Take a look here: https://github.com/mariadb-corporation/mariadb-columnstore-engine/blob/develop/writeengine/wrapper/writeengine.cpp#L4559 (function int WriteEngineWrapper::writeColumnRec).

There are two branches there, one for split write and other for single piece write. The code is quite repetetive, it repeats itself three times: first and second part of split write and the same code goes in single write.

Ideally, we would have two branches: code for first part of split write and single piece code is exactly the same, up to some fixes. We have to have then process second part in split write.

we_colop.h contains more or less identical code in three functions:

    /**

     * @brief Write row(s)

     */

    EXPORT virtual int writeRow(Column& curCol,

                                uint64_t totalRow,

                                const RID* rowIdArray,

                                const void* valArray,

                                      void* oldValArray =0,

                                bool bDelete = false);

    /**

     * @brief Write row(s) for delete  @Bug 1886,2870

     */

    EXPORT virtual int writeRows(Column& curCol,

                                 uint64_t totalRow,

                                 const RIDList& ridList,

                                 const void* valArray,

                                       void* oldValArray = 0,

                                 bool bDelete = false);

    /**

     * @brief Write row(s) for update @Bug 1886,2870

     */

    EXPORT virtual int writeRowsValues(Column& curCol,

                                       uint64_t totalRow,

                                       const RIDList& ridList,

                                       const void* valArray,

                                             void* oldValArray = 0);

These should be unified.

ColStruct in we_type.h (writeengine/shared) does not have proper strict constructor which will set internal strctures right.

For example, in writeengine.cpp there are several calls to Convertor::convertColType(ColStruct* colStruct) function which set colType field using CalpontSystemCatalog tyle field and column width.

The problem is that it is normally fine to assume that all fields are correct - that constructed record is all good and fine. But it is not. That means if your code depends on the write engine's internal type, you have to check whether it is set properly. And if not, you have to call conversion code. And then code after your change can call the same conversion code (due to legacy I describe) and duplicate work for nothing.