> So, what if the Mac switched to using ARM chips instead of Intel? Well, as you can see, Apple could use Bitcode to translate every Bitcode-enabled app on the Mac App Store, without consulting developers, so it would be ready to go on day one.
> Apple doesn't enable Bitcode for submissions to the Mac App Store today
Right. The flag to enable bitcode isn't even visible in Xcode for Mac apps. There's no such thing as Mac/x86 bitcode yet, in or out of the MAS. And:
> ARC appears to use some inline assembly, which means you'll need to disable ARC for a project for arm64-to-x86 translation to succeed right now.
So this approach will only work on Objective-C programs, written in a pre-2011 style -- which I'd bet is a vanishingly small fraction of the apps on the MAS today.
It's a neat exercise in what is possible for a simplest-possible-app, but it's hard to imagine Apple would actually use this approach. It'd require them to ship a couple big new features in Xcode, and then for everyone on the MAS to recompile their apps using those features. Easier all around to just add a new architecture in Xcode the normal way, and have developers compile fat binaries.
That’s fairly easily fixed, certainly if you’re Apple. They already have the equivalent x86 code (Swift and modern Objective-C couldn't exist without it), so they could detect the specific inline ARM code fragment and replace it by the equivalent inline x86 code.
At best, Apple announces mandatory bitcode* for macOS App Store apps at WWDC in 2019; Apple announces the first ARM Macs in 2020 or 2021. So you've got a window of maybe a year or two of apps that don't have to be recompiled to support ARM, and most are likely still being updated since it hasn't been that long since they recompiled to supply Apple with a bitcode version.
This doesn't do anything to solve the big headache of an architecture transition, which are those legacy apps that will never be updated again-- you'd still need an emulation solution like Rosetta or Microsoft's x86 on ARM stuff to support those applications. At best, you're maybe working around the Adobes and Microsofts of the world that move too slowly to have a day-1 update when the new architecture comes out?
* I'll note that Bitcode wasn't mandatory on iOS when it was announced in 2015, and it's still not mandatory there today-- it seems unlikely that Mac Bitcode would be mandatory immediately when they announce it or in the foreseeable future.
If they do Intel->ARM they'll need some sort of migration strategy. Would that involve a Rosetta 2.0?
I’m not sure it’s really worth it today but this support would give Apple air cover for a future intel-arm transition.
If anyone is working on this please email me (address in profile), it's something we slated for a future research project and I'd love to talk.
It's really not too difficult to write a lifter, but it's time consuming. You have to take every instruction from the source instruction set and model its side effects in some IR. To make things simple, "whole binary" lifting tools seem to use a global CPU state (registers, flags, etc...) and so the IR you get is quite different from what you would obtain from a compiled program.
2. A subset of LLVM-IR that's architecture independent. PNacl from Google used this across chip architectures
The universe is teeming with cross architecture IRs. Eventually, LLVM will have a cross platform LLVM-IR, which will possibly be even more impactful than WebAssembly has been in it's short lifespan