); |
| static ossl_inline void lh_$1_doall(LHASH_OF($1) *lh, void (*doall)($1 *)); |
| LHASH_OF($1) |
| EOF |
| } |
| }, |
| |
| ##### |
| # STACK stuff |
| |
| # STACK_OF(foo) is used as a type, but the chandlers won't take it |
| # gracefully, so we expand it here. |
| { regexp => qr/(.*)\bSTACK_OF<<<\((.*?)\)>>>(.*)/, |
| massager => sub { return ("$1struct stack_st_$2$3"); } |
| }, |
| # { regexp => qr/(.*)\bSTACK_OF\((.*?)\)(.*)/, |
| # massager => sub { |
| # my $before = $1; |
| # my $stack_of = "struct stack_st_$2"; |
| # my $after = $3; |
| # if ($after =~ m|^\w|) { $after = " ".$after; } |
| # return ("$before$stack_of$after"); |
| # } |
| # }, |
| { regexp => qr/SKM_DEFINE_STACK_OF<<<\((.*),\s*(.*),\s*(.*)\)>>>/, |
| massager => sub { |
| return (<<"EOF"); |
| STACK_OF($1); |
| typedef int (*sk_$1_compfunc)(const $3 * const *a, const $3 *const *b); |
| typedef void (*sk_$1_freefunc)($3 *a); |
| typedef $3 * (*sk_$1_copyfunc)(const $3 *a); |
| static ossl_inline int sk_$1_num(const STACK_OF($1) *sk); |
| static ossl_inline $2 *sk_$1_value(const STACK_OF($1) *sk, int idx); |
| static ossl_inline STACK_OF($1) *sk_$1_new(sk_$1_compfunc compare); |
| static ossl_inline STACK_OF($1) *sk_$1_new_null(void); |
| static ossl_inline STACK_OF($1) *sk_$1_new_reserve(sk_$1_compfunc compare, |
| int n); |
| static ossl_inline int sk_$1_reserve(STACK_OF($1) *sk, int n); |
| static ossl_inline void sk_$1_free(STACK_OF($1) *sk); |
| static ossl_inline void sk_$1_zero(STACK_OF($1) *sk); |
| static ossl_inline $2 *sk_$1_delete(STACK_OF($1) *sk, int i); |
| static ossl_inline $2 *sk_$1_delete_ptr(STACK_OF($1) *sk, $2 *ptr); |
| static ossl_inline int sk_$1_push(STACK_OF($1) *sk, $2 *ptr); |
| static ossl_inline int sk_$1_unshift(STACK_OF($1) *sk, $2 *ptr); |
| static ossl_inline $2 *sk_$1_pop(STACK_OF($1) *sk); |
| static ossl_inline $2 *sk_$1_shift(STACK_OF($1) *sk); |
| static ossl_inline void sk_$1_pop_free(STACK_OF($1) *sk, |
| sk_$1_freefunc freefunc); |
| static ossl_inline int sk_$1_insert(STACK_OF($1) *sk, $2 *ptr, int idx); |
| static ossl_inline $2 *sk_$1_set(STACK_OF($1) *sk, int idx, $2 *ptr); |
| static ossl_inline int sk_$1_find(STACK_OF($1) *sk, $2 *ptr); |
| static ossl_inline int sk_$1_find_ex(STACK_OF($1) *sk, $2 *ptr); |
| static ossl_inline void sk_$1_sort(STACK_OF($1) *sk); |
| static ossl_inline int sk_$1_is_sorted(const STACK_OF($1) *sk); |
| static ossl_inline STACK_OF($1) * sk_$1_dup(const STACK_OF($1) *sk); |
| static ossl_inline STACK_OF($1) *sk_$1_deep_copy(const STACK_OF($1) *sk, |
| sk_$1_copyfunc copyfunc, |
| sk_$1_freefunc freefunc); |
| static ossl_inline sk_$1_compfunc sk_$1_set_cmp_func(STACK_OF($1) *sk, |
| sk_$1_compfunc compare); |
| EOF |
| } |
| }, |
| { regexp => qr/SKM_DEFINE_STACK_OF_INTERNAL<<<\((.*),\s*(.*),\s*(.*)\)>>>/, |
| massager => sub { |
| return (<<"EOF"); |
| STACK_OF($1); |
| typedef int (*sk_$1_compfunc)(const $3 * const *a, const $3 *const *b); |
| typedef void (*sk_$1_freefunc)($3 *a); |
| typedef $3 * (*sk_$1_copyfunc)(const $3 *a); |
| static ossl_unused ossl_inline $2 *ossl_check_$1_type($2 *ptr); |
| static ossl_unused ossl_inline const OPENSSL_STACK *ossl_check_const_$1_sk_type(const STACK_OF($1) *sk); |
| static ossl_unused ossl_inline OPENSSL_sk_compfunc ossl_check_$1_compfunc_type(sk_$1_compfunc cmp); |
| static ossl_unused ossl_inline OPENSSL_sk_copyfunc ossl_check_$1_copyfunc_type(sk_$1_copyfunc cpy); |
| static ossl_unused ossl_inline OPENSSL_sk_freefunc ossl_check_$1_freefunc_type(sk_$1_freefunc fr); |
| EOF |
| } |
| }, |
