cstyle

#!/usr/bin/perl -w
#
# CDDL HEADER START
#
# The contents of this file are subject to the terms of the
# Common Development and Distribution License (the "License").
# You may not use this file except in compliance with the License.
#
# You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
# or http://www.opensolaris.org/os/licensing.
# See the License for the specific language governing permissions
# and limitations under the License.
#
# When distributing Covered Code, include this CDDL HEADER in each
# file and include the License file at usr/src/OPENSOLARIS.LICENSE.
# If applicable, add the following below this CDDL HEADER, with the
# fields enclosed by brackets "[]" replaced with your own identifying
# information: Portions Copyright [yyyy] [name of copyright owner]
#
# CDDL HEADER END
#
#
# Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
# Use is subject to license terms.
#
# @(#)cstyle 1.58 98/09/09 (from shannon)
#ident	"%Z%%M%	%I%	%E% SMI"
#
# cstyle - check for some common stylistic errors.
#
#	cstyle is a sort of "lint" for C coding style.
#	It attempts to check for the style used in the
#	kernel, sometimes known as "Bill Joy Normal Form".
#
#	There's a lot this can't check for, like proper indentation
#	of code blocks.  There's also a lot more this could check for.
#
#	A note to the non perl literate:
#
#		perl regular expressions are pretty much like egrep
#		regular expressions, with the following special symbols
#
#		\s	any space character
#		\S	any non-space character
#		\w	any "word" character [a-zA-Z0-9_]
#		\W	any non-word character
#		\d	a digit [0-9]
#		\D	a non-digit
#		\b	word boundary (between \w and \W)
#		\B	non-word boundary
#

require 5.0;
use IO::File;
use Getopt::Std;
use strict;

my $usage =
"usage: cstyle [-chpvCP] [-o constructs] file ...
	-c	check continuation indentation inside functions
	-h	perform heuristic checks that are sometimes wrong
	-p	perform some of the more picky checks
	-v	verbose
	-C	don't check anything in header block comments
	-P	check for use of non-POSIX types
	-o constructs
		allow a comma-seperated list of optional constructs:
		    doxygen	allow doxygen-style block comments (/** /*!)
		    splint	allow splint-style lint comments (/*@ ... @*/)
";

my %opts;

if (!getopts("cho:pvCP", \%opts)) {
	print $usage;
	exit 2;
}

my $check_continuation = $opts{'c'};
my $heuristic = $opts{'h'};
my $picky = $opts{'p'};
my $verbose = $opts{'v'};
my $ignore_hdr_comment = $opts{'C'};
my $check_posix_types = $opts{'P'};

my $doxygen_comments = 0;
my $splint_comments = 0;

if (defined($opts{'o'})) {
	for my $x (split /,/, $opts{'o'}) {
		if ($x eq "doxygen") {
			$doxygen_comments = 1;
		} elsif ($x eq "splint") {
			$splint_comments = 1;
		} else {
			print "cstyle: unrecognized construct \"$x\"\n";
			print $usage;
			exit 2;
		}
	}
}

my ($filename, $line, $prev);		# shared globals

my $fmt;
my $hdr_comment_start;

if ($verbose) {
	$fmt = "%s: %d: %s\n%s\n";
} else {
	$fmt = "%s: %d: %s\n";
}

if ($doxygen_comments) {
	# doxygen comments look like "/*!" or "/**"; allow them.
	$hdr_comment_start = qr/^\s*\/\*[\!\*]?$/;
} else {
	$hdr_comment_start = qr/^\s*\/\*$/;
}

# Note, following must be in single quotes so that \s and \w work right.
my $typename = '(int|char|short|long|unsigned|float|double' .
    '|\w+_t|struct\s+\w+|union\s+\w+|FILE)';

