Nightly build crash when runing some GMT tests that call functions in the GDAL lib

[joa-quim]

Hi,

This is not very easy to demonstrate but I' try. Julia nightly now (since some weeks, don't know more) crash when I run the GMT.jl tests.
Digging around I could find a patch that simply consist in accessing the pointer before sending it to the crashing ccall call. An example of such is provided in this line . With that line commented, on Windows I can reproduce the crash with

julia> ds_point = GMT.Gdal.read(joinpath(dirname(pathof(GMT)), "..", "test")*"\\point.geojson")
GDAL Dataset (Driver: GeoJSON/GeoJSON)
File(s):
  C:\Users\joaqu\.julia\dev\GMT\src\..\test\point.geojson

Number of feature layers: 1
  Layer 0: point (Point)


julia> layer = GMT.Gdal.getlayer(ds_point, 0);

julia> layer
Layer: point (1)
  Geometry 0 (`1�f��): [1], POINT (100 0), POINT (100.2785 0.0893), ...

Please submit a bug report with steps to reproduce this fault, and any error messages that follow (in their entirety). Thanks.
Exception: EXCEPTION_ACCESS_VIOLATION at 0x7fff66fa1440 -- OGR_Fld_GetType at C:\programs\compa_libs\DLLs_VC14_64\gdal_w64.dll (unknown line)
in expression starting at none:0
OGR_Fld_GetType at C:\programs\compa_libs\DLLs_VC14_64\gdal_w64.dll (unknown line)
Allocations: 6857399 (Pool: 6853216; Big: 4183); GC: 8

Uncomment that line and the above works. But this is not a Windows only problem. This linux nightly build failed for the same reason, whilst this one with julia 1.7 completed successfully (the linux build failed in this run but for a totally different reason).

[jmkuhn]

I think this is the first commit that displays this issue is: 62ac26a #46651

[vtjnash]

There are no @GC.preserves in that file, so the GC is allowed to delete the object as soon as it concludes that the only field used is .ptr, and thus it (correctly) deletes the object immediately (as instructed)

[joa-quim]

Thanks. Very helpful.