| { regexp => qr/DEFINE_SPECIAL_STACK_OF<<<\((.*),\s*(.*)\)>>>/, |
| massager => sub { return ("SKM_DEFINE_STACK_OF($1,$2,$2)"); }, |
| }, |
| { regexp => qr/DEFINE_STACK_OF<<<\((.*)\)>>>/, |
| massager => sub { return ("SKM_DEFINE_STACK_OF($1,$1,$1)"); }, |
| }, |
| { regexp => qr/DEFINE_SPECIAL_STACK_OF_CONST<<<\((.*),\s*(.*)\)>>>/, |
| massager => sub { return ("SKM_DEFINE_STACK_OF($1,const $2,$2)"); }, |
| }, |
| { regexp => qr/DEFINE_STACK_OF_CONST<<<\((.*)\)>>>/, |
| massager => sub { return ("SKM_DEFINE_STACK_OF($1,const $1,$1)"); }, |
| }, |
| |
| ##### |
| # ASN1 stuff |
| { regexp => qr/DECLARE_ASN1_ITEM<<<\((.*)\)>>>/, |
| massager => sub { |
| return (<<"EOF"); |
| const ASN1_ITEM *$1_it(void); |
| EOF |
| }, |
| }, |
| { regexp => qr/DECLARE_ASN1_ENCODE_FUNCTIONS_only<<<\((.*),\s*(.*)\)>>>/, |
| massager => sub { |
| return (<<"EOF"); |
| int d2i_$2(void); |
| int i2d_$2(void); |
| EOF |
| }, |
| }, |
| { regexp => qr/DECLARE_ASN1_ENCODE_FUNCTIONS<<<\((.*),\s*(.*),\s*(.*)\)>>>/, |
| massager => sub { |
| return (<<"EOF"); |
| int d2i_$3(void); |
| int i2d_$3(void); |
| DECLARE_ASN1_ITEM($2) |
| EOF |
| }, |
| }, |
| { regexp => qr/DECLARE_ASN1_ENCODE_FUNCTIONS_name<<<\((.*),\s*(.*)\)>>>/, |
| massager => sub { |
| return (<<"EOF"); |
| int d2i_$2(void); |
| int i2d_$2(void); |
| DECLARE_ASN1_ITEM($2) |
| EOF |
| }, |
| }, |
| { regexp => qr/DECLARE_ASN1_ALLOC_FUNCTIONS_name<<<\((.*),\s*(.*)\)>>>/, |
| massager => sub { |
| return (<<"EOF"); |
| int $2_free(void); |
| int $2_new(void); |
| EOF |
| }, |
| }, |
| { regexp => qr/DECLARE_ASN1_ALLOC_FUNCTIONS<<<\((.*)\)>>>/, |
| massager => sub { |
| return (<<"EOF"); |
| int $1_free(void); |
| int $1_new(void); |
| EOF |
| }, |
| }, |
| { regexp => qr/DECLARE_ASN1_FUNCTIONS_name<<<\((.*),\s*(.*)\)>>>/, |
| massager => sub { |
| return (<<"EOF"); |
| int d2i_$2(void); |
| int i2d_$2(void); |
| int $2_free(void); |
| int $2_new(void); |
| DECLARE_ASN1_ITEM($2) |
| EOF |
| }, |
| }, |
| { regexp => qr/DECLARE_ASN1_FUNCTIONS<<<\((.*)\)>>>/, |
| massager => sub { return (<<"EOF"); |
| int d2i_$1(void); |
| int i2d_$1(void); |
| int $1_free(void); |
| int $1_new(void); |
| DECLARE_ASN1_ITEM($1) |
| EOF |
| } |
| }, |
| { regexp => qr/DECLARE_ASN1_NDEF_FUNCTION<<<\((.*)\)>>>/, |
| massager => sub { |
| return (<<"EOF"); |
| int i2d_$1_NDEF(void); |
| EOF |
| } |
| }, |
| { regexp => qr/DECLARE_ASN1_PRINT_FUNCTION<<<\((.*)\)>>>/, |
| massager => sub { |
| return (<<"EOF"); |
| int $1_print_ctx(void); |
| EOF |
| } |
| }, |
| { regexp => qr/DECLARE_ASN1_PRINT_FUNCTION_name<<<\((.*),\s*(.*)\)>>>/, |
| massager => sub { |
| return (<<"EOF"); |
| int $2_print_ctx(void); |
| EOF |
| } |
| }, |
| { regexp => qr/DECLARE_ASN1_SET_OF<<<\((.*)\)>>>/, |
| massager => sub { return (); } |
| }, |
| { regexp => qr/DECLARE_ASN1_DUP_FUNCTION<<<\((.*)\)>>>/, |
| massager => sub { |
| return (<<"EOF"); |
| int $1_dup(void); |
| EOF |
| } |
| }, |
| { regexp => qr/DECLARE_ASN1_DUP_FUNCTION_name<<<\((.*),\s*(.*)\)>>>/, |
| massager => sub { |
| return (<<"EOF"); |
| int $2_dup(void); |
| EOF |
| } |
| }, |
| # Universal translator of attributed PEM declarators |
| { regexp => qr/ |
| DECLARE_ASN1 |
| (_ENCODE_FUNCTIONS_only|_ENCODE_FUNCTIONS|_ENCODE_FUNCTIONS_name |
| |_ALLOC_FUNCTIONS_name|_ALLOC_FUNCTIONS|_FUNCTIONS_name|_FUNCTIONS |
| |_NDEF_FUNCTION|_PRINT_FUNCTION|_PRINT_FUNCTION_name |
| |_DUP_FUNCTION|_DUP_FUNCTION_name) |
| _attr |
| <<<\(\s*OSSL_DEPRECATEDIN_(.*?)\s*,(.*?)\)>>> |
| /x, |
| massager => sub { return (<<"EOF"); |
| DECLARE_ASN1$1($3) |
