#!/import/archperf/ws/devtools/4/v9/bin/perl
eval 'exec /import/archperf/ws/devtools/4/v9/bin/perl -S $0 ${1+"$@"}'
if $running_under_some_shell;
use File::Path qw(mkpath);
# Make sure read permissions for all are set:
if (defined umask && (umask() & 0444)) {
use vars qw($opt_D $opt_d $opt_r $opt_l $opt_h $opt_a $opt_Q $opt_e);
die "-r and -a options are mutually exclusive\n" if ($opt_r and $opt_a);
my @inc_dirs = inc_dirs() if $opt_a;
my $Dest_dir = $opt_d || $Config{installsitearch};
die "Destination directory $Dest_dir doesn't exist or isn't a directory\n"
@isatype{@isatype} = (1) x @isatype;
@ARGV = ('-') unless @ARGV;
build_preamble_if_necessary();
my ($t, $tab, %curargs, $new, $eval_index, $dir, $name, $args, $outfile);
my ($incl, $incl_type, $next);
while (defined (my $file = next_file())) {
if (-l $file and -d $file) {
link_if_possible($file) if ($opt_l);
# Recover from header files with unbalanced cpp directives
# $eval_index goes into ``#line'' directives, to help locate syntax errors:
($outfile = $file) =~ s/\.h$/.ph/ || next;
print "$file -> $outfile\n" unless $opt_Q;
if ($file =~ m|^(.*)/|) {
if ($opt_a) { # automagic mode: locate header file in @inc_dirs
open(IN,"$file") || (($Exit = 1),(warn "Can't open $file: $!\n"),next);
open(OUT,">$Dest_dir/$outfile") || die "Can't create $outfile: $!\n";
"require '_h2ph_pre.ph';\n\n",
"no warnings 'redefine';\n\n";
while (defined (local $_ = next_line($file))) {
if (s/^define\s+(\w+)//) {
s/\(\w+\s*\(\*\)\s*\(\w*\)\)\s*(-?\d+)/$1/; # (int (*)(foo_t))0
if (s/^\(([\w,\s]*)\)//) {
foreach my $arg (split(/,\s*/,$args)) {
$arg =~ s/^\s*([^\s].*[^\s])\s*$/$1/;
$args =~ s/\b(\w)/\$$1/g;
$args = "my($args) = \@_;\n$t ";
$new =~ s/(["\\])/\\$1/g; #"]);
$new =~ s/(['\\])/\\$1/g; #']);
"eval \"\\n#line $eval_index $outfile\\n\" . 'sub $name $proto\{\n$t ${args}eval q($new);\n$t}' unless defined(\&$name);\n";
"eval 'sub $name $proto\{\n$t ${args}eval q($new);\n$t}' unless defined(\&$name);\n";
print OUT "unless(defined(\&$name)) {\n sub $name $proto\{\n\t${args}eval q($new);\n }\n}\n";
$new =~ s/(['\\])/\\$1/g; #']);
print OUT $t,"eval \"\\n#line $eval_index $outfile\\n\" . 'sub $name () {",$new,";}' unless defined(\&$name);\n";
print OUT $t,"eval 'sub $name () {",$new,";}' unless defined(\&$name);\n";
# Shunt around such directives as `#define FOO FOO':
next if " \&$name" eq $new;
print OUT $t,"unless(defined(\&$name)) {\n sub $name () {\t",$new,";}\n}\n";
} elsif (/^(include|import|include_next)\s*[<\"](.*)[>\"]/) {
if (($incl_type eq 'include_next') ||
($opt_e && exists($bad_file{$incl}))) {
$t = "\t" x ($tab / 8) . ' ' x ($tab % 8);
print OUT ($t, "my(\@REM);\n");
if ($incl_type eq 'include_next') {
"my(\%INCD) = map { \$INC{\$_} => 1 } ",
"(grep { \$_ eq \"$incl\" } ",
"\@REM = map { \"\$_/$incl\" } ",
"(grep { not exists(\$INCD{\"\$_/$incl\"})",
" and -f \"\$_/$incl\" } \@INC);\n");
"\@REM = map { \"\$_/$incl\" } ",
"(grep {-r \"\$_/$incl\" } \@INC);\n");
"require \"\$REM[0]\" if \@REM;\n");
$t = "\t" x ($tab / 8) . ' ' x ($tab % 8);
