[gdal-dev] OGRLayer::WriteArrowBatch() and CreateFeature() mutually exclusive in the same session

[Even Rouault]

Chris,
> I'm trying to understand implications of OGRLayer::WriteArrowBatch():
>
>> This method and CreateFeature() are mutually exclusive in the same session.
> The context is binding code where the user could potentially call
> either of these methods at any point during the session.
>
> Does "session" refer to the layer or dataset instance, or the GDAL
> session itself?
Dataset and layer instances
>
> In tests with GPKG, using WriteArrowBatch() followed by
> CreateFeature() worked and didn't cause any obvious problems.
yes, because for that driver (actually all drivers except 
Parquet&Arrow), WriteArrowBatch() results in CreateFeature() behind the 
hood. But you'd likely hit issues on Parquet or Arrow datasets, for 
which implementation limitations lead to the general restriction.  And 
as we have nothing in the test suite that checks that mixing 
WriteArrowBatch() & CreateFeature() works even for code where it is OK 
currently, you could be exposed to potential regressions if the 
implementation changed in the future.
>   Should
> binding code guard against this in general?
I'd say up to you. The SWIG bindings don't for example.
>
> Similarly, CreateFieldFromArrowSchema() and CreateField() are mutually
> exclusive in the same session. Is there any issue with using
> CreateField() and WriteArrowBatch() in the same session?

Same as above w.r.t Arrow/Parquet driver. The feature writing logic 
between the Arrow and per-feature strategies takes different code paths. 
I don't say reconciling them would be impossible, but clearly an 
implementation complication as I didn't see the point in supporting 
mixed workflows.

Even


-- 
http://www.spatialys.com
My software is free, but my time generally not.