muscle: factor the field retrieval

[bison.git] / etc / bench.pl.in

diff --git a/etc/bench.pl.in b/etc/bench.pl.in
index 22328cfa91381893cc4820f7c799281d778c7710..30d0a996e611a3b73acb871c9e45132686b44879 100755 (executable)
--- a/etc/bench.pl.in
+++ b/etc/bench.pl.in
@@ -1,6 +1,6 @@
 #! /usr/bin/perl -w
 
 #! /usr/bin/perl -w
 

-# Copyright (C) 2006, 2008 Free Software Foundation, Inc.
+# Copyright (C) 2006, 2008-2013 Free Software Foundation, Inc.

 #
 # This file is part of Bison, the GNU Compiler Compiler.
 #
 #
 # This file is part of Bison, the GNU Compiler Compiler.
 #


@@ -19,22 +19,44 @@
 
 =head1 NAME
 
 
 =head1 NAME
 

-bench.pl - perform benches on Bison parsers.
+bench.pl - bench marks for Bison parsers.

 
 =head1 SYNOPSIS
 
 
 =head1 SYNOPSIS
 

-  ./bench.pl [OPTIONS]... BENCHES
+  ./bench.pl [OPTIONS]... DIRECTIVES

 
 

-=head1 BENCHES
+=head1 DIRECTIVES

 
 

-Specify the set of benches to run.  I<bench-name> should be one of:
+Specify the set of benches to run.  The following grammar defines the
+I<directives>:
+
+   directives ::=
+       directives | directives  -- Alternation
+     | directives & directives  -- Concatenation
+     | [ directives> ]          -- Optional
+     | ( directives> )          -- Parentheses
+     | %b PATH                  -- Use bison at PATH for this bench
+     | #d NAME[=VALUE]          -- %code { #define NAME [VALUE] }
+     | %d NAME[=VALUE]          -- %define NAME ["VALUE"]
+     | %s skeleton              -- %skeleton "skeleton"
+     | directive
+
+Parentheses only group to override precedence.  For instance:
+
+  [ %debug ] & [ %error-verbose ] & [ %define variant ]
+
+will generate eight different cases.
+
+=head1 OPTIONS

 
 =over 4
 
 
 =over 4
 

-=item I<fusion>
+=item B<-b>, B<--bench>
+
+Predefined benches, that is, combimation between a grammar and a I<directives>
+request.

 
 

-Test F<lalr1.cc> with three stacks against F<lalr1-fusion.cc> which
-uses a single one.
+=over 4

 
 =item I<push>
 
 
 =item I<push>
 


@@ -46,12 +68,35 @@ Test the use of variants instead of union in the C++ parser.
 
 =back
 
 
 =back
 

-=head1 OPTIONS
-

 =item B<-c>, B<--cflags>=I<flags>
 
 Flags to pass to the C or C++ compiler.  Defaults to -O2.
 
 =item B<-c>, B<--cflags>=I<flags>
 
 Flags to pass to the C or C++ compiler.  Defaults to -O2.
 

+=item B<-d>, B<--directive>=I<directives>
+
+Add a set of Bison directives to bench against each other.
+
+=item B<-g>, B<--grammar>=I<grammar>
+
+Select the base I<grammar> to use.  Defaults to I<calc>.
+
+=over 4
+
+=item I<calc>
+
+Traditional calculator.
+
+=item I<list>
+
+C++ grammar that uses std::string and std::list.  Can be used with
+or without %define variant.
+
+=item I<triangular>
+
+Artificial grammar with very long rules.
+
+=back
+

 =item B<-h>, B<--help>
 
 Display this message and exit succesfully.  The more verbose, the more
 =item B<-h>, B<--help>
 
 Display this message and exit succesfully.  The more verbose, the more


@@ -104,6 +149,10 @@ The C++ compiler.
 
 Compiler flags (C or C++).
 
 
 Compiler flags (C or C++).
 

+=item C<@directive>
+
+A list of directive sets to measure against each other.
+

 =item C<$iterations>
 
 The number of times the parser is run for a bench.
 =item C<$iterations>
 
 The number of times the parser is run for a bench.


@@ -116,10 +165,13 @@ Verbosity level.
 