# mapping of old types to POSIX compatible types
my %old2posix = (
	'unchar' => 'uchar_t',
	'ushort' => 'ushort_t',
	'uint' => 'uint_t',
	'ulong' => 'ulong_t',
	'u_int' => 'uint_t',
	'u_short' => 'ushort_t',
	'u_long' => 'ulong_t',
	'u_char' => 'uchar_t',
	'quad' => 'quad_t'
);

my $lint_re = qr/\/\*(?:
	ARGSUSED[0-9]*|NOTREACHED|LINTLIBRARY|VARARGS[0-9]*|
	CONSTCOND|CONSTANTCOND|CONSTANTCONDITION|EMPTY|
	FALLTHRU|FALLTHROUGH|LINTED.*?|PRINTFLIKE[0-9]*|
	PROTOLIB[0-9]*|SCANFLIKE[0-9]*|CSTYLED.*?
    )\*\//x;

my $splint_re = qr/\/\*@.*?@\*\//x;

my $warlock_re = qr/\/\*\s*(?:
	VARIABLES\ PROTECTED\ BY|
	MEMBERS\ PROTECTED\ BY|
	ALL\ MEMBERS\ PROTECTED\ BY|
	READ-ONLY\ VARIABLES:|
	READ-ONLY\ MEMBERS:|
	VARIABLES\ READABLE\ WITHOUT\ LOCK:|
	MEMBERS\ READABLE\ WITHOUT\ LOCK:|
	LOCKS\ COVERED\ BY|
	LOCK\ UNNEEDED\ BECAUSE|
	LOCK\ NEEDED:|
	LOCK\ HELD\ ON\ ENTRY:|
	READ\ LOCK\ HELD\ ON\ ENTRY:|
	WRITE\ LOCK\ HELD\ ON\ ENTRY:|
	LOCK\ ACQUIRED\ AS\ SIDE\ EFFECT:|
	READ\ LOCK\ ACQUIRED\ AS\ SIDE\ EFFECT:|
	WRITE\ LOCK\ ACQUIRED\ AS\ SIDE\ EFFECT:|
	LOCK\ RELEASED\ AS\ SIDE\ EFFECT:|
	LOCK\ UPGRADED\ AS\ SIDE\ EFFECT:|
	LOCK\ DOWNGRADED\ AS\ SIDE\ EFFECT:|
	FUNCTIONS\ CALLED\ THROUGH\ POINTER|
	FUNCTIONS\ CALLED\ THROUGH\ MEMBER|
	LOCK\ ORDER:
    )/x;

my $err_stat = 0;		# exit status

if ($#ARGV >= 0) {
	foreach my $arg (@ARGV) {
		my $fh = new IO::File $arg, "r";
		if (!defined($fh)) {
			printf "%s: can not open\n", $arg;
		} else {
			&cstyle($arg, $fh);
			close $fh;
		}
	}
} else {
	&cstyle("<stdin>", *STDIN);
}
exit $err_stat;

my $no_errs = 0;		# set for CSTYLED-protected lines

sub err($) {
	my ($error) = @_;
	unless ($no_errs) {
		printf $fmt, $filename, $., $error, $line;
		$err_stat = 1;
	}
}

sub err_prefix($$) {
	my ($prevline, $error) = @_;
	my $out = $prevline."\n".$line;
	unless ($no_errs) {
		printf $fmt, $filename, $., $error, $out;
		$err_stat = 1;
	}
}

sub err_prev($) {
	my ($error) = @_;
	unless ($no_errs) {
		printf $fmt, $filename, $. - 1, $error, $prev;
		$err_stat = 1;
	}
}

sub cstyle($$) {

my ($fn, $filehandle) = @_;
$filename = $fn;			# share it globally

my $in_cpp = 0;
my $next_in_cpp = 0;

my $in_comment = 0;
my $in_header_comment = 0;
my $comment_done = 0;
my $in_warlock_comment = 0;
my $in_function = 0;
my $in_function_header = 0;
my $in_declaration = 0;
my $note_level = 0;
my $nextok = 0;
my $nocheck = 0;

my $in_string = 0;

my ($okmsg, $comment_prefix);

$line = '';
$prev = '';
reset_indent();

line: while (<$filehandle>) {
	s/\r?\n$//;	# strip return and newline

	# save the original line, then remove all text from within
	# double or single quotes, we do not want to check such text.

