It will be interesting to see if the SaaS subscription/rental model changes this. Or if they feel as beholden to adding features each release as people trying to sell software releases to get upgrade income.
I think the rental model will slow this down, but not get rid of it entirely: not as long as people still want to write blogs about the "new version of our product". I think JetBrains will make a good case study: their products are feature complete and they seemed to struggle to find new features to add lately (plus they have switched to rental model).
What we really need is something similar to API versioning: when GMail has a new version, bump up the version to 24, but keep version 23 accessible to those who prefer it.
What Jetbrains MUST do: improve the performance! Replacing my MBPs spinning rust disk with a SSD helped, but it's still eating up CPU when running on battery with indexing... it's a text editor for fucks sake.
Editor is definitely the core but all the support tools make a big chunk of features and usage. Sure, I can get quite a lot of those features in Sublime with some packages but they aren't that well integrated and easy to use!
That said, consider Microsoft Word. Since forever, it's been doing similar - if not greater - amount of work processing and reprocessing text every keystroke than IDEs contemporary to it. Yet somehow, it is not bloated, it does - and always did - work well on even low-tier machines. Also compare with OpenOffice, which does more-less the same things, only much slower and buggier. If M$ can keep Word functional and performant, I'm pretty sure Eclipse and IntelliJ could be made fast as well.
P.S. as someone who has commit access to LibreOffice (and much to everyone else's chagrin, sometimes uses it), I could be a little bit biased.
OTOH, all Java IDEs I've used so far lag by default. It's often about the little things - like those 1 second delays in menus, or 100ms delays between keystroke and letter appearing on screen, etc. Enough to make the experience frustrating. If I keep using them it's only because some advanced features don't have an equivalent (or easily installable equivalent) in Emacs. But it's a tradeoff between having a powerfool tool that I use every once in a while vs. enduring the constant experience of bloat.
Features of IDEs are cool, but they seriously need to optimize them. And to stop assuming that every developer gets to work on a brand-new machine. Maybe it's true in the US, but it's not true in other places (sometimes because managers don't see a problem with paying developers $lot but then being cheap on the equipment those same developers have to work on).
I personally use LibreOffice quite a bit, but something about the name sounds silly to me, so I avoid speaking it. I usually refer generically to "a word processor" instead.
Keep in mind that JetBrains IDEs started not as a text editor, but as a rename utility.
So really, it's closer to an AST tool with text-editing features.
For the software creator, once you've sold someone your software, the only way to make more money is to sell it to someone else, or create and sell a new version. Eventually you run out of someone-elses to sell to and can make new versions or go out of business.
... or sell a support contract or a subscription model.
I don't like it from a user's point of view, but squaring my preferences as a user with preferences of business isn't easy.
Maybe one day we'll reach the times when programming is done by "programmer-archaeologists", as described by Vernor Vinge in "A Fire Upon the Deep", whose job is to dig through centuries of legacy code covering just about any conceivable need, to find the snippets they need and glue them together. But right now, software gets obsolete just as fast as physical products.
In practice, yes, most things we buy are designed to fail. But in principle 30 years is possible, if anyone is willing to pay for it. Lots of people aren't, for many reasons including time cost of money, fashion, or just not caring.
> Most software has to be replaced after at most 10-15 years, either because of security issues or just because it's no longer supported by hardware.
Yes, so there's the answer for why the software upgrade treadmill exists: it's for software not important enough to run on dedicated separated hardware. Few people are upgrading CNC controllers or power plant controllers or aeroplane controllers because of security issues or because the hardware is no longer available.
Anyway, even on the consumer side the lifespans are rapidly lengthening. In the 1990s running current software on five-year-old computers or using a five-year-old OS would have been basically unheard of in the mainstream. Today a five-year-old computer is only a step behind current and Windows 7 has turned 6 years old and is still extremely widely used.
My mom is still using a cooking stove she bought in 1982, as a second hand purchase. Parts of it have been repaired/replaced, but I don't think they make em like they used to: I doubt a 2016 stove will make it to 2049.
It's not that creating more durable products is significantly more expensive (it could, and was done in the 80s), its that manufacturers cut costs of manufacturing and their B.o.M. in the interest of maximising profit.
The "rot" is because the environment is changing (the hardware you're running it on, other software you need to interact with) or the software is being carelessly updated.
If you need to, software rot can be eliminated. It takes effort but it can be done.
To give an example: you can run a space station with software installed in the 1980s, but you probably can't run a space station with seals or pumps installed in the 1980s.