OpenXLSX is a C++ library for reading, writing, creating and modifying Microsoft Excel® files, with the .xlsx format.
- Motivation
- Ambition
- Compatibility
- Build Instructions
- Current Status
- Performance
- Caveats
- Reference Guide
- Example Programs
- Changes
Many programming languages have the ability to modify Excel files, either natively or in the form of open source libraries. This includes Python, Java and C#. For C++, however, things are more scattered. While there are some libraries, they are generally less mature and have a smaller feature set than for other languages.
Because there are no open source library that fully fitted my needs, I decided to develop the OpenXLSX library.
The ambition is that OpenXLSX should be able to read, write, create and modify Excel files (data as well as formatting), and do so with as few dependencies as possible. Currently, OpenXLSX depends on the following 3rd party libraries:
- PugiXML
- Zippy (C++ wrapper around miniz)
- Boost.Nowide (for opening files with non-ASCII names on Windows)
These libraries are all header-only and included in the repository, i.e. it's not necessary to download and build separately.
Also, focus has been put on speed, not memory usage (although there are options for reducing the memory usage, at the cost of speed; more on that later).
OpenXLSX has been tested on the following platforms/compilers. Note that a '-' doesn't mean that OpenXLSX doesn't work; it just means that it hasn't been tested:
| GCC | Clang | MSVC | |
|---|---|---|---|
| Windows | MinGW | MinGW | + |
| Cygwin | - | - | N/A |
| MacOS | + | + | N/A |
| Linux | + | + | N/A |
The following compiler versions should be able to compile OpenXLSX without errors:
- GCC: Version 7
- Clang: Version 8
- MSVC: Visual Studio 2019
Clang 7 should be able to compile OpenXLSX, but apparently there is a bug in the implementation of std::variant, which causes compiler errors.
Visual Studio 2017 should also work, but hasn't been tested.
OpenXLSX uses CMake as the build system (or build system generator, to be exact). Therefore, you must install CMake first, in order to build OpenXLSX. You can find installation instructions on www.cmake.org.
The OpenXLSX library is located in the OpenXLSX subdirectory to this repo. The OpenXLSX subdirectory is a self-contained CMake project; if you use CMake for your own project, you can add the OpenXLSX folder as a subdirectory to your own project. Alternatively, you can use CMake to generate make files or project files for a toolchain of your choice. Both methods are described in the following.
By far the easiest way to use OpenXLSX in your own project, is to use CMake yourself, and then add the OpenXLSX folder as a subdirectory to the source tree of your own project. Several IDE's support CMake projects, most notably Visual Studio 2019, JetBrains CLion, and Qt Creator. If using Visual Studio, you have to specifically select 'CMake project' when creating a new project.
The main benefit of including the OpenXLSX library as a source subfolder, is that there is no need to locate the library and header files specifically; CMake will take care of that for you. Also, the library will be build using the same configuration (Debug, Release etc.) as your project. In particular, this a benefit on Windows, where is it not possible to use Release libraries in a Debug project (and vice versa) when STL objects are being passed through the library interface, as they are in OpenXLSX. When including the OpenXLSX source, this will not be a problem.
By using the add_subdirectory() command in the CMakeLists.txt file for your project, you can get access to the
headers and library files of OpenXLSX. OpenXLSX can generate either a shared library or a static library. By default
it will produce a shared library, but you can change that in the OpenXLSX CMakeLists.txt file. The library is
located in a namespace called OpenXLSX; hence the full name of the library is OpenXLSX::OpenXLSX.