	$line = $_;

	#
	# C allows strings to be continued with a backslash at the end of
	# the line.  We translate that into a quoted string on the previous
	# line followed by an initial quote on the next line.
	#
	# (we assume that no-one will use backslash-continuation with character
	# constants)
	#
	$_ = '"' . $_		if ($in_string && !$nocheck && !$in_comment);

	#
	# normal strings and characters
	#
	s/'([^\\']|\\[^xX0]|\\0[0-9]*|\\[xX][0-9a-fA-F]*)'/''/g;
	s/"([^\\"]|\\.)*"/\"\"/g;

	#
	# detect string continuation
	#
	if ($nocheck || $in_comment) {
		$in_string = 0;
	} else {
		#
		# Now that all full strings are replaced with "", we check
		# for unfinished strings continuing onto the next line.
		#
		$in_string =
		    (s/([^"](?:"")*)"([^\\"]|\\.)*\\$/$1""/ ||
		    s/^("")*"([^\\"]|\\.)*\\$/""/);
	}

	#
	# figure out if we are in a cpp directive
	#
	$in_cpp = $next_in_cpp || /^\s*#/;	# continued or started
	$next_in_cpp = $in_cpp && /\\$/;	# only if continued

	# strip off trailing backslashes, which appear in long macros
	s/\s*\\$//;

	# an /* END CSTYLED */ comment ends a no-check block.
	if ($nocheck) {
		if (/\/\* *END *CSTYLED *\*\//) {
			$nocheck = 0;
		} else {
			reset_indent();
			next line;
		}
	}

	# a /*CSTYLED*/ comment indicates that the next line is ok.
	if ($nextok) {
		if ($okmsg) {
			err($okmsg);
		}
		$nextok = 0;
		$okmsg = 0;
		if (/\/\* *CSTYLED.*\*\//) {
			/^.*\/\* *CSTYLED *(.*) *\*\/.*$/;
			$okmsg = $1;
			$nextok = 1;
		}
		$no_errs = 1;
	} elsif ($no_errs) {
		$no_errs = 0;
	}

	# check length of line.
	# first, a quick check to see if there is any chance of being too long.
	if (($line =~ tr/\t/\t/) * 7 + length($line) > 80) {
		# yes, there is a chance.
		# replace tabs with spaces and check again.
		my $eline = $line;
		1 while $eline =~
		    s/\t+/' ' x (length($&) * 8 - length($`) % 8)/e;
		if (length($eline) > 80) {
			err("line > 80 characters");
		}
	}

	# ignore NOTE(...) annotations (assumes NOTE is on lines by itself).
	if ($note_level || /\b_?NOTE\s*\(/) { # if in NOTE or this is NOTE
		s/[^()]//g;			  # eliminate all non-parens
		$note_level += s/\(//g - length;  # update paren nest level
		next;
	}

	# a /* BEGIN CSTYLED */ comment starts a no-check block.
	if (/\/\* *BEGIN *CSTYLED *\*\//) {
		$nocheck = 1;
	}

	# a /*CSTYLED*/ comment indicates that the next line is ok.
	if (/\/\* *CSTYLED.*\*\//) {
		/^.*\/\* *CSTYLED *(.*) *\*\/.*$/;
		$okmsg = $1;
		$nextok = 1;
	}
	if (/\/\/ *CSTYLED/) {
		/^.*\/\/ *CSTYLED *(.*)$/;
		$okmsg = $1;
		$nextok = 1;
	}

	# universal checks; apply to everything
	if (/\t +\t/) {
		err("spaces between tabs");
	}
	if (/ \t+ /) {
		err("tabs between spaces");
	}
	if (/\s$/) {
		err("space or tab at end of line");
	}
	if (/[^ \t(]\/\*/ && !/\w\(\/\*.*\*\/\);/) {
		err("comment preceded by non-blank");
	}