| EOF |
| }, |
| }, |
| { regexp => qr/DECLARE_PKCS12_SET_OF<<<\((.*)\)>>>/, |
| massager => sub { return (); } |
| }, |
| |
| ##### |
| # PEM stuff |
| { regexp => qr/DECLARE_PEM(?|_rw|_rw_cb|_rw_const)<<<\((.*?),.*\)>>>/, |
| massager => sub { return (<<"EOF"); |
| #ifndef OPENSSL_NO_STDIO |
| int PEM_read_$1(void); |
| int PEM_write_$1(void); |
| #endif |
| int PEM_read_bio_$1(void); |
| int PEM_write_bio_$1(void); |
| EOF |
| }, |
| }, |
| { regexp => qr/DECLARE_PEM(?|_rw|_rw_cb|_rw_const)_ex<<<\((.*?),.*\)>>>/, |
| massager => sub { return (<<"EOF"); |
| #ifndef OPENSSL_NO_STDIO |
| int PEM_read_$1(void); |
| int PEM_write_$1(void); |
| int PEM_read_$1_ex(void); |
| int PEM_write_$1_ex(void); |
| #endif |
| int PEM_read_bio_$1(void); |
| int PEM_write_bio_$1(void); |
| int PEM_read_bio_$1_ex(void); |
| int PEM_write_bio_$1_ex(void); |
| EOF |
| }, |
| }, |
| { regexp => qr/DECLARE_PEM(?|_write|_write_cb|_write_const)<<<\((.*?),.*\)>>>/, |
| massager => sub { return (<<"EOF"); |
| #ifndef OPENSSL_NO_STDIO |
| int PEM_write_$1(void); |
| #endif |
| int PEM_write_bio_$1(void); |
| EOF |
| }, |
| }, |
| { regexp => qr/DECLARE_PEM(?|_write|_write_cb|_write_const)_ex<<<\((.*?),.*\)>>>/, |
| massager => sub { return (<<"EOF"); |
| #ifndef OPENSSL_NO_STDIO |
| int PEM_write_$1(void); |
| int PEM_write_$1_ex(void); |
| #endif |
| int PEM_write_bio_$1(void); |
| int PEM_write_bio_$1_ex(void); |
| EOF |
| }, |
| }, |
| { regexp => qr/DECLARE_PEM(?|_read|_read_cb)<<<\((.*?),.*\)>>>/, |
| massager => sub { return (<<"EOF"); |
| #ifndef OPENSSL_NO_STDIO |
| int PEM_read_$1(void); |
| #endif |
| int PEM_read_bio_$1(void); |
| EOF |
| }, |
| }, |
| { regexp => qr/DECLARE_PEM(?|_read|_read_cb)_ex<<<\((.*?),.*\)>>>/, |
| massager => sub { return (<<"EOF"); |
| #ifndef OPENSSL_NO_STDIO |
| int PEM_read_$1(void); |
| int PEM_read_$1_ex(void); |
| #endif |
| int PEM_read_bio_$1(void); |
| int PEM_read_bio_$1_ex(void); |
| EOF |
| }, |
| }, |
| # Universal translator of attributed PEM declarators |
| { regexp => qr/ |
| DECLARE_PEM |
| ((?:_rw|_rw_cb|_rw_const|_write|_write_cb|_write_const|_read|_read_cb) |
| (?:_ex)?) |
| _attr |
| <<<\(\s*OSSL_DEPRECATEDIN_(.*?)\s*,(.*?)\)>>> |
| /x, |
| massager => sub { return (<<"EOF"); |
| DECLARE_PEM$1($3) |
| EOF |
| }, |
| }, |
| |
| # OpenSSL's declaration of externs with possible export linkage |
| # (really only relevant on Windows) |
| { regexp => qr/OPENSSL_(?:EXPORT|EXTERN)/, |
| massager => sub { return ("extern"); } |
| }, |
| |
| # Spurious stuff found in the OpenSSL headers |
| # Usually, these are just macros that expand to, well, something |
| { regexp => qr/__NDK_FPABI__/, |
| massager => sub { return (); } |
| }, |
| ); |
| |
| my $anoncnt = 0; |
| |
| my @chandlers = ( |
| ################################################################## |
| # C stuff |
| |
| # extern "C" of individual items |
| # Note that the main parse function has a special hack for 'extern "C" {' |
| # which can't be done in handlers |
| # We simply ignore it. |
| { regexp => qr/^extern "C" (.*(?:;|>>>))/, |
| massager => sub { return ($1); }, |
| }, |
| # any other extern is just ignored |
| { regexp => qr/^\s* # Any spaces before |
| extern # The keyword we look for |
| \b # word to non-word boundary |
| .* # Anything after |
| ; |
| /x, |
| massager => sub { return (); }, |
| }, |
| # union, struct and enum definitions |
| # Because this one might appear a little everywhere within type |
| # definitions, we take it out and replace it with just |
