case $CONFIG in '') if test ! -f config.sh; then ln ../config.sh . || \ ln ../../config.sh . || \ ln ../../../config.sh . || \ (echo "Can't find config.sh."; exit 1) fi 2>/dev/null . ./config.sh ;; esac : This forces SH files to create target in same directory as SH file. : This is so that make depend always knows where to find SH derivatives. case "$0" in */*) cd `expr X$0 : 'X\(.*\)/'` ;; esac echo "Extracting h2ph (with variable substitutions)" : This section of the file will have variable substitutions done on it. : Move anything that needs config subs from !NO!SUBS! section to !GROK!THIS!. : Protect any dollar signs and backticks that you do not want interpreted : by putting a backslash in front. You may delete these comments. rm -f h2ph $spitshell >h2ph <>h2ph <<'!NO!SUBS!' chdir '/usr/include' || die "Can't cd /usr/include"; @isatype = split(' ',<-"); } else { ($outfile = $file) =~ s/\.h$/.ph/ || next; print "$file -> $outfile\n"; if ($file =~ m|^(.*)/|) { $dir = $1; if (!-d "$perlincl/$dir") { mkdir("$perlincl/$dir",0777); } } open(IN,"$file") || ((warn "Can't open $file: $!\n"),next); open(OUT,">$perlincl/$outfile") || die "Can't create $outfile: $!\n"; } while () { chop; while (/\\$/) { chop; $_ .= ; chop; } if (s:/\*:\200:g) { s:\*/:\201:g; s/\200[^\201]*\201//g; # delete single line comments if (s/\200.*//) { # begin multi-line comment? $_ .= '/*'; $_ .= ; redo; } } if (s/^#\s*//) { if (s/^define\s+(\w+)//) { $name = $1; $new = ''; s/\s+$//; if (s/^\(([\w,\s]*)\)//) { $args = $1; if ($args ne '') { foreach $arg (split(/,\s*/,$args)) { $arg =~ s/^\s*([^\s].*[^\s])\s*$/$1/; $curargs{$arg} = 1; } $args =~ s/\b(\w)/\$$1/g; $args = "local($args) = \@_;\n$t "; } s/^\s+//; do expr(); $new =~ s/(["\\])/\\$1/g; if ($t ne '') { $new =~ s/(['\\])/\\$1/g; print OUT $t, "eval 'sub $name {\n$t ${args}eval \"$new\";\n$t}';\n"; } else { print OUT "sub $name {\n ${args}eval \"$new\";\n}\n"; } %curargs = (); } else { s/^\s+//; do expr(); $new = 1 if $new eq ''; if ($t ne '') { $new =~ s/(['\\])/\\$1/g; print OUT $t,"eval 'sub $name {",$new,";}';\n"; } else { print OUT $t,"sub $name {",$new,";}\n"; } } } elsif (/^include\s+<(.*)>/) { ($incl = $1) =~ s/\.h$/.ph/; print OUT $t,"require '$incl';\n"; } elsif (/^ifdef\s+(\w+)/) { print OUT $t,"if (defined &$1) {\n"; $tab += 4; $t = "\t" x ($tab / 8) . ' ' x ($tab % 8); } elsif (/^ifndef\s+(\w+)/) { print OUT $t,"if (!defined &$1) {\n"; $tab += 4; $t = "\t" x ($tab / 8) . ' ' x ($tab % 8); } elsif (s/^if\s+//) { $new = ''; $inif = 1; do expr(); $inif = 0; print OUT $t,"if ($new) {\n"; $tab += 4; $t = "\t" x ($tab / 8) . ' ' x ($tab % 8); } elsif (s/^elif\s+//) { $new = ''; $inif = 1; do expr(); $inif = 0; $tab -= 4; $t = "\t" x ($tab / 8) . ' ' x ($tab % 8); print OUT $t,"}\n${t}elsif ($new) {\n"; $tab += 4; $t = "\t" x ($tab / 8) . ' ' x ($tab % 8); } elsif (/^else/) { $tab -= 4; $t = "\t" x ($tab / 8) . ' ' x ($tab % 8); print OUT $t,"}\n${t}else {\n"; $tab += 4; $t = "\t" x ($tab / 8) . ' ' x ($tab % 8); } elsif (/^endif/) { $tab -= 4; $t = "\t" x ($tab / 8) . ' ' x ($tab % 8); print OUT $t,"}\n"; } } } print OUT "1;\n"; } sub expr { while ($_ ne '') { s/^(\s+)// && do {$new .= ' '; next;}; s/^(0x[0-9a-fA-F]+)// && do {$new .= $1; next;}; s/^(\d+)// && do {$new .= $1; next;}; s/^("(\\"|[^"])*")// && do {$new .= $1; next;}; s/^'((\\"|[^"])*)'// && do { if ($curargs{$1}) { $new .= "ord('\$$1')"; } else { $new .= "ord('$1')"; } next; }; s/^sizeof\s*\(([^)]+)\)/{$1}/ && do { $new .= '$sizeof'; next; }; s/^([_a-zA-Z]\w*)// && do { $id = $1; if ($id eq 'struct') { s/^\s+(\w+)//; $id .= ' ' . $1; $isatype{$id} = 1; } elsif ($id eq 'unsigned') { s/^\s+(\w+)//; $id .= ' ' . $1; $isatype{$id} = 1; } if ($curargs{$id}) { $new .= '$' . $id; } elsif ($id eq 'defined') { $new .= 'defined'; } elsif (/^\(/) { s/^\((\w),/("$1",/ if $id =~ /^_IO[WR]*$/i; # cheat $new .= " &$id"; } elsif ($isatype{$id}) { if ($new =~ /{\s*$/) { $new .= "'$id'"; } elsif ($new =~ /\(\s*$/ && /^[\s*]*\)/) { $new =~ s/\(\s*$//; s/^[\s*]*\)//; } else { $new .= $id; } } else { if ($inif && $new !~ /defined\($/) { $new .= '(defined(&' . $id . ') ? &' . $id . ' : 0)'; } else { $new .= ' &' . $id; } } next; }; s/^(.)// && do {$new .= $1; next;}; } } ############################################################################## # These next few lines are legal in both Perl and nroff. .00 ; # finish .ig 'di \" finish diversion--previous line must be blank .nr nl 0-1 \" fake up transition to first page again .nr % 0 \" start at page 1 '; __END__ ############# From here on it's a standard manual page ############ .TH H2PH 1 "August 8, 1990" .AT 3 .SH NAME h2ph \- convert .h C header files to .ph Perl header files .SH SYNOPSIS .B h2ph [headerfiles] .SH DESCRIPTION .I h2ph converts any C header files specified to the corresponding Perl header file format. It is most easily run while in /usr/include: .nf cd /usr/include; h2ph * sys/* .fi If run with no arguments, filters standard input to standard output. .SH ENVIRONMENT No environment variables are used. .SH FILES /usr/include/*.h .br /usr/include/sys/*.h .br etc. .SH AUTHOR Larry Wall .SH "SEE ALSO" perl(1) .SH DIAGNOSTICS The usual warnings if it can't read or write the files involved. .SH BUGS Doesn't construct the %sizeof array for you. .PP It doesn't handle all C constructs, but it does attempt to isolate definitions inside evals so that you can get at the definitions that it can translate. .PP It's only intended as a rough tool. You may need to dicker with the files produced. .ex !NO!SUBS! chmod 755 h2ph $eunicefix h2ph rm -f h2ph.man ln h2ph h2ph.man