Apple has transitioned Macs from Intel to ARM (M1/M2) chips. In the process it has provided an emulation layer (Rosetta2) to ensure that the new ARM Macs can still run applications created for Intel Macs. The emulation works very well, but is quoted to be some 20% slower than running native ARM binaries. That may not seem like a lot, but it is significant on processor intensive applications such as my own data wrangling software, which often processes datasets with millions of rows through complex sequences of merging, splitting, reformatting, filtering and reshaping. Also people who have just spent a small fortune on a shiny new ARM Mac can get grumpy about not having a native ARM binary to run on it. So I have been investigating moving Easy Data Transform from an Intel binary to a Universal (‘fat'[1]) binary containing both Intel and ARM binaries. This is a process familiar from moving my seating planner software for Mac from PowerPC to Intel chips some years ago. Hopefully I will have retired before the next chip change on the Mac.

My software is built on-top of the excellent Qt cross-platfom framework. Qt announced support for Mac Universal binaries in Qt 6.2 and Qt 5.15.9. I am sticking with Qt 5 for now, because it better supports multiple text encodings and because I don’t see any particular advantage to switching to Qt 6 yet. But, there is a wrinkle. Qt 5.15.3 and later are only available to Qt customers with commercial licenses. I want to use the QtCharts component in Easy Data Transform v2, and QtCharts requires a commercial license (or GPL, which is a no-go for me). I also want access to all the latest bug fixes for Qt 5. So I decided to switch from the free LGPL license and buy a commercial Qt license. Thankfully I was eligible for the Qt small business license which is currently $499 per year. The push towards commercial licensing is controversial with Qt developers, but I really appreciate Qt and all the work that goes into it, so I am happy to support the business (not enough to pay the eye-watering fee for a full enterprise license though!).

Moving from producing an Intel binary using LGPL Qt to producing a Universal binary using commercial Qt involved several major stumbling points that took me hours and a lot of googling to sort out. I’m going to spell them out here to save you that pain. You’re welcome.

• The latest Qt 5 LTS releases are not available via the Qt maintenance tool if you have open source Qt installed. After you buy your commercial licence you need to delete your open source installation and all the associated license files. Here is the information I got from Qt support:

I assume that you were previously using open source version, is that correct?

Qt 5.15.10 should be available through the maintenance tool but it is required to remove the old open source installation completely and also remove the open source license files from your system.

So, first step is to remove the old Qt installation completely. Then remove the old open source licenses which might exist. Instructions for removing the license files:

****************************
Unified installer/maintenancetool/qtcreator will save all licenses (downloaded from the used Qt Account) inside the new qtlicenses.ini file. You need to remove the following files to fully reset the license information.

Windows
"C:/Users/%USERNAME%/AppData/Roaming/Qt/qtlicenses.ini"
"C:/Users/%USERNAME%/AppData/Roaming/Qt/qtaccount.ini"

Linux
"/home/$USERNAME/.local/share/Qt/qtlicenses.ini"
"/home/$USERNAME/.local/share/Qt/qtaccount.ini"

OS X
"/Users/$USERNAME/Library/Application Support/Qt/qtlicenses.ini"
"/Users/$USERNAME/Library/Application Support/Qt/qtaccount.ini"

As a side note: If the files above cannot be found $HOME/.qt-license(Linux/macOS) or %USERPROFILE%\.qt-license(Windows) file is used as a fallback. .qt-license file can be downloaded from Qt Account. https://account.qt.io/licenses
Be sure to name the Qt license file as ".qt-license" and not for example ".qt-license.txt".

***********************************************************************

After removing the old installation and the license files, please download the new online installer via your commercial Qt Account.
You can login there at:
https://login.qt.io/login

After installing Qt with commercial license, it should be able to find the Qt 5.15.10 also through the maintenance tool in addition to online installer.

• Then you need to download the commercial installer from your online Qt account and reinstall all the Qt versions you need. Gigabytes of it. Time to drink some coffee. A lot of coffee.

• In your .pro file you need to add:

macx {
QMAKE_APPLE_DEVICE_ARCHS = x86_64 arm64
}

• Note that the above doubles the build time of your application, so you probably don’t want it set for day to day development.
• You can use macdeployqt to create your deployable Universal .app but, and this is the critical step that took me hours to work out, you need to use <QtDir>/macos/bin/macdeployqt not <QtDir>/clang_64/bin/macdeployqt . Doh!
• You can check the .app is Universal using the lipo command, e.g.:

lipo -detailed_info EasyDataTransform.app/Contents/MacOS/EasyDataTransform

• I was able to use my existing practise of copying extra files (third party libraries, help etc) into the .app file and then digitally signing everything using codesign –deep [2]. Thankfully the only third party library I use apart from Qt (the excellent libXL library for Excel) is available as a Universal framework.
• I notarize the application, as before.

I did all the above on an Intel iMac using the latest Qt 5 LTS release (Qt 5.15.10) and XCode 13.4 on macOS 12. I then tested it on an ARM MacBook Air. No doubt you can also build Universal binaries on an ARM Mac.

Unsurprisingly the Universal app is substantially larger than the Intel-only version. My Easy Data Transform .dmg file (which also includes a lot of help documentation) went from ~56 MB to ~69 MB. However that is still positively anorexic compared to many bloated modern apps (looking at you Electron).

A couple of tests I did on an ARM MacBook Air showed ~16% improvement in performance. For example joining two 500,000 row x 10 column tables went from 4.5 seconds to 3.8 seconds. Obviously the performance improvement depends on the task and the system. One customer reported batch processing 3,541 JSON Files and writing the results to CSV went from 12.8 to 8.1 seconds, a 37% improvement.

[1] I’m not judging.

[2] Apparently the use of –deep is frowned on by Apple. But it works (for now anyway). Bite me, Apple.