 =cut
 
 
 =cut
 

+my $bench;

 my $bison = $ENV{'BISON'} || '@abs_top_builddir@/tests/bison';
 my $cc = $ENV{'CC'} || 'gcc';
 my $cxx = $ENV{'CXX'} || 'g++';
 my $cflags = '-O2';
 my $bison = $ENV{'BISON'} || '@abs_top_builddir@/tests/bison';
 my $cc = $ENV{'CC'} || 'gcc';
 my $cxx = $ENV{'CXX'} || 'g++';
 my $cflags = '-O2';

+my @directive = ();
+my $grammar = 'calc';

 my $iterations = -1;
 my $verbose = 1;
 
 my $iterations = -1;
 my $verbose = 1;
 


@@ -147,22 +199,14 @@ sub verbose($$)
 
 Format the list of directives for Bison for bench named C<$bench>.
 
 
 Format the list of directives for Bison for bench named C<$bench>.
 

-The special fake C<%variant> directive requests the use of
-Boost.Variants instead of a regular union.  So don't pass it, it is
-not a valid directive.
-

 =cut
 
 sub directives($@)
 {
   my ($bench, @directive) = @_;
 =cut
 
 sub directives($@)
 {
   my ($bench, @directive) = @_;

-  my $res = "/* Directives for bench `$bench'. */\n";
-  for my $d (@directive)
-    {
-      $res .= $d . "\n"
-        unless $d eq '%variant';
-    }
-  $res .= "/* End of directives for bench `$bench'. */\n";
+  my $res = "/* Directives for bench '$bench'. */\n";
+  $res .= join ("\n", @directive) . "\n";
+  $res .= "/* End of directives for bench '$bench'. */\n";

   return $res;
 }
 
   return $res;
 }
 


@@ -203,7 +247,6 @@ sub generate_grammar_triangular ($$@)
     or die;
   print $out <<EOF;
 %error-verbose
     or die;
   print $out <<EOF;
 %error-verbose

-%debug

 %{
 #include <stdio.h>
 #include <stdlib.h>
 %{
 #include <stdio.h>
 #include <stdlib.h>


@@ -242,8 +285,8 @@ for my $size (1 .. $max)
   {
     use Text::Wrap;
     print $out wrap ("| ", "   ",
   {
     use Text::Wrap;
     print $out wrap ("| ", "   ",

-                    (map { "\"$_\"" } (1 .. $size)),
-                    " END \n"),
+                     (map { "\"$_\"" } (1 .. $size)),
+                     " END \n"),

                "    { \$\$ = $size; }\n";
   };
 print $out ";\n";
                "    { \$\$ = $size; }\n";
   };
 print $out ";\n";


@@ -275,7 +318,9 @@ yyerror (const char *msg)
 int
 main (void)
 {
 int
 main (void)
 {

+#if YYDEBUG

   yydebug = !!getenv ("YYDEBUG");
   yydebug = !!getenv ("YYDEBUG");

+#endif

   return yyparse ();
 }
 EOF
   return yyparse ();
 }
 EOF


@@ -319,12 +364,16 @@ sub generate_grammar_calc ($$@)
   my ($base, $max, @directive) = @_;
   my $directives = directives ($base, @directive);
 
   my ($base, $max, @directive) = @_;
   my $directives = directives ($base, @directive);
 

+  # Putting this request here is stupid, since the input will be
+  # generated each time we generate a grammar.
+  calc_input ('calc', 200);
+

   my $out = new IO::File ">$base.y"
     or die;
   print $out <<EOF;
 %{
   my $out = new IO::File ">$base.y"
     or die;
   print $out <<EOF;
 %{

+#include <assert.h>

 #include <stdio.h>
 #include <stdio.h>

-

 #include <stdlib.h>
 #include <string.h>
 #include <ctype.h>
 #include <stdlib.h>
 #include <string.h>
 #include <ctype.h>


@@ -363,7 +412,7 @@ static int yylex (void);
 %token <ival> NUM "number"
 %type  <ival> exp
 
 %token <ival> NUM "number"
 %type  <ival> exp
 

-%nonassoc '=' /* comparison           */
+%nonassoc '=' /* comparison            */

 %left '-' '+'
 %left '*' '/'
 %left NEG     /* negation--unary minus */
 %left '-' '+'
 %left '*' '/'
 %left NEG     /* negation--unary minus */


@@ -480,9 +529,8 @@ yylex (void)
 static int
 power (int base, int exponent)
 {
 static int
 power (int base, int exponent)
 {

+  assert (0 <= exponent);

   int res = 1;
   int res = 1;

