Package: firebird2.0 Version: 2.0.0 Severity: serious I was going to use "grave", but there is no really danger of data loss, only very hard to detect, ugly and surprising incompatibility with upstream's binary database files, possibly leading to incorect query results. I (as maintainer) certainly don't want such incompatible version to be released as "stable".
The thread discussing this on upstream mailing list starts at http://sourceforge.net/mailarchive/forum.php?thread_name=200710311235.13978.peshkoff%40yaroslavl.ru&forum_name=firebird-devel (approximatelly. SF mailing list web archives suck) In short, from version 2.0 Firebird uses ICU for collation. Collation means representing any given string via a binary sequence that can be easily compared (byte-by-byte). This binary representation is used in Firebird indices, in the database file. >From time to time, ICU developers find errors in the binary representations and fix them. The problem is that Debian packages of firebird2.0 use the system-wide ICU (3.6), while upstream uses an older version (3.0), bundled together with Firebird sources. So any database created by Debian packages that uses unicode collations would contain in its indices binary representations that are silently incompatible with upstream. For Debian, I'll fix this by using the bundled ICU again. This means alwo enabling rpaths. Ugly, but it will work. Users will have to backup and restore their databases. Upstream, on the other hand implemented a better approach for the next versions of Firebird (2.1+) - each collation remembers with which version of ICU it was created and any attempt to use it with incompatible ICU version results in an error. Backup&restore agains solves the incompatibility. this approach will again allow Debian packages to use system-wide ICU without uncaught incompatibilities. Instead of backup&restore, a more fine-grained method would be to re-create the affected indices: "alter index $index_name active" should do it. -- dam -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
