source: other-projects/nightly-tasks/diffcol/trunk/diffcol/gdbdiff.pm@ 27743

package gdbdiff;

BEGIN {
    die "GSDLHOME not set\n" unless defined $ENV{'GSDLHOME'};
    die "GSDLOS not set\n" unless defined $ENV{'GSDLOS'};
    unshift (@INC, "$ENV{'GSDLHOME'}/perllib");
    unshift (@INC, "$ENV{'GSDLHOME'}/perllib/cpan");
}

use util;
use diffutil;
use Text::Diff;
use Cwd;

if ($^O =~ m/mswin/i) {
    require Win32; # for working out Windows Long Filenames from Win 8.3 short filenames
}

sub readin_gdb
{
    my ($cmd) = @_;

    open(PIN,"$cmd|") 
    || die "Unable to open pipe to $cmd: $!\n";

    my $text_content = "";

    while (defined (my $line = <PIN>)) {
    $text_content .= $line;
    }

    close(PIN);
    return $text_content;
}

# for debugging. Prints txt contents of db to file
sub print_string_to_file 
{
    my ($text, $outfile) = @_;
    
    open(FOUT, ">$outfile") or die "ERROR failed to write to $outfile: $!\n";
    print FOUT $text;
    close(FOUT);
}

sub test_gdb
{
    my ($full_modeldb, $full_testdb,$strColName) = @_;

   # print "Now is testing database\n";
    my ($dbname, $dirname, $suffix)= &File::Basename::fileparse($full_testdb, "\\.[^\\.]+\$");
   
    # need to sort text output of both test and model col database files, to normalise them for the comparison
    # the -sort option to db2txt was added specifically to support diffcol
    my $model_cmd = "db2txt -sort $full_modeldb 2>&1";
    my $test_cmd  = "db2txt -sort $full_testdb 2>&1";

    my $model_text = readin_gdb($model_cmd);
    my $test_text = readin_gdb($test_cmd);

#   my $savepath = &getcwd."/../"; # TASK_HOME env var does not exist at this stage, but it's one level up from current directory
#   print_string_to_file($test_text, $savepath.$dbname."_test.out");    
#   print_string_to_file($model_text, $savepath.$dbname."_model.out");

    # filter out the fields that can be ignored in the two database files
    # The total_numbytes field can vary depending on how many backslashes exist in the urls in the main body text, as each
    # of these windows slashes get escaped with another backslash, and the resulting string is used as key into rel link db
    my $ignore_line_re = "\n<(lastmodified|lastmodifieddate|oailastmodified|oailastmodifieddate|ex.File.FileModifyDate|ex.File.FilePermissions|total_numbytes)>([^\n])*";
    $model_text =~ s/$ignore_line_re//g;
    $test_text =~ s/$ignore_line_re//g;

    # tmp dirs have subdirs with random numbers in name, remove subdir
    $model_text =~ s@(tmp[\\\/])(\d*[\\\/])@$1@g;
    $test_text =~ s@(tmp[\\\/])(\d*[\\\/])@$1@g;

    # if the OS doesn't match and one of them is windows, extra work needs to be done to bring the db files
    # in test and model collection to an even base for comparison
    
    my $testIsWin = &isDBWindowsSensitive($dbname, $test_text);
    my $modelIsWin = &isDBWindowsSensitive($dbname, $model_text);
    
    if($testIsWin == $modelIsWin) { # both linux or both windows, do the basic test we did on linux machines:
            # ignore absolute path prefixes in modelcol and testcol (necessary for archiveinf-doc and -src.gdb files)

            # Remember the original model col on SVN could have been built anywhere, 
            # and in the gdb files, absolute paths are stored to the collection location. 
            # Crop these paths to the collect/<colname> point.
            
            # Entries are of the form [Entry] or <Entry>. In order to do a sensible diff,
            # need to remove the prefix to the collect/colname folder in any (absolute) path that occurs in Entry
            # E.g. [/full/path/collect/colname/import/file.ext] should become [collect/colname/import/file.ext]
            # Better regex is of the form /BEGIN((?:(?!BEGIN).)*)END/, see http://docstore.mik.ua/orelly/perl/cookbook/ch06_16.htm

            $model_text =~ s@^([^\\/]*(//)*).*(\\|/)(collect(\\|/)$strColName)(.*)$@$1$4$6@mg;
            $test_text =~ s@^([^\\/]*(//)*).*(\\|/)(collect(\\|/)$strColName)(.*)$@$1$4$6@mg;
            #$model_text =~ s@^([^\\//]*).*(\\|/)(collect(\\|/)$strColName)(.*)$@$1$3$5@mg;         
            #$test_text =~ s@^([^\\//]*).*(\\|/)(collect(\\|/)$strColName)(.*)$@$1$3$5@mg;                      
    }
    
    else { # one of the collections was built on windows
        # handling slashes and other differences between a model coll built on one OS (e.g. linux) 
        # and a test collection built and diffed on another OS (windows)
    
        my ($win_text, $lin_text); # references
        if($testIsWin) {
            $win_text = \$test_text;
            $lin_text = \$model_text;
        } else {
            $win_text = \$model_text;
            $lin_text = \$test_text;
        }
        
        if($dbname =~ m/archiveinf-doc/) {
            my $tmp = ""; # rebuild windows file's set of lines after processing them one by one
    
