Set CMAKE_POLICY_VERSION_MINIMUM to 3.5 in the CMakePresets.json file in order to build with CMake 4.0. This is a workaround until we can fix the problems caused by the manylinux docker images being updated. There are two problems, jsoncpp which can be fixed by building as an external project instead of add_subdirectory (and specifying the CMAKE_POLICY_VERSION_MINIMUM for the configure step of the external project) and the CMakeRC resource compiler used for some of the help text which probably have to be patched on download in order to fix it.

Fix a very rare problem for 4D data where we need to eliminate layers that have 4D chunks because they cannot be serialized. This happens when you have a trace-mapped channel with more than one mapped value (e.g. SEGYTraceHeader) and a layer with a dimensiongroup that doesn't include dimension 0 (only Dimensions_123 for 4D data). Update GetBaseLayer/GetTopLayer so they work also when a layer has been eliminated.

Clarify how the unit of the Sample axis descriptor is used to discriminate between time and depth domain data.

Make sure we destroy the VolumeDataPageAccessors created by the VolumeDataRequestProcessor when the VDS is closed.

Make sure we include the Curl backend if needed by any enabled IOManager. Change the DISABLE_CURL_IOMANAGER to effectively work like 'DISABLE_HTTP_IOMANAGER' which is what it really should have been from the start.

Alena Chaikouskaya authored 5 months ago and Morten Ofstad committed 3 months ago

When DISABLE_AZURE_PRESIGNED_IOMANAGER / DISABLE_CURL_IOMANAGER are set
to on, Debug build fails while Release builds fine.

This seems to be caused by LTO being enabled by default for Release
builds and disabled for Debug ones.

The actual issue is that additional source files have to be
marked as "header file only" for linking to succeed.

Alena Chaikouskaya authored 5 months ago and Morten Ofstad committed 3 months ago

AWS compilation and compilation of IOManagerDmsFactory started to depend
on IOManagerCurl after DMS Factory code introduction, but the setting
was not updated.

Also the description is updated to better reflect purpose of the
setting.

Ignore traces with primary/secondary keys that are inconsistent with the segment they belong to or unsorted compared to the other traces in the segment. This avoids out-of-bounds access when trying to put them into the volume data chunks.

Add dependency on Python targets being built to the doc generation process to ensure it does not start building the docs before the python module can be imported by conf.py.

Reset the written region and clear the copied to chunk list when a page is written back. Check for m_layer being null in LimitPageListSize() so it doesn't crash when a VDS that has dirty pages is closed.

Johan Simon Seland authored 7 months ago and Morten Ofstad committed 7 months ago

This makes OpenVDS combileable with Visual Studio 2022 17.11.
This version of Visual Studio triggers errors due to the use of (nonstandard) checked iterators.
Consequently, use of these were removed in later versions of fmt.

Use an older implementation of the register_exception from pybind11 when compiling against Python 3.6 to work around the new implementation hanging when used with Python 3.6.

Change the defaultMaxHostConnections constant defintion to an anonymous enum to avoid linker problems.

Lower the max number of host connections to 32 which was the old default. Renamed maxConcurrentRequests parameter to maxHostConnections to reflect what it actually does.

Add uploads/downloads directly to the processingRequests queue and use the CURLMOPT_MAX_HOST_CONNECTIONS feature to limit the number of connections instead of having a queuedRequests queue in the CurlHandle. This simplifies the implementation and avoids a case where requests were not moved from the queuedRequests to the pendingRequests when there were cancellations.

Add a new ConfigureThreadCount() helper function in ThreadPool.h and use it to make the thread count for the IOManagerAzureSdkForCpp configurable via the OPENVDS_AZURE_IOMANAGER_THREAD_COUNT environment variable.

Fix the parameters passed to the Python wrapper for requestProjectedVolumeSubset so they match the native method.

Remove broken compare function from VolumeDataChunk and change pendingDownloadRequests to be an unordered_map instead of map. Add a unique identifier to the VDS handle and use it to make the VolumeDataChunk hash function robust (if we were to make a data structure that contained chunks from different VDSs).

Check that the number of components in the serialized data is valid (matches the number of components of the layer) before deserializing.

Update documentation to make it clear how units are expected to be used in VDS metadata and how to handle the situatioen where the unit of the CRS differs from the unit of the survey coordinates.

Update curl to 7.86.0 to avoid a crash with how the Azure SDK re-uses CONNECT_ONLY connections. Update Azure SDK to 1.11.3 that includes the patch we were previously applying manually.