"warn(\$\@) if \$\@;\n");
print OUT $t,"require '$incl';\n";
} elsif (/^ifdef\s+(\w+)/) {
print OUT $t,"if(defined(&$1)) {\n";
$t = "\t" x ($tab / 8) . ' ' x ($tab % 8);
} elsif (/^ifndef\s+(\w+)/) {
print OUT $t,"unless(defined(&$1)) {\n";
$t = "\t" x ($tab / 8) . ' ' x ($tab % 8);
print OUT $t,"if($new) {\n";
$t = "\t" x ($tab / 8) . ' ' x ($tab % 8);
$t = "\t" x ($tab / 8) . ' ' x ($tab % 8);
print OUT $t,"}\n elsif($new) {\n";
$t = "\t" x ($tab / 8) . ' ' x ($tab % 8);
$t = "\t" x ($tab / 8) . ' ' x ($tab % 8);
print OUT $t,"} else {\n";
$t = "\t" x ($tab / 8) . ' ' x ($tab % 8);
$t = "\t" x ($tab / 8) . ' ' x ($tab % 8);
} elsif(/^undef\s+(\w+)/) {
print OUT $t, "undef(&$1) if defined(&$1);\n";
} elsif(/^error\s+(".*")/) {
print OUT $t, "die($1);\n";
} elsif(/^error\s+(.*)/) {
print OUT $t, "die(\"", quotemeta($1), "\");\n";
} elsif(/^warning\s+(.*)/) {
print OUT $t, "warn(\"", quotemeta($1), "\");\n";
} elsif(/^ident\s+(.*)/) {
} elsif (/^\s*(typedef\s*)?enum\s*(\s+[a-zA-Z_]\w*\s*)?/) { # { for vi
until(/\{[^}]*\}.*;/ || /;/) {
last unless defined ($next = next_line($file));
# drop "#define FOO FOO" in enums
$next =~ s/^\s*#\s*define\s+(\w+)\s+\1\s*$//;
# #defines in enums (aliases)
$next =~ s/^\s*#\s*define\s+(\w+)\s+(\w+)\s*$/$1 = $2,/;
print OUT "# $next\n" if $opt_D;
s/#\s*if.*?#\s*endif//g; # drop #ifdefs
next unless /^\s?(typedef\s?)?enum\s?([a-zA-Z_]\w*)?\s?\{(.*)\}\s?([a-zA-Z_]\w*)?\s?;/;
(my $enum_subs = $3) =~ s/\s//g;
my @enum_subs = split(/,/, $enum_subs);
foreach my $enum (@enum_subs) {
my ($enum_name, $enum_value) = $enum =~ /^([a-zA-Z_]\w*)(=.+)?$/;
$enum_val = (length($enum_value) ? $enum_value : $enum_val + 1);
"eval(\"\\n#line $eval_index $outfile\\n",
"sub $enum_name () \{ $enum_val; \}\") ",
"unless defined(\&$enum_name);\n");
"eval(\"sub $enum_name () \{ $enum_val; \}\") ",
"unless defined(\&$enum_name);\n");
} elsif (/^(?:__extension__\s+)?(?:extern|static)\s+(?:__)?inline(?:__)?\s+/
and !/;\s*$/ and !/{\s*}\s*$/)
# This is a hack to parse the inline functions in the glibc headers.
# Warning: massive kludge ahead. We suppose inline functions
# are mainly constructed like macros.
last unless defined ($next = next_line($file));
undef $_, last if $next =~ /__THROW\s*;/
or $next =~ /^(__extension__|extern|static)\b/;
print OUT "# $next\n" if $opt_D;
last if $next =~ /^}|^{.*}\s*$/;
next if not defined; # because it's only a prototype
s/\b(__extension__|extern|static|(?:__)?inline(?:__)?)\b//g;
and print OUT "# some #ifdef were dropped here -- fill in the blanks\n";
if (s/^(?:\w|\s|\*)*\s(\w+)\s*//) {
warn "name not found"; next; # shouldn't occur...
if (s/^\(([^()]*)\)\s*(\w+\s*)*//) {
for my $arg (split /,/, $1) {
if ($arg =~ /(\w+)\s*$/) {
? "my(" . (join ',', map "\$$_", @args) . ") = \@_;\n$t "
my $proto = @args ? '' : '() ';
s/\breturn\b//g; # "return" doesn't occur in macros usually...
# try to find and perlify local C variables
our @local_variables = (); # needs to be a our(): (?{...}) bug workaround
my $typelist = join '|', keys %isatype;
(?:(?:__)?const(?:__)?\s+)?
