Improve read and write performance umbrella issue

[tfussell]

There have been several users reporting performance problems over the years. I believe the current data structure for cells involving a vector of structs containing cell data and a hash map of cell references pointing to those cell structs is the cause of these problems. Some ideas for improving it:

• Make cell structs smaller
• Improve cache locality by storing cells in contiguous memory
• Use different data structures during serialization/deserialization and the intermediate representation in memory
• Optimize style handling, reference counting, and garbage collection
• Reuse cell data for cells that are identical
• Optionally disable slow features like styles
• Use multiple data structure depending on hints given by the user: prefer row-major or column-major traversal, sparse vs. dense matrix, insert speed vs lookup speed, number of cells

The solution will require deciding on which factors are most important: read speed, write speed, insert speed, etc, writing benchmarks for these factors, and implementing it in the code.

[Blakjak88]

Are there any unintentional copies of the data structures being made in the code that could be quickly eating memory and ultimately slowing stuff down?

As I have never come across such poor performance so far in my use of xlnt, do you have an example of a such an xlsx file that I could use to take a look at the code? Or a method to generate one?