perl/regen/regcomp.pl

#!/usr/bin/perl -w
#
#
# Regenerate (overwriting only if changed):
#
#    pod/perldebguts.pod
#    regnodes.h
#
# from information stored in
#
#    regcomp.sym
#    regexp.h
#
# pod/perldebguts.pod is not completely regenerated.  Only the table of
# regexp nodes is replaced; other parts remain unchanged.
#
# Accepts the standard regen_lib -q and -v args.
#
# This script is normally invoked from regen.pl.

BEGIN {
    # Get function prototypes
    require 'regen/regen_lib.pl';
}
use strict;

# NOTE I don't think anyone actually knows what all of these properties mean,
# and I suspect some of them are outright unused. This is a first attempt to
# clean up the generation so maybe one day we can move to something more self
# documenting. (One might argue that an array of hashes of properties would
# be easier to use.)
#
# Why we use the term regnode and nodes, and not say, opcodes, I am not sure.

# General thoughts:
# 1. We use a single continuum to represent both opcodes and states,
#    and in regexec.c we switch on the combined set.
# 2. Opcodes have more information associated to them, states are simpler,
#    basically just an identifier/number that can be used to switch within
#    the state machine.
# 3. Some opcode are order dependent.
# 4. Output files often use "tricks" to reduce diff effects. Some of what
#    we do below is more clumsy looking than it could be because of this.

# Op/state properties:
#
# Property      In      Descr
# ----------------------------------------------------------------------------
# name          Both    Name of op/state
# id            Both    integer value for this opcode/state
# optype        Both    Either 'op' or 'state'
# line_num          Both    line_num number of the input file for this item.
# type          Op      Type of node (aka regkind)
# code          Op      what code is associated with this node (???)
# args          Op      what type of args the node has (which regnode struct)
# flags         Op      (???)
# longj         Op      Whether this node is a longjump
# comment       Both    Comment about node, if any
# pod_comment   Both    Special comments for pod output (preceding lines in def)

# Global State
my @all;    # all opcodes/state
my %all;    # hash of all opcode/state names

my @ops;    # array of just opcodes
my @states; # array of just states

my $longest_name_length= 0; # track lengths of names for nicer reports
my (%type_alias);           # map the type (??)

# register a newly constructed node into our state tables.
# ensures that we have no name collisions (on name anyway),
# and issues the "id" for the node.
sub register_node {
    my ($node)= @_;

    if ( $all{ $node->{name} } ) {
        die "Duplicate item '$node->{name}' in regcomp.sym line $node->{line_num} "
            . "previously defined on line $all{ $node->{name} }{line_num}\n";
    } elsif (!$node->{optype}) {
        die "must have an optype in node ", Dumper($node);
    } elsif ($node->{optype} eq "op") {
        push @ops, $node;
    } elsif ($node->{optype} eq "state") {
        push @states, $node;
    } else {
        die "Uknown optype '$node->{optype}' in ", Dumper($node);
    }
    $node->{id}= 0 + @all;
    push @all, $node;
    $all{ $node->{name} }= $node;
}

# Parse and add an opcode definition to the global state.
# An opcode definition looks like this:
#
#                             +- args
#                             | +- flags
#                             | | +- longjmp
# Name        Type       code | | | ; comment
# --------------------------------------------------------------------------
# IFMATCH     BRANCHJ,    off 1 . 2 ; Succeeds if the following matches.
# UNLESSM     BRANCHJ,    off 1 . 2 ; Fails if the following matches.
# SUSPEND     BRANCHJ,    off 1 V 1 ; "Independent" sub-RE.
# IFTHEN      BRANCHJ,    off 1 V 1 ; Switch, should be preceded by switcher.
# GROUPP      GROUPP,     num 1     ; Whether the group matched.
#
# Not every opcode definition has all of these. We should maybe make this
# nicer/easier to read in the future. Also note that the above is tab
# sensitive.

sub parse_opcode_def {
    my ( $text, $line_num, $pod_comment )= @_;
    my $node= {
        line_num    => $line_num,
        pod_comment => $pod_comment,
        optype      => "op",
    };

