Not really, Google applies the same disposable approach to supporting ChromeOS on bog-standard x86 hardware. I have an x86 Chromebox that Google suddenly dropped all support for - including security patches - with a generic message that "your device is out of its support window". This basically rendered it useless as a ChromeOS device, because you can't browse the public web with an unpatched web browser. Yet I can run up-to-date Linux in Crouton.
I don't see how this is technically any different than Microsoft deciding to stop all Windows 10 patches on PCs older than a certain age not because of any blocking compatibility issue, but just because they've passed a certain age threshold and they don't want to do it anymore.
I used to evangelize ChromeOS as a great, simple solution for non-technical users who just want to browse the web without the hassle of maintaining a full-fledged PC, but I won't do that if Chrome devices are part of the smartphone "time based support model" rather than the "compatibility based support" model we've had on PCs for decades. A PC for basic use can easily last 10 years, OEMs who arbitrarily shorten that are not helping their customers.
After TI left the Android SoC space, most OEMs and Carriers are beholden to Qualcomm, who write their own drivers and will never mainline them, if they can help it: long upgrade cycles will hurt Qualcomms bottomline, planned obsolescence via kernel incompatibility is how they achieve that. I don't think customers will upgrade from Snapdragon X05 to Snapdragon X20 for the minor gains in performance, if their old device could get the latest Android features.
1. https://android-developers.googleblog.com/2018/11/an-update-...
Not incidentally, "Qual" is german for "torture".
Done. You don't have to support the device for an eternity and, as a bonus, you can benefit from OS improvements to your drivers/etc. Hell, you and other manufactures using the same broadcom SOC (and hear me out... this might sound crazy) wouldn't have to spend the same engineering effort building drivers!
Of course, that would require manufactures to actually give a damn about driver quality, whereas the current situation allows them to push out the most broken patch imaginable to pass whatever conformance tests google asks for.
That would also cut into manufactures bottom line which is why I suspect they've not gone down this route. Gotta sell them phones every year... (they'd cut support to 1 month if they could)
"If only the bad people would just stop being bad!"
This mantra continually assumes goodwill from people who continue to show that they have none. At some point one needs to stop gaslighting themselves and their users and make the system more robust to bad actors. Manufacturers don't want to mainline their drivers? Fine. You can work on getting drivers mainlined while also stabilizing the interface so that it doesn't injure a billion customers when they don't.
They could force the OEM to play ball, like they require Google Play Services to be pre-installed.
I think Google really has to walk a fine line here. Android OEMs are already not thrilled with their lack of freedom when it comes to changing the experience on their phones. The more restrictions Google adds on, the closer some of them (especially Samsung) will be to cutting ties with the Android name and forking.
They are not powerless. They simply made the decision to treat carriers and device makers as the customers they cared about and not the people using the devices.
Huawei only stopped shipping the Store because they were legally obligated.
Google seems like the natural point for that. Maybe a law is needed? Too bad most law makers are completely tech illiterate.
But I agree, it won't change without some central force to kibosh it.
> Yup, and to that point, the fix is really stupid simple, push the drivers into the linux kernel mainline and give google (or whoever) a kernel build profile for your device.
Yeaaaaahhh... I don't see most companies doing that.
1. It is a lot of work to clean up your driver (that will only ever be used on one product) enough to be accepted into the mainline kernel. It takes time, effort and skill. Companies don't want to pay for that, because it doesn't contribute to bottom line. The MO is: get the product out... yesterday so we can sell it and make some money.
2. Even if it was accepted, does that mean it is going to work forever afterwards because the Linux kernel fairies / pixies / whatever will maintain it? No. If you're going push out a new release to end users, you'd better make damn sure it works. So that means each Android release has to go through a long round of QA. Who's going to pay for that?
There's only a couple ways out of the current mess we are in. Only one is likely to actually work.