| # 'union|struct|enum name' while registering it. |
| # This makes use of the parser trick to surround the outer braces |
| # with <<< and >>> |
| { regexp => qr/(.*) # Anything before ($1) |
| \b # word to non-word boundary |
| (union|struct|enum) # The word used ($2) |
| (?:\s([[:alpha:]_]\w*))? # Struct or enum name ($3) |
| <<<(\{.*?\})>>> # Struct or enum definition ($4) |
| (.*) # Anything after ($5) |
| ; |
| /x, |
| massager => sub { |
| my $before = $1; |
| my $word = $2; |
| my $name = $3 |
| || sprintf("__anon%03d", ++$anoncnt); # Anonymous struct |
| my $definition = $4; |
| my $after = $5; |
| my $type = $word eq "struct" ? 'S' : 'E'; |
| if ($before ne "" || $after ne ";") { |
| if ($after =~ m|^\w|) { $after = " ".$after; } |
| return ("$before$word $name$after;", |
| "$word $name", $type, "", "$word$definition", all_conds()); |
| } |
| # If there was no before nor after, make the return much simple |
| return ("", "$word $name", $type, "", "$word$definition", all_conds()); |
| } |
| }, |
| # Named struct and enum forward declarations |
| # We really just ignore them, but we need to parse them or the variable |
| # declaration handler further down will think it's a variable declaration. |
| { regexp => qr/^(union|struct|enum) ([[:alpha:]_]\w*);/, |
| massager => sub { return (); } |
| }, |
| # Function returning function pointer declaration |
| # This sort of declaration may have a body (inline functions, for example) |
| { regexp => qr/(?:(typedef)\s?)? # Possible typedef ($1) |
| ((?:\w|\*|\s)*?) # Return type ($2) |
| \s? # Possible space |
| <<<\(\* |
| ([[:alpha:]_]\w*) # Function name ($3) |
| (\(.*\)) # Parameters ($4) |
| \)>>> |
| <<<(\(.*\))>>> # F.p. parameters ($5) |
| (?:<<<\{.*\}>>>|;) # Body or semicolon |
| /x, |
| massager => sub { |
| return ("", $3, 'T', "", "$2(*$4)$5", all_conds()) |
| if defined $1; |
| return ("", $3, 'F', "$2(*)$5", "$2(*$4)$5", all_conds()); } |
| }, |
| # Function pointer declaration, or typedef thereof |
| # This sort of declaration never has a function body |
| { regexp => qr/(?:(typedef)\s?)? # Possible typedef ($1) |
| ((?:\w|\*|\s)*?) # Return type ($2) |
| <<<\(\*([[:alpha:]_]\w*)\)>>> # T.d. or var name ($3) |
| <<<(\(.*\))>>> # F.p. parameters ($4) |
| ; |
| /x, |
| massager => sub { |
| return ("", $3, 'T', "", "$2(*)$4", all_conds()) |
| if defined $1; |
| return ("", $3, 'V', "$2(*)$4", "$2(*)$4", all_conds()); |
| }, |
| }, |
| # Function declaration, or typedef thereof |
| # This sort of declaration may have a body (inline functions, for example) |
| { regexp => qr/(?:(typedef)\s?)? # Possible typedef ($1) |
| ((?:\w|\*|\s)*?) # Return type ($2) |
| \s? # Possible space |
| ([[:alpha:]_]\w*) # Function name ($3) |
| <<<(\(.*\))>>> # Parameters ($4) |
| (?:<<<\{.*\}>>>|;) # Body or semicolon |
| /x, |
| massager => sub { |
| return ("", $3, 'T', "", "$2$4", all_conds()) |
| if defined $1; |
| return ("", $3, 'F', $2, "$2$4", all_conds()); |
| }, |
| }, |
| # Variable declaration, including arrays, or typedef thereof |
| { regexp => qr/(?:(typedef)\s?)? # Possible typedef ($1) |
| ((?:\w|\*|\s)*?) # Type ($2) |
| \s? # Possible space |
| ([[:alpha:]_]\w*) # Variable name ($3) |
| ((?:<<<\[[^\]]*\]>>>)*) # Possible array declaration ($4) |
| ; |
| /x, |
| massager => sub { |
| return ("", $3, 'T', "", $2.($4||""), all_conds()) |
| if defined $1; |
| return ("", $3, 'V', $2.($4||""), $2.($4||""), all_conds()); |
| }, |
| }, |
| ); |
| |
| # End handlers are almost the same as handlers, except they are run through |
