[gdal-dev] Could vector reproject be affected if alternative BLAS/LAPACK implementations are in use?

[Even Rouault]

Chris,

BLAS/LAPACK is not a direct dependency of GDAL nor PROJ.  BLAS is a 
dependency of libarmadillo, which is an optional dependency of GDAL, but 
not involved in vector reprojection. I'd suggest :

- running ldd libgdal.so and check there are not 2 PROJ versions linked

- running under Valgrind to see if there is something interesting showing up

- running under gdb and showing the stack trace where it crashes

Even

Le 14/08/2025 à 00:29, Chris Toney via gdal-dev a écrit :
> Hi,
> Is it possible that vector reproject (via `GDALVectorTranslate()`)
> could be affected if alternative BLAS/LAPACK implementations are
> present (i.e, from Intel MKL or OpenBLAS)?
>
> "ERROR 1: Reprojection failed" may originate from
> https://github.com/OSGeo/gdal/blob/e48cc4f31f43cf872c036d2c4e65e7368166c34a/apps/ogr2ogr_lib.cpp#L6378.
> This is followed by seg fault.
>
> The error is reported with GDAL 3.10.3, GEOS 3.13.0, PROJ 9.6.2 on
> Fedora 42. The question is motivated by the issue at:
> https://github.com/USDAForestService/gdalraster/issues/769.
>
> The same code/inputs test without issue on a wide range of different
> platforms in several CI and package management scenarios. The error
> message and seg fault appear to only occur (sometimes not always) when
> the alternative BLAS/LAPACK are present.
>
> The code that generates the error is at:
> https://github.com/USDAForestService/gdalraster/blob/d7c620b74a987cbc3f96dec245d37809e09f858a/src/gdal_exp.cpp#L1791
>
> Input and output datasets are GPKG and reprojection is from NAD83 GCS
> (EPSG:4269) to EPSG:32100.
>
> Thanks.
>
> Chris Toney
> _______________________________________________
> gdal-dev mailing list
> gdal-dev at lists.osgeo.org
> https://lists.osgeo.org/mailman/listinfo/gdal-dev

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