	# is this the beginning or ending of a function?
	# (not if "struct foo\n{\n")
	if (/^{$/ && $prev =~ /\)\s*(const\s*)?(\/\*.*\*\/\s*)?\\?$/) {
		$in_function = 1;
		$in_declaration = 1;
		$in_function_header = 0;
		$prev = $line;
		next line;
	}
	if (/^}\s*(\/\*.*\*\/\s*)*$/) {
		if ($prev =~ /^\s*return\s*;/) {
			err_prev("unneeded return at end of function");
		}
		$in_function = 0;
		reset_indent();		# we don't check between functions
		$prev = $line;
		next line;
	}
	if (/^\w*\($/) {
		$in_function_header = 1;
	}

	if ($in_warlock_comment && /\*\//) {
		$in_warlock_comment = 0;
		$prev = $line;
		next line;
	}

	# a blank line terminates the declarations within a function.
	# XXX - but still a problem in sub-blocks.
	if ($in_declaration && /^$/) {
		$in_declaration = 0;
	}

	if ($comment_done) {
		$in_comment = 0;
		$in_header_comment = 0;
		$comment_done = 0;
	}
	# does this looks like the start of a block comment?
	if (/$hdr_comment_start/) {
		$in_comment = 1;
		/^(\s*)\//;
		$comment_prefix = $1;
		if ($comment_prefix eq "") {
			$in_header_comment = 1;
		}
		$prev = $line;
		next line;
	}
	# are we still in the block comment?
	if ($in_comment) {
		if (/^$comment_prefix \*\/$/) {
			$comment_done = 1;
		} elsif (/\*\//) {
			$comment_done = 1;
			err("improper block comment close")
			    unless ($ignore_hdr_comment && $in_header_comment);
		} elsif (!/^$comment_prefix \*[ \t]/ &&
		    !/^$comment_prefix \*$/) {
			err("improper block comment")
			    unless ($ignore_hdr_comment && $in_header_comment);
		}
	}

	if ($in_header_comment && $ignore_hdr_comment) {
		$prev = $line;
		next line;
	}

	# check for errors that might occur in comments and in code.

	# allow spaces to be used to draw pictures in header comments.
	if (/^\t+ [^ \t\*]/ || /^\t+  \S/ || /^\t+   \S/) {
		err("continuation line not indented by 4 spaces");
	}
	if (/$warlock_re/ && !/\*\//) {
		$in_warlock_comment = 1;
		$prev = $line;
		next line;
	}
	if (/^\s*\/\*./ && !/^\s*\/\*.*\*\// && !/$hdr_comment_start/) {
		err("improper first line of block comment");
	}

	if ($in_comment) {	# still in comment, don't do further checks
		$prev = $line;
		next line;
	}

	if ((/[^(]\/\*\S/ || /^\/\*\S/) &&
	    !(/$lint_re/ || ($splint_comments && /$splint_re/))) {
		err("missing blank after open comment");
	}
	if (/\S\*\/[^)]|\S\*\/$/ &&
	    !(/$lint_re/ || ($splint_comments && /$splint_re/))) {
		err("missing blank before close comment");
	}
	if (/\/\/\S/) {		# C++ comments
		err("missing blank after start comment");
	}
	# check for unterminated single line comments, but allow them when
	# they are used to comment out the argument list of a function
	# declaration.
	if (/\S.*\/\*/ && !/\S.*\/\*.*\*\// && !/\(\/\*/) {
		err("unterminated single line comment");
	}

	if (/^(#else|#endif|#include)(.*)$/) {
		$prev = $line;
		if ($picky) {
			my $directive = $1;
			my $clause = $2;
			# Enforce ANSI rules for #else and #endif: no noncomment
			# identifiers are allowed after #endif or #else.  Allow
			# C++ comments since they seem to be a fact of life.
			if ((($1 eq "#endif") || ($1 eq "#else")) &&
			    ($clause ne "") &&
			    (!($clause =~ /^\s+\/\*.*\*\/$/)) &&
			    (!($clause =~ /^\s+\/\/.*$/))) {
				err("non-comment text following " .
				    "$directive (or malformed $directive " .
				    "directive)");
			}
		}
		next line;
	}