    # first split the line into three, the initial NAME, a middle part
    # that we call "desc" which contains various (not well documented) things,
    # and a comment section.
    @{$node}{qw(name desc comment)}= /^(\S+)\s+([^\t]+?)\s*;\s*(.*)/
        or die "Failed to match $_";

    # the content of the "desc" field from the first step is extracted here:
    @{$node}{qw(type code args flags longj)}= split /[,\s]\s*/, $node->{desc};

    $node->{$_} //= "" for qw(type code args flags longj);

    register_node($node); # has to be before the type_alias code below

    if ( !$all{ $node->{type} } and !$type_alias{ $node->{type} } ) {

        #warn "Regop type '$node->{type}' from regcomp.sym line $line_num"
        #     ." is not an existing regop, and will be aliased to $node->{name}\n"
        #    if -t STDERR;
        $type_alias{ $node->{type} }= $node->{name};
    }

    $longest_name_length= length $node->{name}
        if length $node->{name} > $longest_name_length;
}

# parse out a state definition and add the resulting data
# into the global state. may create multiple new states from
# a single definition (this is part of the point).
# Format for states:
# REGOP \t typelist [ \t typelist]
# typelist= namelist
#         = namelist:FAIL
#         = name:count
# Eg:
# WHILEM          A_pre,A_min,A_max,B_min,B_max:FAIL
# BRANCH          next:FAIL
# CURLYM          A,B:FAIL
#
# The CURLYM definition would create the states:
# CURLYM_A, CURLYM_A_fail, CURLYM_B, CURLYM_B_fail
sub parse_state_def {
    my ( $text, $line_num, $pod_comment )= @_;
    my ( $type, @lists )= split /\s+/, $text;
    die "No list? $type" if !@lists;
    foreach my $list (@lists) {
        my ( $names, $special )= split /:/, $list, 2;
        $special ||= "";
        foreach my $name ( split /,/, $names ) {
            my $real=
                $name eq 'resume'
                ? "resume_$type"
                : "${type}_$name";
            my @suffix;
            if ( !$special ) {
                @suffix= ("");
            }
            elsif ( $special =~ /\d/ ) {
                @suffix= ( 1 .. $special );
            }
            elsif ( $special eq 'FAIL' ) {
                @suffix= ( "", "_fail" );
            }
            else {
                die "unknown :type ':$special'";
            }
            foreach my $suffix (@suffix) {
                my $node= {
                    name        => "$real$suffix",
                    optype      => "state",
                    type        => $type || "",
                    comment     => "state for $type",
                    line_num    => $line_num,
                };
                register_node($node);
            }
        }
    }
}

sub process_flags {
    my ( $flag, $varname, $comment )= @_;
    $comment= '' unless defined $comment;

    my @selected;
    my $bitmap= '';
    for my $node (@ops) {
        my $set= $node->{flags} && $node->{flags} eq $flag ? 1 : 0;

        # Whilst I could do this with vec, I'd prefer to do longhand the arithmetic
        # ops in the C code.
        my $current= do {
            no warnings;
            ord substr $bitmap, ( $node->{id} >> 3 );
        };
        substr( $bitmap, ( $node->{id} >> 3 ), 1 )=
            chr( $current | ( $set << ( $node->{id} & 7 ) ) );

        push @selected, $node->{name} if $set;
    }
    my $out_string= join ', ', @selected, 0;
    $out_string =~ s/(.{1,70},) /$1\n    /g;

    my $out_mask= join ', ', map { sprintf "0x%02X", ord $_ } split '', $bitmap;

    return $comment . <<"EOP";
#define REGNODE_\U$varname\E(node) (PL_${varname}_bitmask[(node) >> 3] & (1 << ((node) & 7)))

#ifndef DOINIT
EXTCONST U8 PL_${varname}\[] __attribute__deprecated__;
#else
EXTCONST U8 PL_${varname}\[] __attribute__deprecated__ = {
    $out_string
};
#endif /* DOINIT */

#ifndef DOINIT
EXTCONST U8 PL_${varname}_bitmask[];
#else
EXTCONST U8 PL_${varname}_bitmask[] = {
    $out_mask
};
#endif /* DOINIT */
EOP
}

sub read_definition {
    my ( $file )= @_;
    my ( $seen_sep, $pod_comment )= "";
    open my $in_fh, "<", $file
        or die "Failed to open '$file' for reading: $!";
    while (<$in_fh>) {