The following snippet is a minimum CMakeLists.txt file for your own project, that includes OpenXLSX as a subdirectory. Note that the output location of the binaries are set to a common directory. On Linux and MacOS, this is not really required, but on Windows, this will make your life easier, as you would otherwise have to copy the OpenXLSX shared library file to the location of your executable in order to run.
cmake_minimum_required(VERSION 3.15)
project(MyProject)
set(CMAKE_CXX_STANDARD 17)
# Set the build output location to a common directory
set(CMAKE_ARCHIVE_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR}/output)
set(CMAKE_LIBRARY_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR}/output)
set(CMAKE_RUNTIME_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR}/output)
add_subdirectory(OpenXLSX)
add_executable(MyProject main.cpp)
target_link_libraries(MyProject OpenXLSX::OpenXLSX)Using the above, you should be able to compile and run the following code, which will generate a new Excel file named 'Spreadsheet.xlsx':
#include <OpenXLSX.hpp>
using namespace OpenXLSX;
int main() {
XLDocument doc;
doc.create("Spreadsheet.xlsx");
auto wks = doc.workbook().worksheet("Sheet1");
wks.cell("A1").value() = "Hello, OpenXLSX!";
doc.save();
return 0;
}If you wish to produce the OpenXLSX binaries and include them in your project yourself, it can be done using CMake and a compiler toolchain of your choice.
From the command line, navigate the OpenXLSX subdirectory of the project root, and execute the following commands:
mkdir build
cd build
cmake ..
The last command will configure the project. This will configure the project using the default toolchain. If you
want to specify the toolchain, type cmake -G "<toolchain>" .. with <toolchain> being the toolchain you wish to
use, for example "Unix Makefiles", "Ninja", "Xcode", or "Visual Studio 16 2019". See the CMake documentation for
details.
Finally, you can build the library using the command:
cmake --build . --target OpenXLSX --config Release
You can change the --target and --config arguments to whatever you wish to use.
When built, you can install it using the following command:
cmake --install .
This command will install the library and header files to the default location on your platform (usually /usr/local/ on Linux and MacOS, and C:\Program Files on Windows). You can set a different location using the --prefix argument.
Note that depending on the platform, it may not be possible to install both debug and release libraries. On Linux and MacOS, this is not a big issue, as release libraries can be used for both debug and release executables. Not so for Windows, where the configuration of the library must be the same as for the executable linking to it. For that reason, on Windows, it is much easier to just include the OpenXLSX source folder as a subdirectory to your CMake project; it will save you a lot of headaches.
OpenXLSX is still work in progress. The following is a list of features which have been implemented and should be working properly:
- Create/open/save files
- Read/write/modify cell contents
- Copy cells and cell ranges
- Copy worksheets
- Cell ranges and iterators
- Row ranges and iterators
Features related to formatting, plots and figures have not been implemented, and are not planned to be in the near future.
It should be noted, that creating const XLDocument objects, is currently not working!
The table below is the output from a benchmark (using the Google Benchmark library), which shows that read/write access can be done at a rate of around 4,000,000 cells per second. Floating point numbers are somewhat lower, due to conversion to/from strings in the .xml file.
Run on (16 X 2300 MHz CPU s)
CPU Caches:
L1 Data 32 KiB (x8)
L1 Instruction 32 KiB (x8)
L2 Unified 256 KiB (x8)
L3 Unified 16384 KiB (x1)
Load Average: 2.46, 2.25, 2.19
---------------------------------------------------------------------------
Benchmark Time CPU Iterations UserCounters...
---------------------------------------------------------------------------
BM_WriteStrings 2484 ms 2482 ms 1 items=8.38861M items_per_second=3.37956M/s
BM_WriteIntegers 1949 ms 1949 ms 1 items=8.38861M items_per_second=4.30485M/s
BM_WriteFloats 4720 ms 4719 ms 1 items=8.38861M items_per_second=1.77767M/s
BM_WriteBools 2167 ms 2166 ms 1 items=8.38861M items_per_second=3.87247M/s
BM_ReadStrings 1883 ms 1882 ms 1 items=8.38861M items_per_second=4.45776M/s
BM_ReadIntegers 1641 ms 1641 ms 1 items=8.38861M items_per_second=5.11252M/s
BM_ReadFloats 4173 ms 4172 ms 1 items=8.38861M items_per_second=2.01078M/s
BM_ReadBools 1898 ms 1898 ms 1 items=8.38861M items_per_second=4.4205M/s
An .xlsx file is essentially a bunch of .xml files in a .zip archive. Internally, OpenXLSX uses the miniz library to compress/decompress the .zip archive, and it turns out that miniz has an upper limit regarding the file sizes it can handle.