(?{ push @local_variables, $1 })
(?:(?:__)?const(?:__)?\s+)?
(?{ push @local_variables, $1 })
$new =~ s/&$_\b/\$$_/g for @local_variables;
$new =~ s/(["\\])/\\$1/g; #"]);
# now that's almost like a macro (we hope)
$Is_converted{$file} = 1;
if ($opt_e && exists($bad_file{$file})) {
unlink($Dest_dir . '/' . $outfile);
queue_includes_from($file) if $opt_a;
if ($opt_e && (scalar(keys %bad_file) > 0)) {
warn "Was unable to convert the following files:\n";
warn "\t" . join("\n\t",sort(keys %bad_file)) . "\n";
$new = '"(assembly code)"' and return if /\b__asm__\b/; # freak out.
$joined_args = join('|', keys(%curargs));
s/^\&\&// && do { $new .= " &&"; next;}; # handle && operator
s/^\&([\(a-z\)]+)/$1/i; # hack for things that take the address of
s/^(\s+)// && do {$new .= ' '; next;};
if (length $hex > 8 && !$Config{use64bitint}) {
# Croak if nv_preserves_uv_bits < 64 ?
$new .= hex(substr($hex, -8)) +
2**32 * hex(substr($hex, 0, -8));
# The above will produce "errorneus" code
# if the hex constant was e.g. inside UINT64_C
# macro, but then again, h2ph is an approximation.
s/^(-?\d+\.\d+E[-+]?\d+)[FL]?//i && do {$new .= $1; next;};
s/^(\d+)\s*[LU]*//i && do {$new .= $1; next;};
s/^("(\\"|[^"])*")// && do {$new .= $1; next;};
s/^'((\\"|[^"])*)'// && do {
# replace "sizeof(foo)" with "{foo}"
# also, remove * (C dereference operator) to avoid perl syntax
# problems. Where the %sizeof array comes from is anyone's
# guess (c2ph?), but this at least avoids fatal syntax errors.
# Behavior is undefined if sizeof() delimiters are unbalanced.
# This code was modified to able to handle constructs like this:
# sizeof(*(p)), which appear in the HP-UX 10.01 header files.
my $lvl = 1; # already saw one open paren
# tack { on the front, and skip it in the loop
# find balanced closing paren
while ($index <= length($_) && $lvl > 0) {
$lvl++ if substr($_, $index, 1) eq "(";
$lvl-- if substr($_, $index, 1) eq ")";
# tack } on the end, replacing )
substr($_, $index - 1, 1) = "}";
# remove pesky * operators within the sizeof argument
substr($_, 0, $index - 1) =~ s/\*//g;
/\(([\w\s]+)[\*\s]*\)\s*[\w\(]/ && do {
foreach (split /\s+/, $1) { # Make sure all the words are types,
unless($isatype{$_} or $_ eq 'struct' or $_ eq 'union'){
s/\([\w\s]+[\*\s]*\)// && next; # then eliminate them.
# struct/union member, including arrays:
s/^([_A-Z]\w*(\[[^\]]+\])?((\.|->)[_A-Z]\w*(\[[^\]]+\])?)+)//i && do {
$id =~ s/(\.|(->))([^\.\-]*)/->\{$3\}/g;
$id =~ s/\b([^\$])($joined_args)/$1\$$2/g if length($joined_args);
while($id =~ /\[\s*([^\$\&\d\]]+)\]/) {
if(exists($curargs{$index})) {
$id =~ s/\[\s*([^\$\&\d\]]+)\]/[$index]/;
s/^([_a-zA-Z]\w*)// && do {
if ($id eq 'struct' || $id eq 'union') {
} elsif ($id =~ /^((un)?signed)|(long)|(short)$/) {
while (s/^\s+(\w+)//) { $id .= ' ' . $1; }
$new .= '->' if /^[\[\{]/;
} elsif ($id eq 'defined') {
s/^\s*\((\w),/("$1",/ if $id =~ /^_IO[WR]*$/i; # cheat
} elsif ($isatype{$id}) {
} elsif ($new =~ /\(\s*$/ && /^[\s*]*\)/) {
if ($inif && $new !~ /defined\s*\($/) {
$new .= '(defined(&' . $id . ') ? &' . $id . ' : 0)';
s/^(.)// && do { if ($1 ne '#') { $new .= $1; } next;};
my $pre_sub_tri_graphs = 1;
