[osmosis-dev] pbf problems under windows

[Scott Crosby]

On Tue, Dec 14, 2010 at 10:33 AM, Frederik Ramm <frederik at remote.org> wrote:
> Hi,
>
>   now I have the files ready for your inspection, on
>
> http://www.geofabrik.de/tmp/
>
> MD5 sums:
>
> 9d278ee963925ed34701431b8573e0bf  europe-created-on-linux.osm.pbf
> c4186d9197225e97f5bc935be09881e8  europe-created-on-windows.osm.pbf
>
> file sizes:
>
> 4448335290 2010-12-14 14:24 europe-created-on-linux.osm.pbf
> 4448335278 2010-12-14 11:38 europe-created-on-windows.osm.pbf

I've run a test over all of the files, computing a CRC of each block,
and prettyprinting every block that differs. The two files are
identical except for the reported osmosis version:

-Block      0 (OSMHeader) = 100267638
+Block      0 (OSMHeader) = 921813027
 bbox {
   left: -32733609999
   right: 45000000000
   top: 75000000000
   bottom: 31093990000
 }
 required_features: "OsmSchema-V0.6"
 required_features: "DenseNodes"
-writingprogram: "SNAPSHOT-r24507"
+writingprogram: "0.38"
 source: "http://www.openstreetmap.org/api/0.6"

 Block      1 (OSMData) = 3592976518
 Block      2 (OSMData) = 2353804896
 Block      3 (OSMData) = 257915217

///

If you skip the first 150 bytes of -linux and the first 138 bytes of
-windows, the two files *are* identical.

crosby at functor:/mnt/map$ tail -c +150 europe-created-on-linux.osm.pbf | md5sum
89a13dad1c8f81d802214d7d7bbb1871  -

crosby at functor:/mnt/map$ tail -c +138 europe-created-on-windows.osm.pbf | md5sum
89a13dad1c8f81d802214d7d7bbb1871  -

///

I'm at a loss. It doesn't seem possible that the windows-generated
file would work and the linux-generated file wouldn't.

I want to think the problem is some form of non-deterministic failure
in the windows JVM; bigger files take longer and are more likely to
hit the bug.

Does the same breakage happen of pbf2osm is run on the same file in windows?

Scott