Regex VM finished but untested.

src/backends/regexvm_asm.pl

@@ -0,0 +1,112 @@
+#!/usr/bin/perl -w
+
+use strict;
+# The input file consists of a sequence of blocks, which can be parsed
+# as SVM test cases, RVM test cases, or C functions. Each block starts
+# with a header line, then a sequence of options, and finally text in
+# a format defined by the block type.
+#
+# Header lines start with "+TYPE", optionally followed by a name. This
+# name is semantically meaningful for SVM and RVM blocks; it
+# determines the name of the test case.
+
+# A C block's name is not used, and it takes no options. The body
+# (which continues until the first line that looks like a header), is
+# just passed straight through into the C source.
+
+# SVM blocks' names are the GLib test case name. The underlying
+# function's name is derived by substituting invalid characters with
+# '_'. Note that this can result in collisions (eg, /foo_bar/baz
+# collides with /foo/bar_baz). If this happens, it's your own damn
+# fault; rename the blocks. SVM blocks take three different options:
+# @input, @output, and @pre. The @input pragma's argument is a
+# C-quoted string that gets passed into the VM as the input string,
+# and @output is a C-quoted string that is compared against
+# h_write_result_unamb.  @pre lines are prepended verbatim to the
+# function body (with the @pre stripped, of course); they can be used
+# to initialize environment values.
+#
+# SVM instructions consist of either two or four fields:
+#
+#     input_pos opcode [arg env]
+#
+# input_pos and opcode correspond to the fields in HRVMTrace.  arg and
+# env are used to populate an HSVMAction; arg is the function, and env
+# is the object whose address should be used as the env.
+
+# RVM blocks are very similar to SVM blocks; the name and options are
+# handled exactly the same way. The assembly text is handled slightly
+# differently; the format is:
+#
+#     [label:] opcode [arg ...]
+#
+# For FORK and GOTO, the arg should be a label that is defined
+# elsewhere.
+#
+# For ACTION, the arguments are handled the same way as with SVM.
+#
+# MATCH takes two arguments, each of which can be any C integer
+# constant (not including character constants), which form the lower
+# and upper bounds of the matched character, respectively.
+#
+# No other RVM instructions take an argument.
+
+# At the beginning of any line, comments preceeded by '#' are allowed;
+# they are replaced by C++ comments and inserted in the nearest valid
+# location in the output.
+
+my $mode == "TOP";
+
+# common regexes:
+my $re_ident = qr/[A-Za-z_][A-Za-z0-9_]*/;
+my $re_cstr = qr/"(?:[^\\"]|\\["'abefnrtv0\\]|\\x[0-9a-fA-F]{2}|\\[0-7]{3})*"/;
+
+
+my %svm = (
+    name => sub {
+	my ($env, $name) = @_;
+	$env->{name} = $name;
+    },
+    pragma => sub {
+	my ($env, $name, $val) = @_;
+	if ($name eq "input") {
+	    chomp($env->{input} = $val);
+	} elsif ($name eq "output") {
+	    chomp($env->{output} = $val);
+	} elsif ($name eq "pre") {
+	    # Do I have the ref precedence right here?
+	    push(@$env->{pre}, $val);
+	} else {
+	    warn "Invalid SVM pragma";
+	}
+    },
+    body => sub {
+	my ($env, $line) = @_;
+	my ($ipos, $op, $arg, $argenv);
+	if ($line =~ /^\s*(\d+)\s+(PUSH|NOP|ACTION|CAPTURE|ACCEPT)(?:\s+($re_ident)\s+($re_ident))?/) {
+	    if ($2 eq "PUSH") {
+		# TODO: implement all the opcodes
+	    }
+	}
+    }
+    );
+
+
+while (<>) {
+    if (/^+(C|RVM|SVM)/) {
+	$mode = $1;
+    }
+    
+    if ($mode eq "TOP") {
+	if (/^#(.*)/) {
+	    print "// $1";
+	    next;
+	}
+    } elsif ($mode eq "SVM") {
+    } elsif ($mode eq "RVM") {
+    } elsif ($mode eq "C") {
+    }
+    
+}
+
+