        # Special pod comments
        if (/^#\* ?/) { $pod_comment .= "# $'"; }

        # Truly blank lines possibly surrounding pod comments
        elsif (/^\s*$/) { $pod_comment .= "\n" }

        next if /\A\s*#/ || /\A\s*\z/;

        s/\s*\z//;
        if (/^-+\s*$/) {
            $seen_sep= 1;
            next;
        }

        if ($seen_sep) {
            parse_state_def( $_, $., $pod_comment );
        }
        else {
            parse_opcode_def( $_, $., $pod_comment );
        }
        $pod_comment= "";
    }
    close $in_fh;
    die "Too many regexp/state opcodes! Maximum is 256, but there are ", 0 + @all,
        " in file!"
        if @all > 256;
}

# use fixed width to keep the diffs between regcomp.pl recompiles
# as small as possible.
my ( $width, $rwidth, $twidth )= ( 22, 12, 9 );

sub print_state_defs {
    my ($out)= @_;
    printf $out <<EOP,
/* Regops and State definitions */

#define %*s\t%d
#define %*s\t%d

EOP
        -$width,
        REGNODE_MAX => $#ops,
        -$width, REGMATCH_STATE_MAX => $#all;

    my %rev_type_alias= reverse %type_alias;
    for my $node (@ops) {
        printf $out "#define\t%*s\t%d\t/* %#04x %s */\n",
            -$width, $node->{name}, $node->{id}, $node->{id}, $node->{comment};
        if ( defined( my $alias= $rev_type_alias{ $node->{name} } ) ) {
            printf $out "#define\t%*s\t%d\t/* %#04x %s */\n",
                -$width, $alias, $node->{id}, $node->{id}, "type alias";
        }
    }

    print $out "\t/* ------------ States ------------- */\n";
    for my $node (@states) {
        printf $out "#define\t%*s\t(REGNODE_MAX + %d)\t/* %s */\n",
            -$width, $node->{name}, $node->{id} - $#ops, $node->{comment};
    }
}

sub print_regkind {
    my ($out)= @_;
    print $out <<EOP;

/* PL_regkind[] What type of regop or state is this. */

#ifndef DOINIT
EXTCONST U8 PL_regkind[];
#else
EXTCONST U8 PL_regkind[] = {
EOP
    use Data::Dumper;
    foreach my $node (@all) {
        print Dumper($node) if !defined $node->{type} or !defined( $node->{name} );
        printf $out "\t%*s\t/* %*s */\n",
            -1 - $twidth, "$node->{type},", -$width, $node->{name};
        print $out "\t/* ------------ States ------------- */\n"
            if $node->{id} == $#ops and $node->{id} != $#all;
    }

    print $out <<EOP;
};
#endif
EOP
}

sub wrap_ifdef_print {
    my $out= shift;
    my $token= shift;
    print $out <<EOP;

#ifdef $token
EOP
    $_->($out) for @_;
    print $out <<EOP;
#endif /* $token */

EOP
}

sub print_regarglen {
    my ($out)= @_;
    print $out <<EOP;

/* regarglen[] - How large is the argument part of the node (in regnodes) */

static const U8 regarglen[] = {
EOP

    foreach my $node (@ops) {
        my $size= 0;
        $size= "EXTRA_SIZE(struct regnode_$node->{args})" if $node->{args};

        printf $out "\t%*s\t/* %*s */\n", -37, "$size,", -$rwidth, $node->{name};
    }

    print $out <<EOP;
};
EOP
}

sub print_reg_off_by_arg {
    my ($out)= @_;
    print $out <<EOP;

/* reg_off_by_arg[] - Which argument holds the offset to the next node */

static const char reg_off_by_arg[] = {
EOP

    foreach my $node (@ops) {
        my $size= $node->{longj} || 0;

        printf $out "\t%d,\t/* %*s */\n", $size, -$rwidth, $node->{name};
    }

    print $out <<EOP;
};

EOP
}

sub print_reg_name {
    my ($out)= @_;
    print $out <<EOP;

