Digging in a bit deeper, I figured two things: While the new database format is SQLite3, the old was some binary storing VCARDs which where clearly readable. So I came up with a small PHP snippet:

#!/usr/bin/php

<?php

# -=[ UserConf ]=-

$evodb = 'addressbook.db'; // (path and) name to the old address database

$name  = 'business';       // name of the address book (used as filename)

$separate_vcards = FALSE;  // if TRUE, each VCARD will go to a separate file



#-=[ Do not touch below lines ]=-

$re_vcard = '!(BEGIN:VCARD.*?END:VCARD)!ims';



$evo = file_get_contents($evodb);

preg_match_all($re_vcard,$evo,$vcards);

$cards = count($vcards[1]);

$new = '';

for ($i=0;$i<$cards;++$i) {

  if ($separate_vcards) {

    file_put_contents("${name}_${i}.vcf",$vcards[1][$i]);

  } else {

    $new .= $vcards[1][$i] . "



";

  }

}

if (!$separate_vcards) file_put_contents("${name}.vcf",$new);

?>

This is not 100% perfect, but better than losing all those data. The script runs through the old-style database, and tries to fetch all VCARDs stored there. Those are either exported to separate files (each VCARD in one file, if $separate_vcards = TRUE), or to one collection (otherwise).

As those are supposed to be plain-text files, one can check them and correct them easily (line-breaks, binary garbage) -- and finally using Evolutions Import-Feature (found in the "File" menu) to import the data ("single file") into an existing addressbook (create an empty one before if you want a new addressbook).

For me this restored about 90% of my addresses fine. For the remaining 10%, some where mixed-up or broken (remember, even the old format was binary -- and the snippet just works in "ASCII mode").

Hope this helps somebody else (if so, don't forget to up-vote this solution).