That's not a network effect.

They're clearly not going to be able to afford $100m/year in free CI.

• You have to tediously git submodule update --init --recursive every time you checkout a commit. There's an option to do it automatically but it's super buggy and will break your .git directory.
• Switching between branches that have different sets of submodules doesn't really work. Git won't remove/recreate the submodules like it will for normal directories. Worst case is changing a directory to a submodule or vice versa.
• If you're working on a feature that spans several submodules you have to switch branches in all of them instead of once.
• Making co-dependant changes across submodules is a nightmare.
• If you're using submodules for first party code (not uncommon), it basically creates a new public interface where you didn't have one before. Now you have to worry about backwards compatibility and testing becomes much harder. Monorepos don't have that problem.

The list goes on... Some of them depend on exactly what you're using them for.

The slightly frustrating thing is that there isn't a great answer for what to use instead. Git subtree has its own problems. Language-specific package managers do too. There aren't any good languages agnostic package managers I know of.

I'm really hoping Jujutsu will come up with a better solution because it is possible. But it's hard, and they are constrained by Git compatibility so I won't hold my breath.

Worktrees are great, and a good reason (if you needed any more) to avoid submodules like the plague. Worktrees don't work with submodules.

No Pyright is just a type checker. The IDE features are part of Pylance which is closed source.

You could use uv workspaces. It means you only have one venv though.

A VSCode multi-root workspace might also work.

Yeah I was really surprised by that. Surely "sequelite", given SQL is commonly pronounced "sequel" (c.f. PRQL).

RPI 256GB -> 512GB upgrade: $15

Mac Mini 256GB -> 512GB upgrade: £200!??

Eesh I mean I am at peace with price differentiation but that is wild.

There's also CPC/Farnell but none of those are in the same league as McMaster Carr. Much smaller ranges, worse prices, worse websites, missing CAD models, etc.

Another option is Misumi but they have even worse prices and don't even sell to individuals.

I'd recommend going to McMaster Carr just to see what we are missing out on.

Yeah fair enough. That was a bit mean, sorry.

I think the main issue is that ML isn't useless just because we don't understand how it works. It clearly does work and we can use that. It's hardly unique in that way either. There are a gazillion medicines that work but we don't really know how. We're not going to abandon them just because we don't understand them.

And it's not like people aren't trying to understand how they work; it's just really difficult.

The calculator analogy also makes no sense. You can't build a working speech recognition engine by manually entering equations for phonemes or whatever. That's actually not a million miles away from how speech recognition worked in the 90s and 2000s... or I should say "didn't work".

Nothing else has code completion that even comes close to being that good.

Well, except Visual Studio (for C++), Qt Creator, and every Java IDE in existence.

There aren't many good cross platform GUI toolkits. I mean realistically is there anything other than Qt?