	#
	# delete any comments and check everything else.  Note that
	# ".*?" is a non-greedy match, so that we don't get confused by
	# multiple comments on the same line.
	#
	s/\/\*.*?\*\//
/g;
	s/\/\/.*$/
/;		# C++ comments

	# delete any trailing whitespace; we have already checked for that.
	s/\s*$//;

	# following checks do not apply to text in comments.

	if (/[^<>\s][!<>=]=/ || /[^<>][!<>=]=[^\s,]/ ||
	    (/[^->]>[^,=>\s]/ && !/[^->]>$/) ||
	    (/[^<]<[^,=<\s]/ && !/[^<]<$/) ||
	    /[^<\s]<[^<]/ || /[^->\s]>[^>]/) {
		err("missing space around relational operator");
	}
	if (/\S>>=/ || /\S<<=/ || />>=\S/ || /<<=\S/ || /\S[-+*\/&|^%]=/ ||
	    (/[^-+*\/&|^%!<>=\s]=[^=]/ && !/[^-+*\/&|^%!<>=\s]=$/) ||
	    (/[^!<>=]=[^=\s]/ && !/[^!<>=]=$/)) {
		# XXX - should only check this for C++ code
		# XXX - there are probably other forms that should be allowed
		if (!/\soperator=/) {
			err("missing space around assignment operator");
		}
	}
	if (/[,;]\S/ && !/\bfor \(;;\)/) {
		err("comma or semicolon followed by non-blank");
	}
	# allow "for" statements to have empty "while" clauses
	if (/\s[,;]/ && !/^[\t]+;$/ && !/^\s*for \([^;]*; ;[^;]*\)/) {
		err("comma or semicolon preceded by blank");
	}
	if (/^\s*(&&|\|\|)/) {
		err("improper boolean continuation");
	}
	if (/\S   *(&&|\|\|)/ || /(&&|\|\|)   *\S/) {
		err("more than one space around boolean operator");
	}
	if (/\b(for|if|while|switch|sizeof|return|case)\(/) {
		err("missing space between keyword and paren");
	}
	if (/(\b(for|if|while|switch|return)\b.*){2,}/ && !/^#define/) {
		# multiple "case" and "sizeof" allowed
		err("more than one keyword on line");
	}
	if (/\b(for|if|while|switch|sizeof|return|case)\s\s+\(/ &&
	    !/^#if\s+\(/) {
		err("extra space between keyword and paren");
	}
	# try to detect "func (x)" but not "if (x)" or
	# "#define foo (x)" or "int (*func)();"
	if (/\w\s\(/) {
		my $s = $_;
		# strip off all keywords on the line
		s/\b(for|if|while|switch|return|case|sizeof)\s\(/XXX(/g;
		s/#elif\s\(/XXX(/g;
		s/^#define\s+\w+\s+\(/XXX(/;
		# do not match things like "void (*f)();"
		# or "typedef void (func_t)();"
		s/\w\s\(+\*/XXX(*/g;
		s/\b($typename|void)\s+\(+/XXX(/og;
		if (/\w\s\(/) {
			err("extra space between function name and left paren");
		}
		$_ = $s;
	}
	# try to detect "int foo(x)", but not "extern int foo(x);"
	# XXX - this still trips over too many legitimate things,
	# like "int foo(x,\n\ty);"
#		if (/^(\w+(\s|\*)+)+\w+\(/ && !/\)[;,](\s|
)*$/ &&
#		    !/^(extern|static)\b/) {
#			err("return type of function not on separate line");
#		}
	# this is a close approximation
	if (/^(\w+(\s|\*)+)+\w+\(.*\)(\s|
)*$/ &&
	    !/^(extern|static)\b/) {
		err("return type of function not on separate line");
	}
	if (/^\s*return\W[^;]*;/ && !/^\s*return\s*\(.*\);/) {
		err("unparenthesized return expression");
	}
	if (/\bsizeof\b/ && !/\bsizeof\s*\(.*\)/) {
		err("unparenthesized sizeof expression");
	}
	if (/\(\s/) {
		err("whitespace after left paren");
	}
	# allow "for" statements to have empty "continue" clauses
	if (/\s\)/ && !/^\s*for \([^;]*;[^;]*; \)/) {
		err("whitespace before right paren");
	}
	if (/^\s*\(void\)[^ ]/) {
		err("missing space after (void) cast");
	}
	if (/\S{/ && !/{{/) {
		err("missing space before left brace");
	}
	if ($in_function && /^\s+{/ &&
	    ($prev =~ /\)\s*$/ || $prev =~ /\bstruct\s+\w+$/)) {
		err("left brace starting a line");
	}
	if (/}(else|while)/) {
		err("missing space after right brace");
	}
	if (/}\s\s+(else|while)/) {
		err("extra space after right brace");
	}
	if (/\b_VOID\b|\bVOID\b|\bSTATIC\b/) {
		err("obsolete use of VOID or STATIC");
	}
	if (/\b$typename\*/o) {
		err("missing space between type name and *");
	}
	if (/^\s+#/) {
		err("preprocessor statement not in column 1");
	}
	if (/^#\s/) {
		err("blank after preprocessor #");
	}
	if (/!\s*(strcmp|strncmp|bcmp)\s*\(/) {
		err("don't use boolean ! with comparison functions");
	}