            # convert short filenames to long perl: 
            # http://www.mombu.com/programming/perl/t-convert-dos-83-filenames-to-win32-long-filenames-using-perl-525448.html
            for my $line (split /^/, $$win_text) { # split the string into newlines
                
                # assoc-file and meta-file contain filepaths, ensure these are long windows file paths now (will later convert to linux slashes)    
                if($line =~ m@^<(assoc-file|meta-file)>(.*)(\s+)@s) { 
                    $line = $2; # may be a short file name
                    # perhaps test here if it is a shortfilename? should match /CAPS....~number(.ext)/
                
                    $line = "<$1>".&Win32::GetLongPathName($line)."$3"; # make it a long file name and prefix assoc-file/meta-file tagname to it again                  
                }
                $tmp .= $line;
            }
            $$win_text = $tmp;          
        }
        
        
        # index gdb file
        if($dbname =~ m/$strColName/) {
            my $tmp = ""; # rebuild windows file's set of lines after processing them one by one
            for my $line (split /^/, $$win_text) { # split the string into newlines
                
                if($line =~ m@^<ex.File.Directory>(.*)@s) { # word-pdf collection contains double windows backslashes
                    (my $escaped_path = $1) =~ s@\\\\@\\@g;             
                    $line = "<ex.File.Directory>$escaped_path";
                } 
                elsif($line =~ m@^<Title>(.*)@s) {
#                   print STDERR "***** TITLE: |$1|\n";
                
                    # word-pdf collection: Title of ps files contain new lines at end when
                    # GreenstoneXMLPlugin::xml_end_tag() writes the Title back out after utf8 decode
                    # if($metadata_name eq "Title") { $metadata_value =~ s/[\n\r]*$//; }
                
                    (my $title = $1) =~ s@(\r|\n|\\n)*$@@; # get rid of trailing newlines/carriage returns
                    $line = "<Title>$title\n"; # add single newline                 
                }
                $tmp .= $line;
            }
            $$win_text = $tmp;          
        }
        
        
        # slashes in windows metadata text need to be turned into linux style slashes
        $$win_text =~ s@\\@/@g; #$$win_text =~ s@\\([^n|r|\|"])@/$1@g; # filepath something\rtf remains something\rtf
        
        # cut down absolute paths to files to just collect/colname/.../file, same as before
        $$lin_text =~ s@^([^\\/]*(//)*).*(\\|/)(collect(\\|/)$strColName)(.*)$@$1$4$6@mg; # $$lin_text =~ s@^([^\\\/]*(//)?).*(\\|/)(collect(\\|/)$strColName)(.*)$@$1$4$6@mg;
        $$win_text =~ s@^([^\\/]*(//)*).*(\\|/)(collect(\\|/)$strColName)(.*)$@$1$4$6@mg;       
        
        # for the windows text, need to further get rid of the driveletter after [ or <meta>
        $$win_text =~ s@^(\[|<[^>]*>)[a-zA-Z]:collect@$1collect@mg;     
        
    } # end of equalising differences between a windows collection's db file and linux coll's db file

    # now can go back to using $model_text and $test_text
#   print_string_to_file($test_text, $savepath.$dbname."_test.out");    
#   print_string_to_file($model_text, $savepath.$dbname."_model.out");
    
    my $report_type = "OldStyle"; # Can not change this type.
    my $diff_gdb = diff \$model_text, \$test_text, { STYLE => $report_type };
    
    # leaving the ignore regex as it used to be in the following, in case it helps with single line comparisons
    $diff_gdb = &diffutil::GenerateOutput($diff_gdb,"^<(lastmodified|lastmodifieddate|oailastmodified|oailastmodifieddate|ImageSize)>.*");

    if($diff_gdb eq "")
    {
    return "";
    }
    else
    {
    return "Difference Report: Differences found in the Database file: \n$diff_gdb";
    }
    # Call diff?
}

# returns true if the contents are windows AND it matters for the diffing on the db that it's windows
# For col.gdb it does not seem to matter so far, if it is generated on a windows machine and to be compared to a linux-generated col.gdb
sub isDBWindowsSensitive 
{
    my ($dbtailname, $db_contents) = @_; # db filename without suffix
    
    #if($dbtailname !~ m/archiveinf/) { # only archiveinf-doc and archive-inf source need special Windows processing, not col.gdb
    #   return 0;
    #}
    return ($db_contents =~ m/\\/) ? 1 : 0; # windows slashes detected. Better test would be: [Something\something] OR <tag>something\something
    # for doc.xml:
    #     <Metadata name="gsdlsourcefilename">import/html_files/cleves.html</Metadata>
}

1;