-  if (exponent < 0)
-    exit (3);

   for (/* Niente */; exponent; --exponent)
     res *= base;
   return res;
   for (/* Niente */; exponent; --exponent)
     res *= base;
   return res;


@@ -496,6 +544,10 @@ main (int argc, const char **argv)
   int count = 0;
   int status;
 
   int count = 0;
   int status;
 

+#if YYDEBUG
+  yydebug = !!getenv ("YYDEBUG");
+#endif
+

   input = fopen ("calc.input", "r");
   if (!input)
     {
   input = fopen ("calc.input", "r");
   if (!input)
     {


@@ -516,136 +568,177 @@ EOF
 
 ##################################################################
 
 
 ##################################################################
 

-=item C<generate_grammar_variant ($base, $max, @directive)>
+=item C<generate_grammar_list ($base, $max, @directive)>

 
 

-Generate a Bison file F<$base.y> that uses, or not, the Boost.Variants
-depending on the C<@directive>.
+Generate a Bison file F<$base.y> for a C++ parser that uses C++
+objects (std::string, std::list).  Tailored for using %define variant.

 
 =cut
 
 
 =cut
 

-sub generate_grammar_variant ($$@)
+sub generate_grammar_list ($$@)

 {
   my ($base, $max, @directive) = @_;
   my $directives = directives ($base, @directive);
 {
   my ($base, $max, @directive) = @_;
   my $directives = directives ($base, @directive);

-  my $variant = grep { $_ eq '%variant' } @directive;
-
+  my $variant = grep { /%define "?variant"?/ } @directive;
+  my $token_ctor = grep { /%define "?api.token.constructor"?/ } @directive;

   my $out = new IO::File ">$base.y"
     or die;
   print $out <<EOF;
 %language "C++"
 %defines
   my $out = new IO::File ">$base.y"
     or die;
   print $out <<EOF;
 %language "C++"
 %defines

+%locations

 $directives
 
 $directives
 

-%code requires // variant.h
+%code requires // *.h

 {
 #include <string>
 }
 
 {
 #include <string>
 }
 

-%code // variant.c
+%code // *.c

 {
 #include <algorithm>
 #include <iostream>
 #include <sstream>
 
 {
 #include <algorithm>
 #include <iostream>
 #include <sstream>
 

-// Prototype of the yylex function providing subsequent tokens.
-static yy::parser::token_type yylex(yy::parser::semantic_type* yylval);
-

 #define STAGE_MAX    ($max * 10) // max = $max
 #define STAGE_MAX    ($max * 10) // max = $max

+
+#define USE_TOKEN_CTOR $token_ctor

 #define USE_VARIANTS $variant
 #define USE_VARIANTS $variant

-#if USE_VARIANTS
-# define IF_VARIANTS(True, False) True
+
+  // Prototype of the yylex function providing subsequent tokens.
+  static
+#if USE_TOKEN_CTOR
+  yy::parser::symbol_type yylex();

 #else
 #else

-# define IF_VARIANTS(True, False) False
+  yy::parser::token_type yylex(yy::parser::semantic_type* yylvalp,
+                               yy::parser::location_type* yyllocp);

 #endif
 #endif

+
+  // Conversion to string.
+  template <typename T>
+    inline
+    std::string
+    string_cast (const T& t)
+  {
+    std::ostringstream o;
+    o << t;
+    return o.str ();
+  }

 }
 }

+
+%token END_OF_FILE 0

 EOF
 
   if ($variant)
     {
       print $out <<'EOF';
 EOF
 
   if ($variant)
     {
       print $out <<'EOF';

-%code variant {int,std::string}

 %token <std::string> TEXT
 %token <int> NUMBER
 %printer { std::cerr << "Number: " << $$; } <int>
 %printer { std::cerr << "Text: " << $$; } <std::string>
 %token <std::string> TEXT
 %token <int> NUMBER
 %printer { std::cerr << "Number: " << $$; } <int>
 %printer { std::cerr << "Text: " << $$; } <std::string>

-%token END_OF_FILE 0

 %type <std::string> text result
 
 %%
 result:
 %type <std::string> text result
 
 %%
 result:

-  text                 { /* Throw away the result. */ }
+  text                  { /* Throw away the result. */ }

 ;
 
 text:
 ;
 
 text:

-  /* nothing */                { /* This will generate an empty string */ }
-| text TEXT            { std::swap($$,$1); $$.append($2); }
-| text NUMBER          {
-                         std::swap($$,$1);
-                          std::ostringstream ss;
-                         ss << ' ' << $2;
-                         $$.append(ss.str());
-                        }
+  /* nothing */         { /* This will generate an empty string */ }
+| text TEXT             { std::swap ($$, $2); }
+| text NUMBER           { $$ = string_cast($2); }

 ;
 EOF
     }
   else
     {
 ;
 EOF
     }
   else
     {

-      # Not using Boost variants.
+      # Not using Bison variants.

       print $out <<'EOF';
 %union {int ival; std::string* sval;}
 %token <sval> TEXT
 %token <ival> NUMBER
 %printer { std::cerr << "Number: " << $$; } <ival>
 %printer { std::cerr << "Text: " << *$$; } <sval>
       print $out <<'EOF';
 %union {int ival; std::string* sval;}
 %token <sval> TEXT
 %token <ival> NUMBER
 %printer { std::cerr << "Number: " << $$; } <ival>
 %printer { std::cerr << "Text: " << *$$; } <sval>

-%token END_OF_FILE 0

 %type <sval> text result
 
 %%
 result:
 %type <sval> text result
 
 %%
 result:

-  text                 { delete $1; }
+  text                  { delete $1; }

 ;
 
 text:
 ;
 
 text:

-  /* nothing */                { $$ = new std::string; }
-| text TEXT            { $$->append(*$2); delete $2; }
-| text NUMBER          {
-                         std::ostringstream ss;
-                         ss << ' ' << $2;
-                         $$->append(ss.str());
-                        }
+  /* nothing */         { $$ = new std::string; }
+| text TEXT             { delete $1; $$ = $2; }
+| text NUMBER           { delete $1; $$ = new std::string (string_cast ($2)); }

 ;
 EOF
     }
 
   print $out <<'EOF';
 %%
 ;
 EOF
     }
 
   print $out <<'EOF';
 %%

+#
+

 static
 static

-yy::parser::token_type
-yylex(yy::parser::semantic_type* yylval)
+#if USE_TOKEN_CTOR
+yy::parser::symbol_type yylex()
+#else
+yy::parser::token_type yylex(yy::parser::semantic_type* yylvalp,
+                             yy::parser::location_type* yyllocp)
+#endif

 {
 {

+  typedef yy::parser::location_type location_type;
+  typedef yy::parser::token token;

   static int stage = -1;
   ++stage;
   if (stage == STAGE_MAX)
   static int stage = -1;
   ++stage;
   if (stage == STAGE_MAX)

-    return yy::parser::token::END_OF_FILE;
+    {
+#if USE_TOKEN_CTOR
+      return yy::parser::make_END_OF_FILE (location_type ());
+#else
+      *yyllocp = location_type ();
+      return token::END_OF_FILE;
+#endif
+    }

   else if (stage % 2)
     {
   else if (stage % 2)
     {

-      IF_VARIANTS(yylval->build<int>(), yylval->ival) = stage;
-      return yy::parser::token::NUMBER;
+#if USE_TOKEN_CTOR
+      return yy::parser::make_NUMBER (stage, location_type ());
+#else
+# if defined ONE_STAGE_BUILD
+      yylvalp->build(stage);
+# elif USE_VARIANTS
+      yylvalp->build<int>() = stage;
+# else
+      yylvalp->ival = stage;
+# endif
+      *yyllocp = location_type ();
+      return token::NUMBER;
+#endif

     }
   else
     {
     }
   else
     {

-      IF_VARIANTS(yylval->build<std::string>() =, yylval->sval = new) std::string("A string.");
-      return yy::parser::token::TEXT;
+#if USE_TOKEN_CTOR
+      return yy::parser::make_TEXT ("A string.", location_type ());
+#else
+# if defined ONE_STAGE_BUILD
+      yylvalp->build(std::string("A string."));
+# elif USE_VARIANTS
+      yylvalp->build<std::string>() = std::string("A string.");
+# else
+      yylvalp->sval = new std::string("A string.");
+# endif
+      *yyllocp = location_type ();
+      return token::TEXT;
+#endif

     }
   abort();
 }
 
 // Mandatory error function
 void
     }
   abort();
 }
 
 // Mandatory error function
 void

-yy::parser::error(const yy::parser::location_type& yylloc,
-                  const std::string& message)
+yy::parser::error(const yy::parser::location_type& loc, const std::string& msg)