Option 1: Release the IP! Have basically all phone SoCs (system on chip) use the same IP blocks for the display, I2C, SPI, memory controllers, GPIO, various sensors, power management, etc. Then you can be more like a PC where the system can basically boot up, and you can then download additional drivers as needed. This will never happen.
Option 2: Fuchsia style microkernel. Create a device driver API for the various SoC hardware blocks, and make it good and flexible enough that you never, ever have to change the API. The device driver binary is written once during initial product development, and can be carried into newer OS updates with no changes at all.
People need to realize that the SoC world is not like the PC world. Hardware interfaces change all the time, with every release of a new family member. Some stuff doesn't change much, like a USB host controller, because the SoC world has standardized that somewhat. Other stuff like power management is a total pain.
You know, your PC has BIOS/UEFI and PCIE busses and other ways for kernel to discover what it is really running on, which drivers to load and so on.
ARM devices have none of these (except for servers, where Redhat had enough clout to say, that if your ARMv8 server has no UEFI, it will have no RHEL). Not only they add up to energy consumption (it is extra hardware, after all), they add up in the BOM as well (again, it is extra hardware). So we are stuck with kernels targeting only a known hardware (which today devicetree describes to some degree) and where a poke on a wrong port could mean that the entire device hangs. And after all, no consumers base their decision on whether a third party can replace the operating system, so why make a worse (=worse battery) and more expensive (=more BOM) than the competition?
Maybe it's the profit motive.
If companies are not incentivized to change, then nothing is going to change. If companies pushing OS and hardware refreshes every year means people are going to buy more phones, I'm not sure why they would have any reason to stop.
Anecdotally, I needed a new phone because my Pixel XL was getting 4 hours of standby time and was constantly overheating. I had just gotten back from an overseas trip and was nearly stranded several times from finding myself in remote places with my backup battery dead. In the end, I settled for a OnePlus 7 (ironic, I know). I would rather have used the Pixel for many more years. hate not having a headphone jack or bezel, and I know that some product manager is going to use my purchase as evidence that those "features" were the correct move and continue to propagate the belief that nobody wants headphone jacks or bezels anymore. But I was in serious need of a smartphone that was actually reliable over one that had a headphone jack or bezel, so I got one without those features anyway.
People like me are now reliant on smartphones to do things like make calls, navigate unknown places or look up things on cellular data. In my book, if people need to have a cellphone in order to function in modern society, or are at least strongly incentivized to have one, then planned obsolescence only increases profits. Companies can market this as "upgrading" to new phones with a festureset that doesn't have to be a significant improvement over the last generation, and if people are used to this and don't mind spending the money to upgrade, then this will continue. And there are surely a lot of people who will buy the lastest model anyway even if they don't absolutely need to upgrade.
Embedded development is hard. Really hard. And much of that difficulty is intrinsic to the problem space, so it isn't going away anytime soon. But much of it is just embedded developers being masochists and embedded tool developers being sadists. For phones, the pain is probably 90% avoidable. 90 percent!
No, I don't know why we live with it either.
The reason your phone is different than your PC is that the PC's OS has all the drivers for everything built in, then loads whichever drivers your specific machine needs. Embedded Linux images are built specific to one model of hardware, with weird crappy OEM-provided drivers baked in and nowhere else available. It doesn't _have_ to be this way. But it is.
So to support your phone, the vendor basically maintains a custom-patched kernel with a bizarre build process. And that's done more or less independently for every model they're supporting. It's a huge headache!
Why deep embedded is this way is a little more understandable. (It's also where I prefer to work if I'm doing software; I don't like embedded Linux. It's just not my thing. So if I'm wrong when talking about it, that's why.) Deep embedded is often working with very limited hardware, custom circuits, and general strangeness. It also can serve the industrial, automotive, or medical markets, which naturally tend to longer product cycles. So the product is generally less standard, meaning it makes more sense (at least to me) why the support flow is so bespoke.