-- Based on checks and defaults used by various testing frameworks (primarily GTest and Catch)
-- Moved the header into detail where it should have been to start with (oops)
- Nested unordered_map was the cause of a significant memory/cell spike.
- added cell_reference hash, lookup by cell_reference to make memory usage proportional to cell count only
- Cut time to write a sheet with many rows by not calling highest_row inside a loop over the rows (On^2 -> On)
- Observation: more memory is used / cell as the number of rows increases
- valid options are 11, 14, and 17
- default is 14
- cmake will error if an invalid value is provided
- requires cmake >= 3.10.* to take effect in visual studio
-- while unlikely to become an issue, ordering of ctors across source files is undefined and debugging issues related to it is not easy so just avoid that issue
- hyperlinks to cells and ranges are complete
- hyperlink::display is now set as well as the cell value (in excel these can be different)
-- if a cell is empty, display is equal to value text
-- if a cell has a value, display can be just about anything
- This version copies excel in that display is completely ignored once value is set
- All hyperlink tests are now part of the cell test suite (not the worksheet test suite which the majority were previously located)
* operator-- completes the naive bidirectional iterator requirements
* tests for various construction and assignment behaviours
* tests for iteration and dereferencing behaviours
Users may still want to dereference a const iterator (note: not a const_iterator).
Also use the "reference" typedef to ensure there is only 1 source of information
For rule of 5/0, where no implementation is required, all 5 operations have been declared as defaulted. This is less likely to forget definitions for all 5 if required
- removed forwarding of copy ctor to assignment (which was defaulted already) in favour of defaulted copy ctor
- added defaulted move assignment/ctor and destructor
Changed workbook reference to a pointer to allow tests to compile (reference isn't rebindable so defaulted assignment is equivalent to deleted)
Users may still want to dereference a const iterator (note: not a const_iterator).
Also use the "reference" typedef to ensure there is only 1 source of information
For rule of 5/0, where no implementation is required, all 5 operations have been declared as defaulted. This is less likely to forget definitions for all 5 if required
- removed forwarding of copy ctor to assignment (which was defaulted already) in favour of defaulted copy ctor
- added defaulted move assignment/ctor and destructor
Users may still want to derederence a const iterator (note: not a const_iterator).
Also use the "reference" typedef to ensure there is only 1 source of information
For rule of 5/0, where no implementation is required, all 5 operations have been declared as defaulted. This is less likely to forget definitions for all 5 if required
- removed forwarding of copy to assignment (which was defaulted already) in favour of defaulted copy ctor
- added defaulted move assignment/ctor and destructor
workbook::operator== was comparing the value of the raw pointer held by two std::unique_ptr's. By definition, this is always false in a well behaved program (if it's true, things go bang...). This then led to adding equality operators to nearly every other struct/class in xlnt to support workbook::operator==
workbook::load and the non-default ctors for loading data from a file are tested using the now functional equality operator
NOTE: a large number of copy ctors need updates/fixing. Many should be defaulted
Issue #298
- all 4 are simply duplicating existing behaviour, but perhaps we can get a more optimal version in future
- istream ctor is intended as an extension point that can then be used to create free/static functions to work with any future data source (vector), while the path ctor is a convenience function for the common case (from file)
-- roundtripping namespaces requires modifications of the parser used for the entire worksheet as there does not appear to be a way to track the namespace changes without listening/registering for the event. This (ofcourse) breaks lots of other things...
