[OSM-dev] Osmosis data corruption on Debian Jessie/Testing

[Brett Henderson]

I suspect that attempting to detect the underlying XML runtime would be
brittle.  Another option might be to embed that bit of data in Osmosis
itself and do a self test before attempting to execute any XML tasks.

I'm surprised that this is still an issue in standard Java.  I tried
raising tickets against Sun Java before it moved under Oracle but never got
a response.  I gave up, embedded Xerces in the main Osmosis distribution,
and then forgot about it.

On 5 March 2015 at 10:16, Jochen Topf <jochen at remote.org> wrote:

> Hi!
>
> Just spent a few hours debugging this problem: The way Osmosis is packaged
> on Debian Jessie seems to be wrong. It doesn't use the Xerces XML parser
> but seems to fall back to Java default XML parser which mangles Unicode
> characters.
>
> This can lead to data corruption (and has for me today) when using Osmosis
> for planet updates etc.
>
> You can test whether this bug is on your system, too: Download the XML
> for this node: http://www.openstreetmap.org/node/3382756758. Then run
> it through osmosis:
>
>     osmosis --rx 3382756758.osm --wx out.osm
>
> Compare the two files, you'll see the musical notation character doubling
> in the second case when your Osmosis is broken. The fix is simple: Add
> a line "load /usr/share/java/xercesImpl.jar" to /etc/osmosis/plexus.conf.
> As I understand this, it tells Java to load Xerces replacing the built-in
> XML parser.
>
> I have opened a bug with Debian.
>
> Arguably Osmosis should somehow detect when Xerces isn't found and return
> an
> error instead of using a different implemenation. But I don't know enough
> about
> Java to say whether thats possible.
>
> Jochen
> --
> Jochen Topf  jochen at remote.org  http://www.jochentopf.com/
> +49-173-7019282
>
> _______________________________________________
> dev mailing list
> dev at openstreetmap.org
> https://lists.openstreetmap.org/listinfo/dev
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstreetmap.org/pipermail/dev/attachments/20150305/5811eda9/attachment.html>