From 3752be09d3a3c4142a3dadbc306f8985a4234489 Mon Sep 17 00:00:00 2001 From: Paul Poulain Date: Mon, 11 Jun 2012 15:38:35 +0200 Subject: [PATCH] Revert "Subroutine prototypes used at line XXX, column 1. See page 194 of PBP." This reverts commit 583abead1b3d537f78a3d558981b2e9abb2791ff. The translator does not work anymore after this patch has been applied, so reverting it --- misc/translator/TmplTokenizer.pm | 54 ++++++++++++++++---------------- 1 file changed, 27 insertions(+), 27 deletions(-) diff --git a/misc/translator/TmplTokenizer.pm b/misc/translator/TmplTokenizer.pm index 2cdff317e3..6129f8d37e 100644 --- a/misc/translator/TmplTokenizer.pm +++ b/misc/translator/TmplTokenizer.pm @@ -48,21 +48,21 @@ use vars qw( $serial ); ############################################################################### -sub FATAL_P {'fatal-p'} -sub SYNTAXERROR_P {'syntaxerror-p'} +sub FATAL_P () {'fatal-p'} +sub SYNTAXERROR_P () {'syntaxerror-p'} -sub FILENAME {'input'} -#sub HANDLE {'handle'} +sub FILENAME () {'input'} +#sub HANDLE () {'handle'} -#sub READAHEAD {'readahead'} -sub LINENUM_START {'lc_0'} -sub LINENUM {'lc'} -sub CDATA_MODE_P {'cdata-mode-p'} -sub CDATA_CLOSE {'cdata-close'} -#sub PCDATA_MODE_P {'pcdata-mode-p'} # additional submode for CDATA -sub JS_MODE_P {'js-mode-p'} # cdata-mode-p must also be true +#sub READAHEAD () {'readahead'} +sub LINENUM_START () {'lc_0'} +sub LINENUM () {'lc'} +sub CDATA_MODE_P () {'cdata-mode-p'} +sub CDATA_CLOSE () {'cdata-close'} +#sub PCDATA_MODE_P () {'pcdata-mode-p'} # additional submode for CDATA +sub JS_MODE_P () {'js-mode-p'} # cdata-mode-p must also be true -sub ALLOW_CFORMAT_P {'allow-cformat-p'} +sub ALLOW_CFORMAT_P () {'allow-cformat-p'} sub new { shift; @@ -137,10 +137,10 @@ BEGIN { # Perl quoting is really screwed up, but this common subexp is way too long $js_EscapeSequence = q{\\\\(?:['"\\\\bfnrt]|[^0-7xu]|[0-3]?[0-7]{1,2}|x[\da-fA-F]{2}|u[\da-fA-F]{4})}; } -sub parenleft { '(' } -sub parenright { ')' } +sub parenleft () { '(' } +sub parenright () { ')' } -sub _split_js { +sub _split_js ($) { my ($s0) = @_; my @it = (); while (length $s0) { @@ -186,13 +186,13 @@ sub _split_js { return @it; } -sub STATE_UNDERSCORE { 1 } -sub STATE_PARENLEFT { 2 } -sub STATE_STRING_LITERAL { 3 } +sub STATE_UNDERSCORE () { 1 } +sub STATE_PARENLEFT () { 2 } +sub STATE_STRING_LITERAL () { 3 } # XXX This is a crazy hack. I don't want to write an ECMAScript parser. # XXX A scanner is one thing; a parser another thing. -sub _identify_js_translatables { +sub _identify_js_translatables (@) { my @input = @_; my @output = (); # We mark a JavaScript translatable string as in C, i.e., _("literal") @@ -229,7 +229,7 @@ sub _identify_js_translatables { ############################################################################### -sub string_canon ) { +sub string_canon ($) { my $s = shift; # Fold all whitespace into single blanks $s =~ s/\s+/ /g; @@ -238,7 +238,7 @@ sub string_canon ) { } # safer version used internally, preserves new lines -sub string_canon_safe { +sub string_canon_safe ($) { my $s = shift; # fold tabs and spaces into single spaces $s =~ s/[\ \t]+/ /gs; @@ -361,7 +361,7 @@ sub next_token { # function taken from old version # used by tmpl_process3 -sub parametrize { +sub parametrize ($$$$) { my($fmt_0, $cformat_p, $t, $f) = @_; my $it = ''; if ($cformat_p) { @@ -454,12 +454,12 @@ sub parametrize { # Other simple functions (These are not methods) -sub blank_p { +sub blank_p ($) { my($s) = @_; return $s =~ /^(?:\s|\ $re_end_entity|$re_tmpl_var|$re_xsl)*$/osi; } -sub trim { +sub trim ($) { my($s0) = @_; my $l0 = length $s0; my $s = $s0; @@ -468,7 +468,7 @@ sub trim { return wantarray? (substr($s0, 0, $l1), $s, substr($s0, $l0 - $l2)): $s; } -sub quote_po { +sub quote_po ($) { my($s) = @_; # Locale::PO->quote is buggy, it doesn't quote newlines :-/ $s =~ s/([\\"])/\\\1/gs; @@ -477,7 +477,7 @@ sub quote_po { return "\"$s\""; } -sub charset_canon { +sub charset_canon ($) { my($charset) = @_; $charset = uc($charset); $charset = "$1-$2" if $charset =~ /^(ISO|UTF)(\d.*)/i; @@ -510,7 +510,7 @@ use vars qw( @latin1_utf8 ); "\303\270", "\303\271", "\303\272", "\303\273", "\303\274", "\303\275", "\303\276", "\303\277" ); -sub charset_convert { +sub charset_convert ($$$) { my($s, $charset_in, $charset_out) = @_; if ($s !~ /[\200-\377]/s) { # FIXME: don't worry about iso2022 for now ; -- 2.39.5