| # ONCE when the input has been parsed through. These are used to check for |
| # remaining stuff, such as an unfinished #ifdef and stuff like that that the |
| # main parser can't check on its own. |
| my @endhandlers = ( |
| { massager => sub { |
| my %opts = %{$_[0]}; |
| |
| die "Unfinished preprocessor conditions levels: ",scalar(@preprocessor_conds),($opts{filename} ? " in file ".$opts{filename}: ""),$opts{PLACE} |
| if @preprocessor_conds; |
| } |
| } |
| ); |
| |
| # takes a list of strings that can each contain one or several lines of code |
| # also takes a hash of options as last argument. |
| # |
| # returns a list of hashes with information: |
| # |
| # name name of the thing |
| # type type, see the massage handler function |
| # returntype return type of functions and variables |
| # value value for macros, signature for functions, variables |
| # and structs |
| # conds preprocessor conditions (array ref) |
| |
| sub parse { |
| my %opts; |
| if (ref($_[$#_]) eq "HASH") { |
| %opts = %{$_[$#_]}; |
| pop @_; |
| } |
| my %state = ( |
| in_extern_C => 0, # An exception to parenthesis processing. |
| cpp_parens => [], # A list of ending parens and braces found in |
| # preprocessor directives |
| c_parens => [], # A list of ending parens and braces found in |
| # C statements |
| in_string => "", # empty string when outside a string, otherwise |
| # "'" or '"' depending on the starting quote. |
| in_comment => "", # empty string when outside a comment, otherwise |
| # "/*" or "//" depending on the type of comment |
| # found. The latter will never be multiline |
| # NOTE: in_string and in_comment will never be |
| # true (in perl semantics) at the same time. |
| current_line => 0, |
| ); |
| my @result = (); |
| my $normalized_line = ""; # $input_line, but normalized. In essence, this |
| # means that ALL whitespace is removed unless |
| # it absolutely has to be present, and in that |
| # case, there's only one space. |
| # The cases where a space needs to stay present |
| # are: |
| # 1. between words |
| # 2. between words and number |
| # 3. after the first word of a preprocessor |
| # directive. |
| # 4. for the #define directive, between the macro |
| # name/args and its value, so we end up with: |
| # #define FOO val |
| # #define BAR(x) something(x) |
| my $collected_stmt = ""; # Where we're building up a C line until it's a |
| # complete definition/declaration, as determined |
| # by any handler being capable of matching it. |
| |
| # We use $_ shamelessly when looking through @lines. |
| # In case we find a \ at the end, we keep filling it up with more lines. |
| $_ = undef; |
| |
| foreach my $line (@_) { |
| # split tries to be smart when a string ends with the thing we split on |
| $line .= "\n" unless $line =~ m|\R$|; |
| $line .= "#"; |
| |
| # We use ¦undef¦ as a marker for a new line from the file. |
| # Since we convert one line to several and unshift that into @lines, |
| # that's the only safe way we have to track the original lines |
| my @lines = map { ( undef, $_ ) } split $/, $line; |
| |
| # Remember that extra # we added above? Now we remove it |
| pop @lines; |
| pop @lines; # Don't forget the undef |
| |
| while (@lines) { |
| if (!defined($lines[0])) { |
| shift @lines; |
| $state{current_line}++; |
| if (!defined($_)) { |
| $opts{PLACE} = " at ".$opts{filename}." line ".$state{current_line}."\n"; |
| $opts{PLACE2} = $opts{filename}.":".$state{current_line}; |
| } |
| next; |
| } |
| |
| $_ = "" unless defined $_; |
| $_ .= shift @lines; |
| |
| if (m|\\$|) { |
| $_ = $`; |
