I recently had to set up Ubuntu 19.10 on a friend's laptop with a dead keyboard so had to use the on-screen keyboard. For some reason the keyboard wouldn't show up in the installer's text fields so no way to enter username, password, etc. It works everywhere else, but not in the installer, presumably because it's using the wrong toolkit out of the dozens of GUI toolkits available and the rest of the desktop doesn't recognise its text fields properly.

So unfortunately I still have to agree - how come the built-in installer program not play ball with the built-in accessibility features, and why wasn't this tested?

Logging in with the OSK also had a quirk - if you request the OSK on the login screen, common sense would suggest that you also want the OSK to persist once you're logged in... but no - once logged in you have to manually go in settings and enable it again. This is not a big deal, but just having someone think about the UX would've taken care of this.

Finally getting Netflix to work was annoying; Firefox is actually nice enough to let me know that the page requires DRM content and offered to enable it in one click without arguing about licenses or non-free software, but it still didn't work... after searching around it seems like I had to enable the non-free repo and install an extra "libavcodec" package. I figured it out, but I wouldn't blame a casual user if they gave up and said "Linux sucks".

This is correct. The Linux desktop works, but the last ~10% of polish and UX attention is what takes like 95% of the time. It's the little things, the edge cases, the things that just sort of work or break at odd times or in odd ways. The parade of edge case issues that must be fixed to make something really high quality is endless.

That sort of polish is the least fun aspect of programming. It consists of brutal iterative bug fixing, polish, design tweaking, and repeat. It's endless. Since this is the part of programming that is not fun, people rarely do it for free. Linux desktop has basically no economic model so it has no way to raise money to pay people to do the nasty grinding work required to make it truly competitive with a commercial OS.

The other huge problem is unnecessary fragmentation. There are like six popular desktop distributions that are all fairly similar under the hood. There aren't any fundamental differences between them sufficient to merit a fork. So you have to distribute your app many times to target one platform: Linux.

> The Linux desktop is still a clusterfuck today.

This is true enough. But it has nothing whatsoever to do with why it's not more widely used, given that it applies to all desktop OS's (each in their own distinct ways).

> Bill Gates doing bad stuff 20 years ago has nothing to do with it.

Well that's just not true, is it? If you successfully conspire with governments to force your OS on children, and with computer manufacturers to force it on purchasers, then this 'bad stuff' has a positive effect on your market share. It's not the only reason for Microsoft's position, but it's hardly insignificant.

If Linux had been forced on schoolkids and consumers for a couple of decades, do you not think it would be more widely used today?