[gdal-dev] Trying to debug a GDAL crash in "GDALClose(hDataset); "

[Joaquim Luis]

Even,

Thanks, but this was another thing. I fixed a bug in GMT where under  
certain circumstances we were doing some overflowing access to the  
contents of the 'tmp' variable returned by

		GDALRasterIO(hBand, GF_Read, xOrigin, yOrigin, nXSize, nYSize, tmp,
		             nBufXSize, nBufYSize, GDALGetRasterDataType(hBand), 0, 0);


but this whole issue might not be all over yet.

Joaquim

> Le dimanche 15 novembre 2015 01:41:12, Joaquim Luis a écrit :
>> Ok I went ahead I tried it myself. It doesn't crash on OSX
>
> Perhaps it is an issue similar to
> https://trac.osgeo.org/gdal/ticket/5829 ?
>
> Several workarounds :
> - make sure all datasets are closed with GDALClose() before process
> termination (cleanest solution)
> - set the env variable GDAL_DESTROY to NO to prevent the termination  
> routine
> of GDAL from run and attempting to close any remaining opened dataset
>
>>
>> > My bad, I don't know how to produce a stack trace from within the VS
>> > debugger. I have it all visually but find no option to save it as a  
>> file.
>> >
>> > I added a breakpoint at  cpl_atomic_ops.cpp#L56 but it didn't stop  
>> there
>> > after the call to   GDALClose(hDataset);Actually when stepping in  it
>> > jumps directly from GDALClose() into CPLAtomicCompareAndExchange() but
>> > once there (due to crash) I can step back >to other functions.
>> > Sorry if it's not clear.
>> >
>> > Do you have a GMT build with GDAL svn? Could you please try if this
>> > crashes for you
>> >
>> > grdmath -Rd -I1 X = lixo.tiff:GTiff
>> > grdinfo -R0/20/30/50 lixo.tiff
>> >
>> >> nLockCount should be set by
>> >>
>> >> https://trac.osgeo.org/gdal/browser/trunk/gdal/gcore/gdal_priv.h#L598
>> >>
>> >>  
>> https://trac.osgeo.org/gdal/browser/trunk/gdal/port/cpl_atomic_ops.cpp#L
>> >> 56
>> >>
>> >> which uses InterlockedCompareExchange to change it.
>> >>
>> >> Can you give the whole stack trace?
>> >>
>> >> On Fri, Nov 13, 2015 at 6:47 PM, Joaquim Luis <jluis at ualg.pt> wrote:
>> >>> Hi,
>> >>>
>> >>> I'm having strange crashes in GDAL (SVN) when called via GMT. The
>> >>> crashes occur at a call to
>> >>>
>> >>> GDALClose(hDataset);
>> >>>
>> >>> Now, this is only occurs when I try to read a sub-region of a grid  
>> but
>> >>> those are GMT details.
>> >>> It doesn't happen to colleagues on OSX but they are using gdal 1.11.
>> >>> Can't say that's the reason but this used to work for me as well on
>> >>> Windows >>>(I should know because I wrote that code).
>> >>>
>> >>> I built a GDAL debug and can see that the crash happens in
>> >>> cpl_atomic_ops.cpp
>> >>>
>> >>> int CPLAtomicCompareAndExchange(volatile int* ptr, int oldval, int
>> >>> newval)
>> >>> {
>> >>>
>> >>>  return (LONG)InterlockedCompareExchange((volatile LONG*)(ptr),
>> >>>
>> >>> (LONG)newval, (LONG)oldval) == (LONG)oldval;
>> >>> }
>> >>>
>> >>> because the 'ptr' is invalid. That function is called by
>> >>> gdalrasterblock.cpp
>> >>>
>> >>> GDALRasterBlock::DropLockForRemovalFromStorage()
>> >>>
>> >>>        if( CPLAtomicCompareAndExchange(&nLockCount, 0, -1) )
>> >>>
>> >>> and the debugger tells me that
>> >>>
>> >>>        nLockCount      <Unable to read memory>
>> >>>
>> >>> and here I'm lost. Where is nLockCount supposed to be set?
>> >>>
>> >>> Thanks
>> >>>
>> >>> Joaquim
>> >>> _______________________________________________
>> >>> gdal-dev mailing list
>> >>> gdal-dev at lists.osgeo.org
>> >>> http://lists.osgeo.org/mailman/listinfo/gdal-dev
>> >>
>> >> ----
>> >> http://schwehr.org