READ: while (not eof IN) {
if ($pre_sub_tri_graphs) {
# Preprocess all tri-graphs
# including things stuck in quoted string constants.
$in =~ s/\?\?=/#/g; # | ??=| #|
$in =~ s/\?\?\!/|/g; # | ??!| ||
$in =~ s/\?\?'/^/g; # | ??'| ^|
$in =~ s/\?\?\(/[/g; # | ??(| [|
$in =~ s/\?\?\)/]/g; # | ??)| ]|
$in =~ s/\?\?\-/~/g; # | ??-| ~|
$in =~ s/\?\?\//\\/g; # | ??/| \|
$in =~ s/\?\?</{/g; # | ??<| {|
$in =~ s/\?\?>/}/g; # | ??>| }|
if ($in =~ /^\#ifdef __LANGUAGE_PASCAL__/) {
# Tru64 disassembler.h evilness: mixed C and Pascal.
if ($in =~ /^extern inline / && # Inlined assembler.
$^O eq 'linux' && $file =~ m!(?:^|/)asm/[^/]+\.h$!) {
if ($in =~ s/\\$//) { # \-newline
} elsif ($in =~ s/^([^"'\\\/]+)//) { # Passthrough
} elsif ($in =~ s/^(\\.)//) { # \...
} elsif ($in =~ /^'/) { # '...
if ($in =~ s/^('(\\.|[^'\\])*')//) {
} elsif ($in =~ /^"/) { # "...
if ($in =~ s/^("(\\.|[^"\\])*")//) {
} elsif ($in =~ s/^\/\/.*//) { # //...
} elsif ($in =~ m/^\/\*/) { # /*...
# C comment removal adapted from perlfaq6:
if ($in =~ s/^\/\*[^*]*\*+([^\/*][^*]*\*+)*\///) {
} else { # Incomplete /* */
} elsif ($in =~ s/^(\/)//) { # /...
} elsif ($in =~ s/^([^\'\"\\\/]+)//) {
} elsif ($^O eq 'linux' &&
$file =~ m!(?:^|/)linux/byteorder/pdp_endian\.h$! &&
$out =~ s!I DON$!I_DO_NOT_KNOW!;
warn "Cannot parse $file:\n$in\n";
die "Cannot parse:\n$in\n";
last READ if $out =~ /\S/;
# Handle recursive subdirectories without getting a grotesquely big stack.
# Could this be implemented using File::Find?
if ($file eq '-' or -f $file or -l $file) {
print STDERR "Skipping directory `$file'\n";
print STDERR "Skipping `$file': not a file or directory\n";
# Put all the files in $directory into @ARGV for processing.
next if ($_ eq '.' or $_ eq '..');
# expand_glob() is going to be called until $ARGV[0] isn't a
# directory; so push directories, and unshift everything else.
if (-d "$directory/$_") { push @ARGV, "$directory/$_" }
else { unshift @ARGV, "$directory/$_" }
# Given $file, a symbolic link to a directory in the C include directory,
# make an equivalent symbolic link in $Dest_dir, if we can figure out how.
# Otherwise, just duplicate the file or directory.
my $target = eval 'readlink($dirlink)';
if ($target =~ m:^\.\./: or $target =~ m:^/:) {
# The target of a parent or absolute link could leave the $Dest_dir
# hierarchy, so let's put all of the contents of $dirlink (actually,
# the contents of $target) into @ARGV; as a side effect down the
# line, $dirlink will get created as an _actual_ directory.
if (-l "$Dest_dir/$dirlink") {
unlink "$Dest_dir/$dirlink" or
print STDERR "Could not remove link $Dest_dir/$dirlink: $!\n";
if (eval 'symlink($target, "$Dest_dir/$dirlink")') {
print "Linking $target -> $Dest_dir/$dirlink\n";
# Make sure that the link _links_ to something:
if (! -e "$Dest_dir/$target") {
mkpath("$Dest_dir/$target", 0755) or
print STDERR "Could not create $Dest_dir/$target/\n";
print STDERR "Could not symlink $target -> $Dest_dir/$dirlink: $!\n";
# Push all #included files in $file onto our stack, except for STDIN
# and files we've already processed.
return if ($file eq "-");
open HEADER, $file or return;
while (defined($line = <HEADER>)) {
while (/\\$/) { # Handle continuation lines
if ($line =~ /^#\s*include\s+<(.*?)>/) {
push(@ARGV, $1) unless $Is_converted{$1};
