[gdal-dev] Advanced 3.6.1 release and retraction of 3.6.0 ?

[Even Rouault]

Hi,

playing lately with more complex scenarios with ogr2ogr, that time 
involving both a lot of features and multiple layers being converted at 
once to GeoPackage, I've seen error messages being emitted during 
conversions and some slowness due to SQLite3 self-locking itself for 
periods of a few seconds. This is still due to the multithreaded RTree 
implementation. I didn't notice data corruption though, but I won't 
swear it couldn't happen.

My conclusion was that is isn't safe to enable it when multiple layers 
are created in a GeoPackage file. I've thus disabled it in such 
scenarios per https://github.com/OSGeo/gdal/pull/6959. So in the 
meantime setting the OGR_GPKG_ALLOW_THREADED_RTREE environment 
variable/configuration option to NO is still a recommended workaround 
(if you write multiple layers in the same GeoPackage file). I'll 
probably issue a 3.6.2 release at the beginning of next year when 
everyone is back from their break.

Even

Le 13/12/2022 à 16:18, Even Rouault via gdal-dev a écrit :
> Hi,
>
> https://github.com/qgis/QGIS/issues/51188 has been brought to my 
> attention. The issue is that the new background building of the RTree 
> of GeoPackage files introduced in 3.6.0 didn't work well with 
> committing transactions in between, which is easily triggered by 
> ogr2ogr. All features were inserted but with default settings ~ 10% of 
> them lacked an entry in the spatial index, which can be enough to 
> break interactive display and workflows relying on spatial filtering. 
> When using ogr2ogr, I believe the issue can only be seen when creating 
> layers with more than 100 000 features, since that's the default value 
> for the interval at which transactions are committed.
>
> I've a fix ready in https://github.com/OSGeo/gdal/pull/6911. The fix 
> itself is a simple two-liner: 
> https://github.com/OSGeo/gdal/pull/6911/commits/3f5f6225fe82e0c2e0241e4f66bfb861cdf4fe9d
>
> Given the status of GeoPackage being the default format for QGIS, I 
> believe this is a severe enough issue to warrant an advanced 3.6.1 
> release, and an official retraction of 3.6.0.
>
> Thoughts ?
>
> Even
>
-- 
http://www.spatialys.com
My software is free, but my time generally not.