Change Details

Running the to the SoA format (T95965) when reading meshes means that none of the changes from versioning would be applied first. This means important fixes like rBf14995aba70a aren't applied first, so modifications could be done to invalid `CustomData`. One option is to just move all changes to `CustomData` to mesh reading code, but that doesn't align with the purpose of versioning, and adds unnecessary _permanent_ complexity. Instead, just run the mesh versioning in the regular versioning file. Unlike other versioning, we won't put this behind a subversion bump until 4.0. Another option is structuring the change more like `ntreeUpdateAllNew`, but IMO that doesn't fit well with the fact that this is just versioning where we also happen to write with the old format for the time being. --- I noticed this problem when working on D15795. To reproduce, check out `refactor-mesh-selection-generic` and run `ctest -R physics_cloth`.

Converting to the SoA format (T95965) immediately when reading meshes means that none of the changes from versioning would be applied first. This means important fixes like rBf14995aba70a aren't properly applied, so modifications could be done to invalid `CustomData`. One option is to just move all changes to `CustomData` to mesh reading code, but that doesn't align with the purpose of versioning, and adds unnecessary _permanent_ complexity. Instead, just run the mesh versioning in the regular versioning file. Unlike other versioning, we won't put this behind a subversion bump until 4.0. Another option is structuring the change more like `ntreeUpdateAllNew`, but IMO that doesn't fit well with the fact that this is just versioning where we also happen to write with the old format for the time being. --- I noticed this problem when working on D15795. To reproduce, check out `refactor-mesh-selection-generic` and run `ctest -R physics_cloth`.

Running theConverting to the SoA format (T95965) immediately when reading meshes means that none of the changes from versioning would be applied first. This means important fixes like rBf14995aba70a aren't applied firstproperly applied, so modifications could be done to invalid `CustomData`. One option is to just move all changes to `CustomData` to mesh reading code, but that doesn't align with the purpose of versioning, and adds unnecessary _permanent_ complexity. Instead, just run the mesh versioning in the regular versioning file. Unlike other versioning, we won't put this behind a subversion bump until 4.0. Another option is structuring the change more like `ntreeUpdateAllNew`, but IMO that doesn't fit well with the fact that this is just versioning where we also happen to write with the old format for the time being. --- I noticed this problem when working on D15795. To reproduce, check out `refactor-mesh-selection-generic` and run `ctest -R physics_cloth`.