# Determine include directories; $Config{usrinc} should be enough for (all
# non-GCC?) C compilers, but gcc uses an additional include directory.
my $from_gcc = `LC_ALL=C $Config{cc} -v 2>&1`;
if( !( $from_gcc =~ s:^Reading specs from (.*?)/specs\b.*:$1/include:s ) )
$from_gcc = `LC_ALL=C $Config{cc} -print-search-dirs 2>&1`;
if ( !($from_gcc =~ s/^install:\s*([^\s]+[^\s\/])([\s\/]*).*$/$1\/include/s) )
length($from_gcc) ? ($from_gcc, $Config{usrinc}) : ($Config{usrinc});
# Create "_h2ph_pre.ph", if it doesn't exist or was built by a different
sub build_preamble_if_necessary
# Increment $VERSION every time this function is modified:
my $preamble = "$Dest_dir/_h2ph_pre.ph";
# Can we skip building the preamble file?
# Extract version number from first line of preamble:
open PREAMBLE, $preamble or die "Cannot open $preamble: $!";
close PREAMBLE or die "Cannot close $preamble: $!";
# Don't build preamble if a compatible preamble exists:
return if $1 == $VERSION;
my (%define) = _extract_cc_defines();
open PREAMBLE, ">$preamble" or die "Cannot open $preamble: $!";
print PREAMBLE "# This file was created by h2ph version $VERSION\n";
foreach (sort keys %define) {
print PREAMBLE "# $_=$define{$_}\n";
if ($define{$_} =~ /^(\d+)U?L{0,2}$/i) {
"unless (defined &$_) { sub $_() { $1 } }\n\n";
} elsif ($define{$_} =~ /^\w+$/) {
"unless (defined &$_) { sub $_() { &$define{$_} } }\n\n";
"unless (defined &$_) { sub $_() { \"",
quotemeta($define{$_}), "\" } }\n\n";
close PREAMBLE or die "Cannot close $preamble: $!";
# %Config contains information on macros that are pre-defined by the
# system's compiler. We need this information to make the .ph files
# function with perl as the .h files do with cc.
my $allsymbols = join " ",
@Config{'ccsymbols', 'cppsymbols', 'cppccsymbols'};
# Split compiler pre-definitions into `key=value' pairs:
foreach (split /\s+/, $allsymbols) {
/(.+?)=(.+)/ and $define{$1} = $2;
print STDERR "$_: $1 -> $2\n";
##############################################################################
h2ph - convert .h C header files to .ph Perl header files
B<h2ph [-d destination directory] [-r | -a] [-l] [headerfiles]>
converts any C header files specified to the corresponding Perl header file
It is most easily run while in /usr/include:
cd /usr/include; h2ph * sys/*
cd /usr/include; h2ph * sys/* arpa/* netinet/*
cd /usr/include; h2ph -r -l .
The output files are placed in the hierarchy rooted at Perl's
architecture dependent library directory. You can specify a different
hierarchy with a B<-d> switch.
If run with no arguments, filters standard input to standard output.
Put the resulting B<.ph> files beneath B<destination_dir>, instead of
beneath the default Perl library location (C<$Config{'installsitsearch'}>).
Run recursively; if any of B<headerfiles> are directories, then run I<h2ph>
on all files in those directories (and their subdirectories, etc.). B<-r>
and B<-a> are mutually exclusive.
Run automagically; convert B<headerfiles>, as well as any B<.h> files
which they include. This option will search for B<.h> files in all
directories which your C compiler ordinarily uses. B<-a> and B<-r> are
Symbolic links will be replicated in the destination directory. If B<-l>
is not specified, then links are skipped over.
Put ``hints'' in the .ph files which will help in locating problems with
I<h2ph>. In those cases when you B<require> a B<.ph> file containing syntax
errors, instead of the cryptic
[ some error condition ] at (eval mmm) line nnn
you will see the slightly more helpful
[ some error condition ] at filename.ph line nnn
However, the B<.ph> files almost double in size when built using B<-h>.
Include the code from the B<.h> file as a comment in the B<.ph> file.
This is primarily used for debugging I<h2ph>.
``Quiet'' mode; don't print out the names of the files being converted.
No environment variables are used.
The usual warnings if it can't read or write the files involved.
Doesn't construct the %sizeof array for you.
It doesn't handle all C constructs, but it does attempt to isolate
definitions inside evals so that you can get at the definitions
It's only intended as a rough tool.
You may need to dicker with the files produced.
You have to run this program by hand; it's not run as part of the Perl
Doesn't handle complicated expressions built piecemeal, a la:
Doesn't necessarily locate all of your C compiler's internally-defined