The maximum allowable file size for a file in an archive (i.e. en entry in a .zip archive, not the archive itself) is 4 GB (uncompressed). Usually, the largest file in an .xlsx file/archive, will be the .xml files holding the worksheet data. I.e., the worksheet data may not exceed 4 GB. What that translates to in terms of rows and columns depend a lot on the type of data, but 1,048,576 rows x 128 columns filled with 4-digit integers will take up approx. 4 GB. The size of the compressed archive also depends on the data held in the worksheet, as well as the compression algorithm used, but a workbook with a single worksheet of 4 GB will usually have a compressed size of 300-350 MB.
The 4 GB limitation is only related to a single entry in an archive, not the total archive size. That means that if an archive holds multiple entries with a size of 4GB, miniz can still handle it. For OpenXLSX, this means that a workbook with several large worksheets can still be opened.
OpenXLSX uses the PugiXML library for parsing and manipulating .xml files in .xlsx archive. PugiXML is a DOM parser, which reads the entire .xml document into memory. That makes parsing and manipulation incredibly fast.
However, all choices have consequences, and using a DOM parser can also demand a lot of memory. For small spreadsheets, it shouldn't be a problem, but if you need to manipulate large spreadsheets, you may need a lot of memory.
The table below gives an indication of how many columns of data can be handled by OpenXLSX (assuming 1,048,576 rows):
| Columns | |
|---|---|
| 8 GB RAM | 8-16 |
| 16 GB RAM | 32-64 |
| 32 GB RAM | 128-256 |
Your milage may vary. The performance of OpenXLSX will depend on the type of data in the spreadsheet.
Note also that it is recommended to use OpenXLSX in 64-bit mode. While it can easily be used in 32-bit mode, it can only access 4 GB of RAM, which will severely limit the usefulness when handling large spreadsheets.
If memory consumption is an issue for you, you can build the OpenXLSX library in compact mode (look for the ENABLE_COMPACT_MODE in the CMakeLists.txt file), which will enable PugiXML's compact mode. OpenXLSX will then use less memory, but also run slower. See further details in the PugiXML documentation here. A test case run on Linux VM with 8 GB RAM revealed that OpenXLSX could handle a worksheet with 1,048,576 rows x 32 columns in compact mode, versus 1,048,576 rows x 16 columns in default mode.
By far the most questions I get about OpenXLSX on Github, is related to Unicode. It is apparently (and understandably) a source of great confusion for many people.
Early on, I decided that OpenXLSX should focus on the Excel part, and not be a text encoding/conversion utility also. Therefore, all text input/output to OpenXLSX MUST be in UTF-8 encoding... Otherwise it won't work as expected. It may also be necessary that the source code files are saved in UTF-8 format. If, for example, a source file is saved in UTF-16 format, then any string literals will also be in UTF-16. So if you have any hard-coded string literals in your source code, then the source file must also be saved in UTF-8 format.
All string manipulations and usage in OpenXLSX uses the C++ std::string, which is encoding agnostic, but can easily be used for UTF-8 encoding. Also, Excel uses UTF-8 encoding internally (actually, it might be possible to use other encodings, but I'm not sure about that).
For the above reason, if you work with other text encodings, you have to convert to/from UTF-8 yourself. There are a number of options (e.g. Boost.Nowide or Boost.Text). Internally, OpenXLSX uses Boose.Nowide; it has a number of handy features for opening files and converting between std::string and std::wstring etc. I will also suggest that you watch James McNellis' presentation at CppCon 2014, and read Joel Spolsky's blog.