 {
 {

-  std::cerr << yylloc << ": " << message << std::endl;
+  std::cerr << loc << ": " << msg << std::endl;

 }
 
 int main(int argc, char *argv[])
 }
 
 int main(int argc, char *argv[])


@@ -671,12 +764,12 @@ Generate F<$base.y> by calling C<&generate_grammar_$name>.
 sub generate_grammar ($$@)
 {
   my ($name, $base, @directive) = @_;
 sub generate_grammar ($$@)
 {
   my ($name, $base, @directive) = @_;

-  verbose 2, "Generating $base.y\n";
+  verbose 3, "Generating $base.y\n";

   my %generator =
     (
       "calc"       => \&generate_grammar_calc,
   my %generator =
     (
       "calc"       => \&generate_grammar_calc,

+      "list"       => \&generate_grammar_list,

       "triangular" => \&generate_grammar_triangular,
       "triangular" => \&generate_grammar_triangular,

-      "variant"    => \&generate_grammar_variant,

     );
   &{$generator{$name}}($base, 200, @directive);
 }
     );
   &{$generator{$name}}($base, 200, @directive);
 }


@@ -692,7 +785,7 @@ Run, possibly verbosely, the shell C<$command>.
 sub run ($)
 {
   my ($command) = @_;
 sub run ($)
 {
   my ($command) = @_;

-  verbose 2, "$command\n";
+  verbose 3, "$command\n";

   system ("$command") == 0
     or die "$command failed";
 }
   system ("$command") == 0
     or die "$command failed";
 }


@@ -714,38 +807,44 @@ sub compile ($)
 
   my $compiler = $language eq 'C++' ? $cxx : $cc;
 
 
   my $compiler = $language eq 'C++' ? $cxx : $cc;
 

-  run "$bison $base.y -o $base.c";
+  my $my_bison = `sed -ne '/%bison "\\(.*\\)"/{s//\\1/;p;q;}' $base.y`;
+  run ((length $my_bison ? $my_bison : $bison) . " $base.y -o $base.c");

   run "$compiler -o $base $cflags $base.c";
 }
 
 ######################################################################
 
   run "$compiler -o $base $cflags $base.c";
 }
 
 ######################################################################
 

-=item C<bench_grammar ($gram, %bench)>
-
-Generate benches for C<$gram>.  C<$gram> should be C<calc> or
-C<triangle>.  C<%bench> is a hash of the form:
+=item C<bench ($grammar, @token)>

 
 

-  $name => @directive
-
-where C<$name> is the name of the bench, and C<@directive> are the
-Bison directive to use for this bench.  All the benches are compared
-against each other, repeated 50 times.
+Generate benches for the C<$grammar> and the directive specification
+given in the list of C<@token>.

 
 =cut
 
 
 =cut
 

-sub bench_grammar ($%)
+sub bench ($@)

 {
 {

-  my ($gram, %test) = @_;
-
+  my ($grammar, @token) = @_;

   use Benchmark qw (:all :hireswallclock);
 
   use Benchmark qw (:all :hireswallclock);
 

+  my @directive = parse (@token);
+

   # Set up the benches as expected by timethese.
   my %bench;
   # Set up the benches as expected by timethese.
   my %bench;

+  # A counter of directive sets.
+  my $count = 1;
+  for my $d (@directive)
+    {
+      $bench{$count} = $d;
+      printf " %2d. %s\n", $count, join (' ', split ("\n", $d));
+      $count++;
+    };
+

   # For each bench, capture the size.
   my %size;
   # For each bench, capture the size.
   my %size;

-  while (my ($name, $directives) = each %test)
+
+  while (my ($name, $directives) = each %bench)

     {
     {

-      generate_grammar ($gram, $name, @$directives);
+      generate_grammar ($grammar, $name, $directives);

       # Compile the executable.
       compile ($name);
       $bench{$name} = "system ('./$name');";
       # Compile the executable.
       compile ($name);
       $bench{$name} = "system ('./$name');";


@@ -761,7 +860,7 @@ sub bench_grammar ($%)
   # shows only wallclock and the two children times.  'auto' (the
   # default) will act as 'all' unless the children times are both
   # zero, in which case it acts as 'noc'.  'none' prevents output.
   # shows only wallclock and the two children times.  'auto' (the
   # default) will act as 'all' unless the children times are both
   # zero, in which case it acts as 'noc'.  'none' prevents output.

