CRUD apps feel less tedious in Dark

[pbiggar]

A user says:

I really enjoyed Dark! I rarely get excited about CRUD apps but I was excited to try out Dark. There's some novelty effect but the experience was more fun and less boilerplaty than setting up a REST API. Overall, I think Dark is directionally correct and I'm excited for this future. brain dump time :)

The no-deploy was awesome. Being able to make a change and see it live. Trace-driven development isn't something I've experienced before (i wonder what happens to mutations on trace evals). Clearing DB with a REPL was interesting. I love what Dark is doing/exploring here. CRUD apps feel less tedious in Dark.

I'm curious about structural editing myself and I'm happy Dark gave me a concrete experience. It's usable! but I did find myself wanting to just edit text many times :) The error-rails was confusing to me as someone used to handling Result/Options consistently. Is there UI for it? I see something on the right but wish it was more explicit (e.g. Dict::get and Dict::getExn). I'm more convinced that when editing something that's printed as text, I want text control. At times, I wanted to switch to AST mode and drag nodes around instead of working with ctrl + . Wrapping and inserting nodes is not-straightforward. I didn't look into it but is foo.x equivalent to Dict::get foo "x"? Are values of Dict untyped? I wish I could get the static analyzed type of variables on hover. There were times where the canvas location was buggy. I create a new node and it would show up offscreen.

A number of times I had to delete code to reenter the code I wanted. If I want to type if (exprA == exprB), I learned to type == first. This makes sense! but is different than normal code. Typing exprA first could get you into a state where I couldn't get to my desired final expression.

CORS was a pain in the butt but I won't put that on Dark. I think I'd be a little faster debugging with my own backend but not sure.

Wishlist:

• grouping items on canvas. organization that's not spatial
• jump to definition and other modern editor capabilities (seeing immediate structural rename is cool!)
• disabling logs for certain endpoints (sensitive passwords)
• refactoring is a little daunting. hard to get from some A -> B via tree transforms
• errors were confusing at times. not clear where in code it tripped up. It feels good with the live eval when something shows up at the end :)

Concerns:

• flexibility. no access to existing knowledge (table joins, limited to KV, transactions, etc..)
• scale. QPS? scaling/managing table indices. Lot of this is not needed for simple apps.
• changeability. I don't have access to SQL so I'm hesitant to do migrations. Double write? seems tricky to get right. Probably something with feature flags but seems scary to work in prod :)
• portability. If I hit some issue where I need to migrate off, is it doable? How do I access/move data?

Again, love what Dark is doing! Hope this helps and happy to expand on anything.