| next; |
| } |
| |
| if ($opts{debug}) { |
| print STDERR "DEBUG:----------------------------\n"; |
| print STDERR "DEBUG: \$_ = '$_'\n"; |
| } |
| |
| ########################################################## |
| # Now that we have a full line, let's process through it |
| while(1) { |
| unless ($state{in_comment}) { |
| # Begin with checking if the current $normalized_line |
| # contains a preprocessor directive |
| # This is only done if we're not inside a comment and |
| # if it's a preprocessor directive and it's finished. |
| if ($normalized_line =~ m|^#| && $_ eq "") { |
| print STDERR "DEBUG[OPENSSL CPP]: \$normalized_line = '$normalized_line'\n" |
| if $opts{debug}; |
| $opts{debug_type} = "OPENSSL CPP"; |
| my @r = ( _run_handlers($normalized_line, |
| @opensslcpphandlers, |
| \%opts) ); |
| if (shift @r) { |
| # Checking if there are lines to inject. |
| if (@r) { |
| @r = split $/, (pop @r).$_; |
| print STDERR "DEBUG[OPENSSL CPP]: injecting '", join("', '", @r),"'\n" |
| if $opts{debug} && @r; |
| @lines = ( @r, @lines ); |
| |
| $_ = ""; |
| } |
| } else { |
| print STDERR "DEBUG[CPP]: \$normalized_line = '$normalized_line'\n" |
| if $opts{debug}; |
| $opts{debug_type} = "CPP"; |
| my @r = ( _run_handlers($normalized_line, |
| @cpphandlers, |
| \%opts) ); |
| if (shift @r) { |
| if (ref($r[0]) eq "HASH") { |
| push @result, shift @r; |
| } |
| |
| # Now, check if there are lines to inject. |
| # Really, this should never happen, it IS a |
| # preprocessor directive after all... |
| if (@r) { |
| @r = split $/, pop @r; |
| print STDERR "DEBUG[CPP]: injecting '", join("', '", @r),"'\n" |
| if $opts{debug} && @r; |
| @lines = ( @r, @lines ); |
| $_ = ""; |
| } |
| } |
| } |
| |
| # Note: we simply ignore all directives that no |
| # handler matches |
| $normalized_line = ""; |
| } |
| |
| # If the two strings end and start with a character that |
| # shouldn't get concatenated, add a space |
| my $space = |
| ($collected_stmt =~ m/(?:"|')$/ |
| || ($collected_stmt =~ m/(?:\w|\d)$/ |
| && $normalized_line =~ m/^(?:\w|\d)/)) ? " " : ""; |
| |
| # Now, unless we're building up a preprocessor directive or |
| # are in the middle of a string, or the parens et al aren't |
| # balanced up yet, let's try and see if there's a OpenSSL |
| # or C handler that can make sense of what we have so far. |
| if ( $normalized_line !~ m|^#| |
| && ($collected_stmt ne "" || $normalized_line ne "") |
| && ! @{$state{c_parens}} |
| && ! $state{in_string} ) { |
| if ($opts{debug}) { |
| print STDERR "DEBUG[OPENSSL C]: \$collected_stmt = '$collected_stmt'\n"; |
| print STDERR "DEBUG[OPENSSL C]: \$normalized_line = '$normalized_line'\n"; |
| } |
| $opts{debug_type} = "OPENSSL C"; |
| my @r = ( _run_handlers($collected_stmt |
| .$space |
| .$normalized_line, |
| @opensslchandlers, |
| \%opts) ); |
| if (shift @r) { |
| # Checking if there are lines to inject. |
| if (@r) { |
| @r = split $/, (pop @r).$_; |
| print STDERR "DEBUG[OPENSSL]: injecting '", join("', '", @r),"'\n" |
| if $opts{debug} && @r; |
| @lines = ( @r, @lines ); |
| |
| $_ = ""; |
| } |
| $normalized_line = ""; |
| $collected_stmt = ""; |
| } else { |
| if ($opts{debug}) { |
| print STDERR "DEBUG[C]: \$collected_stmt = '$collected_stmt'\n"; |
| print STDERR "DEBUG[C]: \$normalized_line = '$normalized_line'\n"; |
| } |
| $opts{debug_type} = "C"; |
| my @r = ( _run_handlers($collected_stmt |
| .$space |
| .$normalized_line, |
| @chandlers, |
| \%opts) ); |
| if (shift @r) { |
| if (ref($r[0]) eq "HASH") { |
| push @result, shift @r; |