Unicode on Windows is particularly challenging. While UTF-8 is well supported on Linux and MacOS, support on Windows is more limited. For example, output of non-ASCII characters (e.g. Chinese or Japanese characters) to the terminal window will look like gibberish. As mentioned, sometimes you also have to be mindful of the text encoding of the source files themselves. Some users have had problems with OpenXLSX crashing when opening/creating .xlsx files with non-ASCII filenames, where it turned out that the source code for the test program was in a non-UTF-8 encoding, and hence the input string to OpenXLSX was also non-UTF-8. To stay sane, I recommend that source code files are always in UTF-8 files; all IDE's I know of can handle source code files in UTF-8 encoding. Welcome to the wonderful world of unicode on Windows 🤮
An Excel-file is essentially just a bunch of .xml files wrapped in a .zip archive. OpenXLSX uses a 3rd party library to extract the .xml files from the .zip archive. The default library used by OpenXLSX is Zippy, which is an object-oriented wrapper around miniz. The miniz library is fast, and is header-only, which is ideal for OpenXLSX.
However, it is possible to use a different zip-library, if you want to. In rare cases, you may experience stability issues with miniz. In those cases, it may be useful to try a different zip-library.
Using the Zippy/miniz library requires no special efforts; it will work straight out of the box. Using a different zip-library, however, will require some work.
In order to use a different zip-library, you must create a wrapper class that conforms to the interface specified by the IZipArchive class. Note that this is implemented using type erasure, meaning that no inheritance is required; the class just needs to have a conforming interface, thats all. After that, provide an object of the class and supply it to the OpenXLSX constructor.
To see an example of how this is done, take a look at Demo1A in the Examples folder. This example uses a class called CustomZip (using libzip as the zip library) which can be found under Examples/external/CustomZip. In order to build the example program, make sure that libzip (and it's dependencies) is installed on your computer, and enable the OPENXLSX_ENABLE_LIBZIP option in the CMakeLists.txt file in the OpenXLSX root.
As mentioned, the Demo1A example program uses libzip. libzip is a very stable library and widely used. However, my experience is that it is quite slow for large zip files, such as large spreadsheets. For that reason, libzip may not be the ideal solution, but it is useful for showing how a different zip library can be used.
In the 'Examples' folder, you will find several example programs, that illustrates how to use OpenXLSX. Studying those example programs is the best way to learn how to use OpenXLSX. The example programs are annotated, so it should be relatively easy to understand what's going on.
OpenXLSX can now use other zip libraries than the default Zippy/miniz library. See Demo1A as an example of how it's done
This version includes row ranges and iterators. It also support assignment of containers of cell values to XLRow objects. This is significantly faster (up to x2) than using cell ranges or accessing cells by cell references.
The internal architecture of OpenXLSX has been significantly re-designed since the previous version. The reason is that the library was turning into a big ball of mud, and it was increasingly difficult to add features and fix bugs. With the new architecture, it will (hopefully) be easier to manage and to add new features.
Due to the re-design of the architecture, there are a few changes to the public interface. These changes, however, are not significant, and it should be easy to update:
- All internal objects are now handled as values rather than pointers, similar to the approach taken in the underlying PugiXML library. This means that when requesting a certain worksheet from a workbook, the resulting worksheet is not returned as a pointer, but as an object that supports both copying and moving.
- The distinction between interface objects and implementation objects are now gone, as it made it very difficult to manage changes. It was an attempt to implement the pimpl idiom, but it wasn't very effective. In the future, I may try to implement pimpl again, but only if it can be done in a simpler way.
- All member functions have been renamed to begin with a small letter (camelCase), i.e. the member function WorksheetCount() is renamed to worksheetCount(). This was done mostly for cosmetic reasons.
I realise that these changes may cause problems for some users. Because of that, the previous version of OpenXLSX can be found in the "legacy" branch of this repository. However, I strongly recommend that you transition to the new version instead.
- Thanks to @zeux for providing the awesome PugiXML library.
- Thanks to @richgel999 for providing the miniz library.
- Thanks to @tectu for cleaning up my CMakeLists.txt files.