chore(architecture): Disable non 64-bit architecture

[lbartoletti]

Description

Support for 32-bit architectures is becoming increasingly obsolete. For QGIS:

• As stated on the official download page: "Since QGIS 3.20 we only ship 64-bit Windows executables."
• macOS is exclusively 64-bit
• Support for 32-bit systems is gradually deprecating on major distributions such as Debian, Fedora, and FreeBSD.
• 32-bit ARM, RISC-V, Power architectures are rarely used and tested, while newer versions of these architectures run on 64-bit (and also rarely used and tested except arm64, maybe).
• In the past, we also made patches for i386—I still remember the painful -mllvm -inline-threshold=128 ... This will be a good opportunity to start cleaning things up.

To my knowledge, there is no official statement from QGIS explicitly dropping 32-bit support, apart from the information mentioned on the download page. But if there has been an official announcement that I missed, at least this formalizes it. 😄

[github-actions]

🪟 Windows builds

Download Windows builds of this PR for testing.
Debug symbols for this build are available here.
(Built from commit e585570)

🪟 Windows Qt6 builds

Download Windows Qt6 builds of this PR for testing.
(Built from commit e585570)

[m-kuhn]

We still build qfield for 32 bit android devices.

[lbartoletti]

We still build qfield for 32 bit android devices.

@m-kuhn ah mince. Sounds like a good argument against this PR :/
I don't know anything about Android. You still have a lot of 32-bit devices in circulation? Is there an end date for support?

[m-kuhn]

Not a lot, it's marginal. If we lose support it's not the end of the world -- but why hard disable?

[rouault]

More an educational comment than anything, but there are architectures, like CHERI, with 128-bit pointers: https://lwn.net/Articles/909265/ , so if we don't want to exclude them but just 32-bit, the test could be "CMAKE_SIZEOF_VOID_P EQUAL 4"
I've experimented that a bit with GDAL (for fun) ~ 2 years ago, and that mostly worked, except in all places where one serializes a pointer to text ( printf("%p) ) and deserialize it back to pointer ( sscanf("%p)" ).

[nyalldawson]

I think we should close this. There's no strong benefit to dropping 32 bit right now, and a clear downside to doing so.