-  verbose 2, "Running the benches for $gram\n";
+  verbose 3, "Running the benches for $grammar\n";

   my $res = timethese ($iterations, \%bench, 'nop');
 
   # Output the speed result.
   my $res = timethese ($iterations, \%bench, 'nop');
 
   # Output the speed result.


@@ -794,57 +893,32 @@ interfaces.
 
 sub bench_push_parser ()
 {
 
 sub bench_push_parser ()
 {

-  calc_input ('calc', 200);
-  bench_grammar
-    ('calc',
-     (
-      "pull-impure" => [],
-      "pull-pure"   => ['%define api.pure'],
-      "push-impure" => ['%define api.push_pull "both"'],
-      "push-pure"   => ['%define api.push_pull "both"', '%define api.pure'],
-     )
-    );
+  bench ('calc',
+         qw(
+            [ %d api.pure ]
+            &
+            [ %d api.push-pull=both ]
+         ));

 }
 
 ######################################################################
 
 =item C<bench_variant_parser ()>
 
 }
 
 ######################################################################
 
 =item C<bench_variant_parser ()>
 

-Bench the C++ lalr1.cc parser using Boost.Variants or %union.
+Bench the C++ lalr1.cc parser using variants or %union.

 
 =cut
 
 sub bench_variant_parser ()
 {
 
 =cut
 
 sub bench_variant_parser ()
 {

-  bench_grammar
-    ('variant',
-     (
-      "f-union"         => ['%skeleton "lalr1-fusion.cc"'],
-      "f-uni-deb"   => ['%skeleton "lalr1-fusion.cc"', '%debug'],
-      "f-var"       => ['%skeleton "lalr1-fusion.cc"', '%variant'],
-      "f-var-deb" => ['%skeleton "lalr1-fusion.cc"', '%debug', '%variant'],
-      "f-var-dtr"       => ['%skeleton "lalr1-fusion.cc"', '%variant', "%code {\n#define VARIANT_DESTROY\n}"],
-      "f-var-deb-dtr" => ['%skeleton "lalr1-fusion.cc"', '%debug', '%variant', "%code {\n#define VARIANT_DESTROY\n}"],
-     )
-    );
-}
-
-######################################################################
-
-=item C<bench_fusion_parser ()>
-
-Bench the C++ lalr1.cc parser using Boost.Variants or %union.
-
-=cut
-
-sub bench_fusion_parser ()
-{
-  bench_grammar
-    ('variant',
-     (
-      "split"         => [],
-      "fused"         => ['%skeleton "lalr1-fusion.cc"'],
-     )
+  bench ('list',
+         qw(
+            [
+              %d variant
+              &
+              [ #d ONE_STAGE_BUILD | %d api.token.constructor ]
+            ]
+         )

     );
 }
 
     );
 }
 


@@ -863,11 +937,144 @@ sub help ($)
 
