Windows-Server-2003/mergedcomponents/setupinfs/reorder_layout.pl

389 lines
12 KiB
Perl

#!perl -w
#
# reorder_layout.pl -- A tool to reorganize the layout.inx file based on usage
#
# Author: Scott Mackowski (ScottMa)
#
##############################################################################
##############################################################################
# Globals
##############################################################################
my %desiredoutputorder; # Hash to map: number -> {FN, DN}
my %orderedfiles; # Hash to map: {FN, DN} -> number
my %datalines; # Layout.inx lines associated with FN, DN
my $currentfilenumber = 0; # Index of the current file into the sort
# order hashtables.
##############################################################################
# Functions
##############################################################################
#
# This sort routine is for numeric sorting...
#
sub numerically { $a <=> $b; }
#
# Displays usage info for this script and then aborts.
#
sub ShowUsageInformation
{
printf STDERR "Usage: $0 (layout.inx) (usage_file)\n";
die "\n";
} # ShowUsageInformation
#
# Flushes all lines for the current section of the INX file, using the
# desired output order.
#
sub FlushFileSection
{
#
# Walk through the data hash in the desired output order.
# If the file has been seen in the current section (has data),
# write out each of its associated lines in order, then remove
# the entry from all hashes.
#
foreach $recordnumber (sort numerically keys %desiredoutputorder)
{
my $filename = $desiredoutputorder{$recordnumber}{FN};
my $dirnum = $desiredoutputorder{$recordnumber}{DN};
if ( (exists ($datalines{$filename}) )
&& (exists ($datalines{$filename}{$dirnum}) ) )
{
foreach $linenumber (sort numerically keys %{ $datalines{$filename}{$dirnum} })
{
printf OUTPUTFILE ("%s\n", $datalines{$filename}{$dirnum}{$linenumber} );
}
delete $datalines{$filename}{$dirnum};
delete $desiredoutputorder{$recordnumber};
delete $orderedfiles{$filename}{$dirnum};
}
}
#
# Flush the extra lines from the end of the section that
# were read but not associated with any entry.
#
if ($archivedlinecount > 0)
{
foreach $archivedlinecount (sort numerically keys %archivedlines)
{
printf OUTPUTFILE ("%s\n", $archivedlines{$archivedlinecount});
delete $archivedlines{$archivedlinecount};
}
$archivedlinecount = 0;
$lastarchiveassocdirnum = "";
$lastarchiveassocfile = "";
}
}
##############################################################################
# Main
##############################################################################
#
# If we have no commandline arguments, abort with usage
#
if ($#ARGV < 0)
{
&ShowUsageInformation();
}
#
# First, we read the current layout.inx file and extract the list of
# directories and their associated numbers.
#
$currentinputfile = $ARGV[0];
open(INPUTFILE, $currentinputfile)
or die "\nCan't open input file $currentinputfile\n";
$inDirectorySection = 0; # This variable will be set when we are
# in the section of the file that has the
# directory number -> directory mappings.
LINE: while( $line = <INPUTFILE> )
{
# Chomp the trailing newline off; lowercase the line.
chomp $line;
$line = lc $line;
#
# If this line starts with a bracket, we are entering a new section.
# Reset the "inDirectorySection" variable, and only set it if the
# section name matches.
#
if ($line =~ /^\[/)
{
$inDirectorySection = 0;
}
if ($line eq "[winntdirectories]")
{
$inDirectorySection = 1;
next LINE;
}
#
# Skip lines outside of the targetted section and comment lines.
#
next LINE if (! $inDirectorySection);
next LINE if ($line =~ /^@\*/);
#
# Extract the directory name from the line as everything on the
# right side of the = sign, removing leading whitespace.
# Also extract the directory number as the string of digits
# immediately preceding the = sign, removing trailing whitespace.
#
($directory = $line) =~ s/.*=\s*(.*)/$1/;
($dirnum = $line) =~ s/(.*:)?(\d*)\s*=.*/$2/;
#
# If this is a good line, the directory number will have been
# extracted, save this directory into the masterdirectoryhash.
#
if ($dirnum)
{
$directory =~ s/^\"(.*)\"$/$1/;
$directory =~ s/^(.*)\\$/$1/;
$masterdirectoryhash{$directory} = $dirnum;
}
} # while (there is still data in this inputfile)
close (INPUTFILE); # Close the input file...
#
# Now, we read the desired output order file and create the ordering
# for this set of files.
#
$currentinputfile = $ARGV[1];
open(INPUTFILE, $currentinputfile)
or die "\nCan't open input file $currentinputfile\n";
LINE: while( $line = <INPUTFILE> )
{
# Chomp the trailing newline off; lowercase the line.
chomp $line;
$line = lc $line;
#
# Extract the directory name from the line as everything to the
# left of the last backslash (excluding the backslash itself).
# Also extract the filename as everything to the right of the
# last backslash (excluding the backslash itself).
#
($directory = $line) =~ s/(.*)\\.*/$1/;
($filename = $line) =~ s/.*\\(.*)/$1/;
#
# Remove any leading backslash from the directory name.
#
$directory =~ s/^\\(.*)/$1/;
#
# If this is one of the directories we saw in the original layout.inx
# file, add this filename/pathname combonation as the next file to
# place. We save both a forward lookup: {FN, DN} -> number, and a
# ordering lookup: number -> {FN, DN}.
#
if (exists($masterdirectoryhash{$directory}))
{
if ( (! exists($orderedfiles{$filename}))
|| (! exists($orderedfiles{$filename}{$masterdirectoryhash{$directory}})) )
{
$desiredoutputorder{$currentfilenumber}{FN} = $filename;
$desiredoutputorder{$currentfilenumber}{DN} = $masterdirectoryhash{$directory};
$orderedfiles{$filename}{$masterdirectoryhash{$directory}} = $currentfilenumber;
$currentfilenumber++;
}
}
} # while (there is still data in this inputfile)
close (INPUTFILE); # Close the input file...