/* reg_name[] - Opcode/state names in string form, for debugging */

#ifndef DOINIT
EXTCONST char * PL_reg_name[];
#else
EXTCONST char * const PL_reg_name[] = {
EOP

    my $ofs= 0;
    my $sym= "";
    foreach my $node (@all) {
        my $size= $node->{longj} || 0;

        printf $out "\t%*s\t/* $sym%#04x */\n",
            -3 - $width, qq("$node->{name}",), $node->{id} - $ofs;
        if ( $node->{id} == $#ops and @ops != @all ) {
            print $out "\t/* ------------ States ------------- */\n";
            $ofs= $#ops;
            $sym= 'REGNODE_MAX +';
        }
    }

    print $out <<EOP;
};
#endif /* DOINIT */

EOP
}

sub print_reg_extflags_name {
    my ($out)= @_;
    print $out <<EOP;
/* PL_reg_extflags_name[] - Opcode/state names in string form, for debugging */

#ifndef DOINIT
EXTCONST char * PL_reg_extflags_name[];
#else
EXTCONST char * const PL_reg_extflags_name[] = {
EOP

    my %rxfv;
    my %definitions;    # Remember what the symbol definitions are
    my $val= 0;
    my %reverse;
    my $REG_EXTFLAGS_NAME_SIZE= 0;
    foreach my $file ( "op_reg_common.h", "regexp.h" ) {
        open my $in_fh, "<", $file or die "Can't read '$file': $!";
        while (<$in_fh>) {

            # optional leading '_'.  Return symbol in $1, and strip it from
            # comment of line
            if (s/^ \# \s* define \s+ ( _? RXf_ \w+ ) \s+ //xi) {
                chomp;
                my $define= $1;
                my $orig= $_;
                s{ /\* .*? \*/ }{ }x;    # Replace comments by a blank

                # Replace any prior defined symbols by their values
                foreach my $key ( keys %definitions ) {
                    s/\b$key\b/$definitions{$key}/g;
                }

                # Remove the U suffix from unsigned int literals
                s/\b([0-9]+)U\b/$1/g;

                my $newval= eval $_;     # Get numeric definition

                $definitions{$define}= $newval;

                next unless $_ =~ /<</;    # Bit defines use left shift
                if ( $val & $newval ) {
                    my @names= ( $define, $reverse{$newval} );
                    s/PMf_// for @names;
                    if ( $names[0] ne $names[1] ) {
                        die sprintf
                            "ERROR: both $define and $reverse{$newval} use 0x%08X (%s:%s)",
                            $newval, $orig, $_;
                    }
                    next;
                }
                $val |= $newval;
                $rxfv{$define}= $newval;
                $reverse{$newval}= $define;
            }
        }
    }
    my %vrxf= reverse %rxfv;
    printf $out "\t/* Bits in extflags defined: %s */\n", unpack 'B*', pack 'N',
        $val;
    my %multibits;
    for ( 0 .. 31 ) {
        my $power_of_2= 2**$_;
        my $n= $vrxf{$power_of_2};
        my $extra= "";
        if ( !$n ) {

            # Here, there was no name that matched exactly the bit.  It could be
            # either that it is unused, or the name matches multiple bits.
            if ( !( $val & $power_of_2 ) ) {
                $n= "UNUSED_BIT_$_";
            }
            else {

                # Here, must be because it matches multiple bits.  Look through
                # all possibilities until find one that matches this one.  Use
                # that name, and all the bits it matches
                foreach my $name ( keys %rxfv ) {
                    if ( $rxfv{$name} & $power_of_2 ) {
                        $n= $name . ( $multibits{$name}++ );
                        $extra= sprintf qq{ : "%s" - 0x%08x}, $name,
                            $rxfv{$name}
                            if $power_of_2 != $rxfv{$name};
                        last;
                    }
                }
            }
        }
        s/\bRXf_(PMf_)?// for $n, $extra;
        printf $out qq(\t%-20s/* 0x%08x%s */\n), qq("$n",), $power_of_2, $extra;
        $REG_EXTFLAGS_NAME_SIZE++;
    }

    print $out <<EOP;
};
#endif /* DOINIT */

#ifdef DEBUGGING
#  define REG_EXTFLAGS_NAME_SIZE $REG_EXTFLAGS_NAME_SIZE
#endif
EOP