	#
	# We completely ignore, for purposes of indentation:
	#  * lines outside of functions
	#  * preprocessor lines
	#
	if ($check_continuation && $in_function && !$in_cpp) {
		process_indent($_);
	}
	if ($picky) {
		# try to detect spaces after casts, but allow (e.g.)
		# "sizeof (int) + 1", "void (*funcptr)(int) = foo;", and
		# "int foo(int) __NORETURN;"
		if ((/^\($typename( \*+)?\)\s/o ||
		    /\W\($typename( \*+)?\)\s/o) &&
		    !/sizeof\s*\($typename( \*)?\)\s/o &&
		    !/\($typename( \*+)?\)\s+=[^=]/o) {
			err("space after cast");
		}
		if (/\b$typename\s*\*\s/o &&
		    !/\b$typename\s*\*\s+const\b/o) {
			err("unary * followed by space");
		}
	}
	if ($check_posix_types) {
		# try to detect old non-POSIX types.
		# POSIX requires all non-standard typedefs to end in _t,
		# but historically these have been used.
		if (/\b(unchar|ushort|uint|ulong|u_int|u_short|u_long|u_char|quad)\b/) {
			err("non-POSIX typedef $1 used: use $old2posix{$1} instead");
		}
	}
	if ($heuristic) {
		# cannot check this everywhere due to "struct {\n...\n} foo;"
		if ($in_function && !$in_declaration &&
		    /}./ && !/}\s+=/ && !/{.*}[;,]$/ && !/}(\s|
)*$/ &&
		    !/} (else|while)/ && !/}}/) {
			err("possible bad text following right brace");
		}
		# cannot check this because sub-blocks in
		# the middle of code are ok
		if ($in_function && /^\s+{/) {
			err("possible left brace starting a line");
		}
	}
	if (/^\s*else\W/) {
		if ($prev =~ /^\s*}$/) {
			err_prefix($prev,
			    "else and right brace should be on same line");
		}
	}
	$prev = $line;
}

if ($prev eq "") {
	err("last line in file is blank");
}

}

#
# Continuation-line checking
#
# The rest of this file contains the code for the continuation checking
# engine.  It's a pretty simple state machine which tracks the expression
# depth (unmatched '('s and '['s).
#
# Keep in mind that the argument to process_indent() has already been heavily
# processed; all comments have been replaced by control-A, and the contents of
# strings and character constants have been elided.
#