 ######################################################################
 
 
 ######################################################################
 

+# The end of the directives to parse.
+my $eod = "end of directives";
+# The list of tokens parsed by the following functions.
+my @token;
+
+# eat ($EXPECTED)
+# ---------------
+# Check that the current token is $EXPECTED, and move to the next.
+sub eat ($)
+{
+  my ($expected) = @_;
+  die "expected $expected, unexpected: $token[0] (@token)\n"
+    unless $token[0] eq $expected;
+  shift @token;
+}
+
+# Parse directive specifications:
+#   expr: term (| term)*
+#   term: fact (& fact)*
+#   fact: ( expr ) | [ expr ] | dirs
+#   dirs: %s SKELETON | #d NAME[=VALUE] | %d NAME[=VALUE] | directive
+sub parse (@)
+{
+  @token = (@_, $eod);
+  verbose 3, "Parsing: @token\n";
+  my @res = parse_expr ();
+  eat ($eod);
+  return @res;
+}
+
+sub parse_expr ()
+{
+  my @res = parse_term ();
+  while ($token[0] eq '|')
+    {
+      eat ('|');
+      # Alternation.
+      push @res, parse_term ();
+    }
+  return @res;
+}
+
+sub parse_term ()
+{
+  my @res = parse_fact ();
+  while ($token[0] eq '&')
+    {
+      eat ('&');
+      # Cartesian product.
+      my @lhs = @res;
+      @res = ();
+      for my $rhs (parse_fact ())
+        {
+          for my $lhs (@lhs)
+            {
+              push @res, $lhs . ($lhs && $rhs ? "\n" : "") . $rhs;
+            }
+        }
+    }
+  return @res;
+}
+
+sub parse_fact ()
+{
+  my @res;
+  die "unexpected end of expression"
+    unless defined $token[0];
+
+  if ($token[0] eq '(')
+    {
+      eat ('(');
+      @res = parse_expr ();
+      eat (')');
+    }
+  elsif ($token[0] eq '[')
+    {
+      eat ('[');
+      @res = (parse_expr (), '');
+      eat (']');
+    }
+  else
+    {
+      @res = parse_dirs ();
+    }
+  return @res;
+}
+
+sub parse_dirs ()
+{
+  my @res;
+  die "unexpected end of expression"
+    unless defined $token[0];
+
+  if ($token[0] eq '#d')
+    {
+      eat ('#d');
+      $token[0] =~ s/(.*?)=(.*)/$1 $2/;
+      @res = ("%code {\n#define $token[0]\n}");
+      shift @token;
+    }
+  elsif ($token[0] eq '%d')
+    {
+      shift @token;
+      $token[0] =~ s/(.*?)=(.*)/$1 "$2"/;
+      @res = ("%define $token[0]");
+      shift @token;
+    }
+  elsif ($token[0] eq '%s')
+    {
+      shift @token;
+      @res = ("%skeleton \"$token[0]\"");
+      shift @token;
+    }
+  elsif ($token[0] eq '%b')
+    {
+      shift @token;
+      @res = ("/*\n%bison \"$token[0]\"\\\n*/");
+      shift @token;
+    }
+  else
+    {
+      @res = $token[0];
+      shift @token;
+    }
+
+  return @res;
+}
+
+######################################################################
+

 sub getopt ()
 {
   use Getopt::Long;
   my %option = (
 sub getopt ()
 {
   use Getopt::Long;
   my %option = (

+    "b|bench=s"      => \$bench,

     "c|cflags=s"     => \$cflags,
     "c|cflags=s"     => \$cflags,

+    "d|directive=s"  => \@directive,
+    "g|grammar=s"    => \$grammar,

     "h|help"         => sub { help ($verbose) },
     "i|iterations=i" => \$iterations,
     "q|quiet"        => sub { --$verbose },
     "h|help"         => sub { help ($verbose) },
     "i|iterations=i" => \$iterations,
     "q|quiet"        => sub { --$verbose },


@@ -881,17 +1088,46 @@ sub getopt ()
 ######################################################################
 
 getopt;
 ######################################################################
 
 getopt;

+
+# Create the directory we work in.
+mkdir "benches" or die "cannot create benches"
+  unless -d "benches";
+my $count = 1;
+++$count
+  while -d "benches/$count";
+my $dir = "benches/$count";
+mkdir $dir
+  or die "cannot create $dir";
+chdir $dir
+  or die "cannot chdir $dir";
+
+# The following message is tailored to please Emacs' compilation-mode.
+verbose 1, "Entering directory `$dir'\n";

 verbose 1, "Using bison=$bison.\n";
 verbose 1, "Using bison=$bison.\n";

-verbose 1, "Using cc=$cc.\n";
-verbose 1, "Using cxx=$cxx.\n";
-verbose 1, "Using cflags=$cflags.\n";
+verbose 2, "Using cc=$cc.\n";
+verbose 2, "Using cxx=$cxx.\n";
+verbose 2, "Using cflags=$cflags.\n";
+verbose 2, "Grammar: $grammar\n";

 
 

-for my $b (@ARGV)
+
+# Support -b: predefined benches.
+my %bench =
+  (
+   "push"     => \&bench_push_parser,
+   "variant"  => \&bench_variant_parser,
+  );
+
+if (defined $bench)
+{
+  die "invalid argument for --bench: $bench"
+    unless defined $bench{$bench};
+  &{$bench{$bench}}();
+  exit 0;
+}
+else

 {
 {

-  verbose 1, "Running benchmark $b.\n";
-  bench_fusion_parser()  if $b eq "fusion";
-  bench_push_parser()    if $b eq "push";
-  bench_variant_parser() if $b eq "variant";
+  # Launch the bench marking.
+  bench ($grammar, @ARGV);

 }
 
 ### Setup "GNU" style for perl-mode and cperl-mode.
 }
 
 ### Setup "GNU" style for perl-mode and cperl-mode.