Existing python typecheckers are bad, slow and their communities are fragmented. If the community would agree on a single good and fast type checker, everyone will benefit.
Microsoft has been dragging their feet when it comes to updating the LSP spec. Many of their Copilot features are done in VSCode, in fact using private APIs that are not accessible to other extensions.
I am all for everyone adopting LSP, but the reality is harsh.
LSP is amazing but also kinda sucks balls. It’s impossible to run VSCode without a million pops in the corner with a million extension errors. It’s so bad.
And autocomplete is the least interesting thing an LLM can do. Cursor’s UX isn’t the end game but has lots great features.
The ideal UX is still being worked out. It’s good that different people are building tools to try different ideas.
They're not opening listening ports on the local system, they're just ignoring the system's DNS and saying "Take me to this IP and this port" and then doing a DNS lookup themselves
What is an arbitrary TCP port? Ports in isolation from an IP address aren't inherently arbitrary, they're nothing, and the IP:port pair is arbitrary. Once you allow connections to any host on the internet the port doesn't really matter - you can do whatever nefarious shit over port 80. And not allowing apps to connect to external internet servers seems pretty limiting.
The slowness comes from its serialized downloads and package installations. If you install a thing that has 10 dependencies, those 10 dependencies will serially install, likely taking many seconds.
- Original HTML attribute names, `class` instead of `className`, `for` instead of `htmlFor`
- Let expressions and components return multiple elements without the need for `Fragment`.
Could make a JSX 2.0 which would be much closer to actual HTML.
reply