| } |
| |
| # Checking if there are lines to inject. |
| if (@r) { |
| @r = split $/, (pop @r).$_; |
| print STDERR "DEBUG[C]: injecting '", join("', '", @r),"'\n" |
| if $opts{debug} && @r; |
| @lines = ( @r, @lines ); |
| |
| $_ = ""; |
| } |
| $normalized_line = ""; |
| $collected_stmt = ""; |
| } |
| } |
| } |
| if ($_ eq "") { |
| $collected_stmt .= $space.$normalized_line; |
| $normalized_line = ""; |
| } |
| } |
| |
| if ($_ eq "") { |
| $_ = undef; |
| last; |
| } |
| |
| # Take care of inside string first. |
| if ($state{in_string}) { |
| if (m/ (?:^|(?<!\\)) # Make sure it's not escaped |
| $state{in_string} # Look for matching quote |
| /x) { |
| $normalized_line .= $`.$&; |
| $state{in_string} = ""; |
| $_ = $'; |
| next; |
| } else { |
| die "Unfinished string without continuation found$opts{PLACE}\n"; |
| } |
| } |
| # ... or inside comments, whichever happens to apply |
| elsif ($state{in_comment}) { |
| |
| # This should never happen |
| die "Something went seriously wrong, multiline //???$opts{PLACE}\n" |
| if ($state{in_comment} eq "//"); |
| |
| # A note: comments are simply discarded. |
| |
| if (m/ (?:^|(?<!\\)) # Make sure it's not escaped |
| \*\/ # Look for C comment end |
| /x) { |
| $state{in_comment} = ""; |
| $_ = $'; |
| print STDERR "DEBUG: Found end of comment, followed by '$_'\n" |
| if $opts{debug}; |
| next; |
| } else { |
| $_ = ""; |
| next; |
| } |
| } |
| |
| # At this point, it's safe to remove leading whites, but |
| # we need to be careful with some preprocessor lines |
| if (m|^\s+|) { |
| my $rest = $'; |
| my $space = ""; |
| $space = " " |
| if ($normalized_line =~ m/^ |
| \#define\s\w(?:\w|\d)*(?:<<<\([^\)]*\)>>>)? |
| | \#[a-z]+ |
| $/x); |
| print STDERR "DEBUG: Processing leading spaces: \$normalized_line = '$normalized_line', \$space = '$space', \$rest = '$rest'\n" |
| if $opts{debug}; |
| $_ = $space.$rest; |
| } |
| |
| my $parens = |
| $normalized_line =~ m|^#| ? 'cpp_parens' : 'c_parens'; |
| (my $paren_singular = $parens) =~ s|s$||; |
| |
| # Now check for specific tokens, and if they are parens, |
| # check them against $state{$parens}. Note that we surround |
| # the outermost parens with extra "<<<" and ">>>". Those |
| # are for the benefit of handlers who to need to detect |
| # them, and they will be removed from the final output. |
| if (m|^[\{\[\(]|) { |
| my $body = $&; |
| $_ = $'; |
| if (!@{$state{$parens}}) { |
| if ("$normalized_line$body" =~ m|^extern "C"\{$|) { |
| $state{in_extern_C} = 1; |
| print STDERR "DEBUG: found start of 'extern \"C\"' ($normalized_line$body)\n" |
| if $opts{debug}; |
| $normalized_line = ""; |
| } else { |
| $normalized_line .= "<<<".$body; |
| } |
| } else { |
| $normalized_line .= $body; |
| } |
| |
| if ($normalized_line ne "") { |
| print STDERR "DEBUG: found $paren_singular start '$body'\n" |
| if $opts{debug}; |
| $body =~ tr|\{\[\(|\}\]\)|; |
| print STDERR "DEBUG: pushing $paren_singular end '$body'\n" |
| if $opts{debug}; |
| push @{$state{$parens}}, $body; |
| } |
| } elsif (m|^[\}\]\)]|) { |
| $_ = $'; |
| |
| if (!@{$state{$parens}} |
| && $& eq '}' && $state{in_extern_C}) { |
| print STDERR "DEBUG: found end of 'extern \"C\"'\n" |
| if $opts{debug}; |
| $state{in_extern_C} = 0; |
| } else { |
| print STDERR "DEBUG: Trying to match '$&' against '" |
| ,join("', '", @{$state{$parens}}) |
| ,"'\n" |
| if $opts{debug}; |
| die "Unmatched parentheses$opts{PLACE}\n" |
| unless (@{$state{$parens}} |
| && pop @{$state{$parens}} eq $&); |
| if (!@{$state{$parens}}) { |
