source: other-projects/nightly-tasks/diffcol/trunk/diffcol/gdbdiff.pm@ 28172

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, $test_os, $model_os) = @_;

   # 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.out1");   
#   print_string_to_file($model_text, $savepath.$dbname."_model.out1");

    # 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<(FileSize|lastmodified|lastmodifieddate|oailastmodified|oailastmodifieddate|ex.File.FileModifyDate|ex.File.FilePermissions|total_numbytes|ex.Composite.LightValue)>([^\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
    # these tmpdirs are located inside the collection directory
    $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 $testIsWin = ($test_os ne "compute") ? ($test_os eq "windows") : &isDBWindowsSensitive($dbname, $test_text);
    my $modelIsWin = ($model_os ne "compute") ? ($model_os eq "windows") : &isDBWindowsSensitive($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         

                # In the following regex, add any .gdb fieldnames that represent a path and so would contain double backslashes 
                # on Windows (to escape the single backlash of win filepaths). They will be turned into single-backslashes here, 
                # and converted into single forward slashes futher below when the txt version of the win gdb file is normalised 
                # to compare it with the linux version. 
                # E.g. On windows, the Word-PDF collection(s) contains double backslashes in the ex.File.Directory field
                # the MARC-Exploded collection contains double backslashes in the null_file entry field of the .gdb file                
                if($line =~ m@^<(ex.File.Directory|null_file)>(.*)@s) {
                    my ($fieldname, $escaped_path) = ($1, $2);
                    $escaped_path =~ s@\\\\@/@g; #(my $escaped_path = $2) =~ s@\\\\@\\@g;
                    $line = "<$fieldname>$escaped_path";
                } 
                elsif($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. 
            # index\col.gdb uses double backslashes, and single for \n,\t
            #$$win_text =~ s@\\\\@/@g;
        }
        else { # archiveinf gdb file
        
            # slashes in windows metadata text need to be turned into linux style slashes. 
            # In the two archivesinf gdb files, filepaths may use single backslashes
            $$win_text =~ s@\\@/@g; #$$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
    
    # The following block of code is necessary to deal with tmp (html) source files generated when using PDFBox
    # These tmpdirs are located inside the toplevel *greenstone* directory
    (my $gsdlhome_re = $ENV{'GSDLHOME'}) =~ s@\\@\/@g;      
    $gsdlhome_re = ".*" unless $$ENV{'GSDLHOME'};
    my $tmpfile_regex = "<URL>http://$gsdlhome_re/tmp/([^\.]*?)(\..{3,4})"; # $gsdlhome/tmp/randomfilename.html, file ext can be 3 or 4 chars long          
    if($test_text =~ m@$tmpfile_regex@g) {          
        # found a match, replace the tmp file name with "random", keeping the original file extension 
        # in <OrigSource|URL|UTF8URL|gsdlconvertedfilename>
    
        # This code is slightly different from doc.xml because each document has its own doc.xml, so this needs to be done
        # only once for doc.xml, but multiple times in index/col.gdb since it contains the random filenames of all docs in the col  
        #my ($old_tmp_filename, $ext) = ($1, $2);
        
        my $new_tmp_filename = "random";            
    
        
        $tmpfile_regex = "(<(URL|UTF8URL|gsdlconvertedfilename|OrigSource)>(http://)?)($gsdlhome_re)?(/tmp/)?.*?(\..{3,4})";
        if($5) { 
            $test_text =~ s@$tmpfile_regex@$1$5$new_tmp_filename$6@mg;
        } else { # OrigSource contains only the filename
            $test_text =~ s@$tmpfile_regex@$1$new_tmp_filename$6@mg;
        }
        
        # modelcol used a different gsdlhome, but also a tmp dir, so make the same changes to its random filename           
        $tmpfile_regex = "(<(URL|UTF8URL|gsdlconvertedfilename|OrigSource)>(http://)?)(.*)?(/tmp/)?.*?(\..{3,4})";
        if($5) { 
            $model_text =~ s@$tmpfile_regex@$1$5$new_tmp_filename$6@mg;
        } else { # OrigSource contains only the filename
            $model_text =~ s@$tmpfile_regex@$1$new_tmp_filename$6@mg;
        }
        
        # index/col.gdb also has entries for the random tmp file names in the form: [http://research/ak19/GS2bin_5July2013/tmp/F639.html]
        # need to equalise these also. Sadly, when there are multiple intermediate files, their random tmp filenames are not 
        # guaranteed to be generated in the same (alphabetical/numerical) order between model and test collection, so the 
        # HASH OIDs, although all of them accounted for, appear in a different order. So we have to remove the Hash OIDs.
        #$test_text =~ s@\[http://.*/tmp/.*(\..{3,4})\]@tmp/random$1@mg; # HASH OIDs can appear in different order
        #$model_text =~ s@\[http://.*/tmp/.*(\..{3,4})\]@tmp/random$1@mg;
        
        $test_text  =~ s@\[http://[^\n]*?/tmp/.*?(\..{3,4})\]\n<section>([^\n]*?)\n@[tmp/random$1\n<section>$2]\n@sg;
        $model_text =~ s@\[http://[^\n]*?/tmp/.*?(\..{3,4})\]\n<section>([^\n]*?)\n@[tmp/random$1\n<section>$2]\n@sg;
        
        # need to re- sort the keys, now that the absolute paths to tmp locations has been removed
        # so that we get the tmp files in the same order in both model and test collections
        
        # http://stackoverflow.com/questions/1909262/how-can-i-pipe-input-into-a-java-command-from-perl
        open PIPE, "| txt2db model.gdb";
        print PIPE "$model_text";
        close(PIPE);
        open PIPE, "| txt2db test.gdb";
        print PIPE "$test_text";
        close(PIPE);
        
        $model_cmd = " db2txt -sort model.gdb 2>&1";
        $test_cmd  = "db2txt -sort test.gdb 2>&1";
        $model_text = readin_gdb($model_cmd);
        $test_text = readin_gdb($test_cmd);
    }   

    # 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
    
#   return ($db_contents =~ m/\\/) ? 1 : 0; # windows slashes detected. Better test would be: [Something\something] OR <tag>something\something 
    
    if($dbtailname =~ m/^archiveinf-doc/) {
        return ($db_contents =~ m@<src-file>[a-zA-Z]:\\@) ? 1 : 0; # <src-file>C:\path
    } 
    elsif($dbtailname =~ m/^archiveinf-src/) { # <src-file>C:\path
        return ($db_contents =~ m@\[[a-zA-Z]:\\@) ? 1 : 0; # [C:\path]
    } 
    else { # index/col.gdb file
        if ($db_contents =~ m@<URL>http://[a-zA-Z]:/@) { # <URL>http://C:/path
            return 1;
        }       
        elsif ($db_contents =~ m@^(<URL>http://[a-zA-Z]:/)|(<null_file>[^\\]*\\)@m) { # <URL>http://C:/path OR <null_file>CMSwp-all.00000001\\00000035.nul          
            return 1;
        }
        elsif ($db_contents =~ m@^(<ex.File.Directory>[a-zA-Z]:\\\\)@m) { # <ex.File.Directory>C:\\path\\path for OAI collection
            return 1;
        }
        return 0;       
    }   
}

1;