}

sub print_reg_intflags_name {
    my ($out)= @_;
    print $out <<EOP;

/* PL_reg_intflags_name[] - Opcode/state names in string form, for debugging */

#ifndef DOINIT
EXTCONST char * PL_reg_intflags_name[];
#else
EXTCONST char * const PL_reg_intflags_name[] = {
EOP

    my %rxfv;
    my %definitions;    # Remember what the symbol definitions are
    my $val= 0;
    my %reverse;
    my $REG_INTFLAGS_NAME_SIZE= 0;
    foreach my $file ("regcomp.h") {
        open my $fh, "<", $file or die "Can't read $file: $!";
        while (<$fh>) {

            # optional leading '_'.  Return symbol in $1, and strip it from
            # comment of line
            if (
                m/^ \# \s* define \s+ ( PREGf_ ( \w+ ) ) \s+ 0x([0-9a-f]+)(?:\s*\/\*(.*)\*\/)?/xi
                )
            {
                chomp;
                my $define= $1;
                my $abbr= $2;
                my $hex= $3;
                my $comment= $4;
                my $val= hex($hex);
                $comment= $comment ? " - $comment" : "";

                printf $out qq(\t%-30s/* 0x%08x - %s%s */\n), qq("$abbr",),
                    $val, $define, $comment;
                $REG_INTFLAGS_NAME_SIZE++;
            }
        }
    }

    print $out <<EOP;
};
#endif /* DOINIT */

EOP
    print $out <<EOQ;
#ifdef DEBUGGING
#  define REG_INTFLAGS_NAME_SIZE $REG_INTFLAGS_NAME_SIZE
#endif

EOQ
}

sub print_process_flags {
    my ($out)= @_;

    print $out process_flags( 'V', 'varies', <<'EOC');
/* The following have no fixed length. U8 so we can do strchr() on it. */
EOC

    print $out process_flags( 'S', 'simple', <<'EOC');

/* The following always have a length of 1. U8 we can do strchr() on it. */
/* (Note that length 1 means "one character" under UTF8, not "one octet".) */
EOC

}

sub do_perldebguts {
    my $guts= open_new( 'pod/perldebguts.pod', '>' );

    my $node;
    my $code;
    my $name_fmt= '<' x  ( $longest_name_length - 1 );
    my $descr_fmt= '<' x ( 58 - $longest_name_length );
    eval <<EOD or die $@;
format GuTS =
 ^*~~
 \$node->{pod_comment}
 ^$name_fmt ^<<<<<<<<< ^$descr_fmt~~
 \$node->{name}, \$code, \$node->{comment}//''
.
1;
EOD

    my $old_fh= select($guts);
    $~= "GuTS";

    open my $oldguts, "pod/perldebguts.pod"
        or die "$0 cannot open pod/perldebguts.pod for reading: $!";
    while (<$oldguts>) {
        print;
        last if /=for regcomp.pl begin/;
    }

    print <<'END_OF_DESCR';

 # TYPE arg-description [num-args] [longjump-len] DESCRIPTION
END_OF_DESCR
    for my $n (@ops) {
        $node= $n;
        $code= "$node->{code} " . ( $node->{args} || "" );
        $code .= " $node->{longj}" if $node->{longj};
        if ( $node->{pod_comment} ||= "" ) {

            # Trim multiple blanks
            $node->{pod_comment} =~ s/^\n\n+/\n/;
            $node->{pod_comment} =~ s/\n\n+$/\n\n/;
        }
        write;
    }
    print "\n";

    while (<$oldguts>) {
        last if /=for regcomp.pl end/;
    }
    do { print } while <$oldguts>;
    select $old_fh;
    close_and_rename($guts);
}

read_definition("regcomp.sym");
my $out= open_new( 'regnodes.h', '>',
    { by => 'regen/regcomp.pl', from => 'regcomp.sym' } );
print_state_defs($out);
print_regkind($out);
wrap_ifdef_print(
    $out,
    "REG_COMP_C",
    \&print_regarglen,
    \&print_reg_off_by_arg
);
print_reg_name($out);
print_reg_extflags_name($out);
print_reg_intflags_name($out);
print_process_flags($out);
read_only_bottom_close_and_rename($out);

do_perldebguts();