| $normalized_line .= $&.">>>"; |
| } else { |
| $normalized_line .= $&; |
| } |
| } |
| } elsif (m|^["']|) { # string start |
| my $body = $&; |
| $_ = $'; |
| |
| # We want to separate strings from \w and \d with one space. |
| $normalized_line .= " " if $normalized_line =~ m/(\w|\d)$/; |
| $normalized_line .= $body; |
| $state{in_string} = $body; |
| } elsif (m|^\/\*|) { # C style comment |
| print STDERR "DEBUG: found start of C style comment\n" |
| if $opts{debug}; |
| $state{in_comment} = $&; |
| $_ = $'; |
| } elsif (m|^\/\/|) { # C++ style comment |
| print STDERR "DEBUG: found C++ style comment\n" |
| if $opts{debug}; |
| $_ = ""; # (just discard it entirely) |
| } elsif (m/^ (?| (?: 0[xX][[:xdigit:]]+ | 0[bB][01]+ | [0-9]+ ) |
| (?i: U | L | UL | LL | ULL )? |
| | [0-9]+\.[0-9]+(?:[eE][\-\+]\d+)? (?i: F | L)? |
| ) /x) { |
| print STDERR "DEBUG: Processing numbers: \$normalized_line = '$normalized_line', \$& = '$&', \$' = '$''\n" |
| if $opts{debug}; |
| $normalized_line .= $&; |
| $_ = $'; |
| } elsif (m/^[[:alpha:]_]\w*/) { |
| my $body = $&; |
| my $rest = $'; |
| my $space = ""; |
| |
| # Now, only add a space if it's needed to separate |
| # two \w characters, and we also surround strings with |
| # a space. In this case, that's if $normalized_line ends |
| # with a \w, \d, " or '. |
| $space = " " |
| if ($normalized_line =~ m/("|')$/ |
| || ($normalized_line =~ m/(\w|\d)$/ |
| && $body =~ m/^(\w|\d)/)); |
| |
| print STDERR "DEBUG: Processing words: \$normalized_line = '$normalized_line', \$space = '$space', \$body = '$body', \$rest = '$rest'\n" |
| if $opts{debug}; |
| $normalized_line .= $space.$body; |
| $_ = $rest; |
| } elsif (m|^(?:\\)?.|) { # Catch-all |
| $normalized_line .= $&; |
| $_ = $'; |
| } |
| } |
| } |
| } |
| foreach my $handler (@endhandlers) { |
| if ($handler->{massager}) { |
| $handler->{massager}->(\%opts); |
| } |
| } |
| return @result; |
| } |
| |
| # arg1: line to check |
| # arg2...: handlers to check |
| # return undef when no handler matched |
| sub _run_handlers { |
| my %opts; |
| if (ref($_[$#_]) eq "HASH") { |
| %opts = %{$_[$#_]}; |
| pop @_; |
| } |
| my $line = shift; |
| my @handlers = @_; |
| |
| foreach my $handler (@handlers) { |
| if ($handler->{regexp} |
| && $line =~ m|^$handler->{regexp}$|) { |
| if ($handler->{massager}) { |
| if ($opts{debug}) { |
| print STDERR "DEBUG[",$opts{debug_type},"]: Trying to handle '$line'\n"; |
| print STDERR "DEBUG[",$opts{debug_type},"]: (matches /\^",$handler->{regexp},"\$/)\n"; |
| } |
| my $saved_line = $line; |
| my @massaged = |
| map { s/(<<<|>>>)//g; $_ } |
| $handler->{massager}->($saved_line, \%opts); |
| print STDERR "DEBUG[",$opts{debug_type},"]: Got back '" |
| , join("', '", @massaged), "'\n" |
| if $opts{debug}; |
| |
| # Because we may get back new lines to be |
| # injected before whatever else that follows, |
| # and the injected stuff might include |
| # preprocessor lines, we need to inject them |
| # in @lines and set $_ to the empty string to |
| # break out from the inner loops |
| my $injected_lines = shift @massaged || ""; |
| |
| if (@massaged) { |
| return (1, |
| { |
| name => shift @massaged, |
| type => shift @massaged, |
| returntype => shift @massaged, |
| value => shift @massaged, |
| conds => [ @massaged ] |
| }, |
| $injected_lines |
| ); |
| } else { |
| print STDERR "DEBUG[",$opts{debug_type},"]: (ignore, possible side effects)\n" |
| if $opts{debug} && $injected_lines eq ""; |
| return (1, $injected_lines); |
| } |
| } |
| return (1); |
| } |
| } |
| return (0); |
| } |