my $cont_in;		# currently inside of a continuation
my $cont_off;		# skipping an initializer or definition
my $cont_noerr;		# suppress cascading errors
my $cont_start;		# the line being continued
my $cont_base;		# the base indentation
my $cont_first;		# this is the first line of a statement
my $cont_multiseg;	# this continuation has multiple segments

my $cont_special;	# this is a C statement (if, for, etc.)
my $cont_macro;		# this is a macro
my $cont_case;		# this is a multi-line case

my @cont_paren;		# the stack of unmatched ( and [s we've seen

sub
reset_indent()
{
	$cont_in = 0;
	$cont_off = 0;
}

sub
delabel($)
{
	#
	# replace labels with tabs.  Note that there may be multiple
	# labels on a line.
	#
	local $_ = $_[0];

	while (/^(\t*)( *(?:(?:\w+\s*)|(?:case\b[^:]*)): *)(.*)$/) {
		my ($pre_tabs, $label, $rest) = ($1, $2, $3);
		$_ = $pre_tabs;
		while ($label =~ s/^([^\t]*)(\t+)//) {
			$_ .= "\t" x (length($2) + length($1) / 8);
		}
		$_ .= ("\t" x (length($label) / 8)).$rest;
	}

	return ($_);
}

sub
process_indent($)
{
	require strict;
	local $_ = $_[0];			# preserve the global $_

	s/
//g;	# No comments
	s/\s+$//;	# Strip trailing whitespace

	return			if (/^$/);	# skip empty lines

	# regexps used below; keywords taking (), macros, and continued cases
	my $special = '(?:(?:\}\s*)?else\s+)?(?:if|for|while|switch)\b';
	my $macro = '[A-Z_][A-Z_0-9]*\(';
	my $case = 'case\b[^:]*$';

	# skip over enumerations, array definitions, initializers, etc.
	if ($cont_off <= 0 && !/^\s*$special/ &&
	    (/(?:(?:\b(?:enum|struct|union)\s*[^\{]*)|(?:\s+=\s*)){/ ||
	    (/^\s*{/ && $prev =~ /=\s*(?:\/\*.*\*\/\s*)*$/))) {
		$cont_in = 0;
		$cont_off = tr/{/{/ - tr/}/}/;
		return;
	}
	if ($cont_off) {
		$cont_off += tr/{/{/ - tr/}/}/;
		return;
	}

	if (!$cont_in) {
		$cont_start = $line;

		if (/^\t* /) {
			err("non-continuation indented 4 spaces");
			$cont_noerr = 1;		# stop reporting
		}
		$_ = delabel($_);	# replace labels with tabs

		# check if the statement is complete
		return		if (/^\s*\}?$/);
		return		if (/^\s*\}?\s*else\s*\{?$/);
		return		if (/^\s*do\s*\{?$/);
		return		if (/{$/);
		return		if (/}[,;]?$/);

		# Allow macros on their own lines
		return		if (/^\s*[A-Z_][A-Z_0-9]*$/);

		# cases we don't deal with, generally non-kosher
		if (/{/) {
			err("stuff after {");
			return;
		}

		# Get the base line, and set up the state machine
		/^(\t*)/;
		$cont_base = $1;
		$cont_in = 1;
		@cont_paren = ();
		$cont_first = 1;
		$cont_multiseg = 0;

		# certain things need special processing
		$cont_special = /^\s*$special/? 1 : 0;
		$cont_macro = /^\s*$macro/? 1 : 0;
		$cont_case = /^\s*$case/? 1 : 0;
	} else {
		$cont_first = 0;