#
# Finally, we re-read the supplied layout.inx file, and simulataneously
# create the output file. This file has the same name as the input file,
# with ".new" appended to it.
#
$currentinputfile = $ARGV[0];
open(INPUTFILE, $currentinputfile)
or die "\nCan't open input file $currentinputfile\n";
open(OUTPUTFILE, ">" . $currentinputfile . ".new")
or die "\nCan't open output file $currentinputfile.new\n";
$inFileSection = 0; # This variable will be set when we are
# in any section of the file that has the
# actual files' lines to be reordered.
$archivedlinecount = 0; # These variables hold lines that are to
$lastarchiveassocfile = ""; # be associated with the next real line
$lastarchiveassocdirnum = ""; # such as comments, etc.
LINE: while( $line = <INPUTFILE> )
{
# Chomp the trailing newline off.
chomp $line;
#
# If this line starts with a bracket, we are entering a new section.
# Reset the "inFileSection" variable, flush the file section's data,
# and only set it if the section name matches.
#
if ($line =~ /^\[/)
{
$inFileSection = 0;
&FlushFileSection();
}
if ($line =~ /^\[SourceDisksFiles/)
{
$inFileSection = 1;
printf OUTPUTFILE ("%s\n", $line);
next LINE;
}
#
# Skip lines outside of the targetted section, simply copying them
# into the output file.
#
if (! $inFileSection)
{
printf OUTPUTFILE ("%s\n", $line);
next LINE;
}
#
# If the line does not contain an equals sign, assume it is a comment.
#
if ($line !~ /=/)
{
#
# If we are currently associated with an entry (it had comment lines
# above it), add these new lines to the corresponding data hash.
# Otherwise, archive these new lines to be associated later.
#
if ($lastarchiveassocdirnum)
{
$datalines{$lastarchiveassocfile}{$lastarchiveassocdirnum}{scalar(keys %{ $datalines{$lastarchiveassocfile}{$lastarchiveassocdirnum} })} = $line;
}
else
{
$archivedlines{$archivedlinecount} = $line;
$archivedlinecount++;
}
}
else
{
#
# Extract the filename as everything to the left of the equals
# sign (except the prodfilt tags), removing trailing whitespace.
# Also, extract everything to the right of the equals as a set
# of comma-delimited fields.
#
($filename = $line) =~ s/(.*:)?(\S*.?\S*)\s*=.*/$2/;
($fields = $line) =~ s/.*=\s*(.*)/$1/;
#
# Split the fields along the commas and examine the eighth field.
# Extract it as a number that represents the directory entry.
#
my @linefields = split(',', $fields);
$dirnum = $linefields[7];
$dirnum =~ s/\D*//g;
#
# Check the eleventh field. If it exists, use the contents of it,
# stripping any trailing comments, as the real filename.
#
if ($#linefields > 9)
{
if ($linefields[10] ne "")
{
$filename = $linefields[10];
$filename =~ s/;.*//;
}
}
#
# Strip any trailing whitespace from the filename.
#
$filename =~ s/ *$//g;
#
# If there were lines read before this one that need to be
# associated with a data line (such as comments), add those
# to the data hash for this entry now, removing them from
# the temporary archived lines variable.
# Now, set the lastarchiveassoc* variables so trailing
# comments are mapped with this set.
#
if ($archivedlinecount > 0)
{
foreach $archivedlinecount (sort numerically keys %archivedlines)
{
$datalines{$filename}{$dirnum}{scalar(keys %{ $datalines{$filename}{$dirnum} })} = $archivedlines{$archivedlinecount};
delete $archivedlines{$archivedlinecount};
}
$archivedlinecount = 0;
$lastarchiveassocdirnum = $dirnum;
$lastarchiveassocfile = $filename;
}
#
# Reset the lastarchiveassoc* variables so the next comment
# lines are not mapped with this entry, but are instead
# archived for their successor.
#
else
{
$lastarchiveassocdirnum = 0;
$lastarchiveassocfile = "";
}
#
# Add the current line to the data hash for this entry.
#
$datalines{$filename}{$dirnum}{scalar(keys %{ $datalines{$filename}{$dirnum} })} = $line;
#
# If this file has not already been ordered (wasn't in the
# supplied order file and hasn't already been seen), assign
# it the next available slot and record the file as seen.
#
if ( (! exists($orderedfiles{$filename}))
|| (! exists($orderedfiles{$filename}{$dirnum})) )
{
$desiredoutputorder{$currentfilenumber}{FN} = $filename;
$desiredoutputorder{$currentfilenumber}{DN} = $dirnum;
$orderedfiles{$filename}{$dirnum} = $currentfilenumber;
$currentfilenumber++;
}
}
} # while (there is still data in this inputfile)
close (INPUTFILE); # Close the input file...
&FlushFileSection(); # Flush any data from the last section here
close (OUTPUTFILE); # Close the output file...