		# Strings may be pulled back to an earlier (half-)tabstop
		unless ($cont_noerr || /^$cont_base    / ||
		    (/^\t*(?:    )?(?:gettext\()?\"/ && !/^$cont_base\t/)) {
			err_prefix($cont_start,
			    "continuation should be indented 4 spaces");
		}
	}

	my $rest = $_;			# keeps the remainder of the line

	#
	# The split matches 0 characters, so that each 'special' character
	# is processed separately.  Parens and brackets are pushed and
	# popped off the @cont_paren stack.  For normal processing, we wait
	# until a ; or { terminates the statement.  "special" processing
	# (if/for/while/switch) is allowed to stop when the stack empties,
	# as is macro processing.  Case statements are terminated with a :
	# and an empty paren stack.
	#
	foreach $_ (split /[^\(\)\[\]\{\}\;\:]*/) {
		next		if (length($_) == 0);

		# rest contains the remainder of the line
		my $rxp = "[^\Q$_\E]*\Q$_\E";
		$rest =~ s/^$rxp//;

		if (/\(/ || /\[/) {
			push @cont_paren, $_;
		} elsif (/\)/ || /\]/) {
			my $cur = $_;
			tr/\)\]/\(\[/;

			my $old = (pop @cont_paren);
			if (!defined($old)) {
				err("unexpected '$cur'");
				$cont_in = 0;
				last;
			} elsif ($old ne $_) {
				err("'$cur' mismatched with '$old'");
				$cont_in = 0;
				last;
			}

			#
			# If the stack is now empty, do special processing
			# for if/for/while/switch and macro statements.
			#
			next		if (@cont_paren != 0);
			if ($cont_special) {
				if ($rest =~ /^\s*{?$/) {
					$cont_in = 0;
					last;
				}
				if ($rest =~ /^\s*;$/) {
					err("empty if/for/while body ".
					    "not on its own line");
					$cont_in = 0;
					last;
				}
				if (!$cont_first && $cont_multiseg == 1) {
					err_prefix($cont_start,
					    "multiple statements continued ".
					    "over multiple lines");
					$cont_multiseg = 2;
				} elsif ($cont_multiseg == 0) {
					$cont_multiseg = 1;
				}
				# We've finished this section, start
				# processing the next.
				goto section_ended;
			}
			if ($cont_macro) {
				if ($rest =~ /^$/) {
					$cont_in = 0;
					last;
				}
			}
		} elsif (/\;/) {
			if ($cont_case) {
				err("unexpected ;");
			} elsif (!$cont_special) {
				err("unexpected ;")	if (@cont_paren != 0);
				if (!$cont_first && $cont_multiseg == 1) {
					err_prefix($cont_start,
					    "multiple statements continued ".
					    "over multiple lines");
					$cont_multiseg = 2;
				} elsif ($cont_multiseg == 0) {
					$cont_multiseg = 1;
				}
				if ($rest =~ /^$/) {
					$cont_in = 0;
					last;
				}
				if ($rest =~ /^\s*special/) {
					err("if/for/while/switch not started ".
					    "on its own line");
				}
				goto section_ended;
			}
		} elsif (/\{/) {
			err("{ while in parens/brackets") if (@cont_paren != 0);
			err("stuff after {")		if ($rest =~ /[^\s}]/);
			$cont_in = 0;
			last;
		} elsif (/\}/) {
			err("} while in parens/brackets") if (@cont_paren != 0);
			if (!$cont_special && $rest !~ /^\s*(while|else)\b/) {
				if ($rest =~ /^$/) {
					err("unexpected }");
				} else {
					err("stuff after }");
				}
				$cont_in = 0;
				last;
			}
		} elsif (/\:/ && $cont_case && @cont_paren == 0) {
			err("stuff after multi-line case") if ($rest !~ /$^/);
			$cont_in = 0;
			last;
		}
		next;
section_ended:
		# End of a statement or if/while/for loop.  Reset
		# cont_special and cont_macro based on the rest of the
		# line.
		$cont_special = ($rest =~ /^\s*$special/)? 1 : 0;
		$cont_macro = ($rest =~ /^\s*$macro/)? 1 : 0;
		$cont_case = 0;
		next;
	}
	$cont_noerr = 0			if (!$cont_in);
}