rewrite-vaadin-in-react

untitled

tags: learning

content

prompt 1: prompt 2: question:

• “APl layer strategy: Currently the Java service layer (UserService, SequencingRunService, etc.) acts as a BFF that wraps the raw EdgeOS APIs (GSS, IMS, GDS, etc.) with business logic, caching, and data transformation. How do you want the React app to talk to the backend?”

  • 
  • 
  • 
  • 
  • 
  • 
  • 

• what im thinking:

  • “像这种时候, 我就很纠结, 我感觉这context里都是好context, 但是可能又太bloated了, 他最开始explore的时候, 吃了那么多legacy codebase的💩, 该不该留”
  • i think i have an answer now:
  • look at the cache, but what does it mean?
  • 

• the inconvenience? cursor i have to keep myself behind the computer. i have to approve its every single TODO. at least i can just queue up a bunch of “go”s in the chat

challenge

• the biggest challenge right now is the CSS. Routes are fine, all most mapped 100%. Auth is okay, unsure what the original logic is, but for demo, we could have a very simplified version. also, we might not necessarily wanna copy every single logic in the legacy codebase. risking losing some tribal information (which is bound to happen with this kind of rewrite), i am more in favor of cutting down complexity and fats from the legacy codebase

  • I thought with the competency of coding agents I would not have to deal with CSS again, I would not have to center another div.
  • i was wrong.
  • because the whole plan is to rewrite everything. For now the methodology is to have one platform support with a minimum viable product so I can demo it to colleagues. When it comes to this kind of demonstration I think visual representation is very important. Especially when people on the other end might not be 100% technical
  • i just feel that having a bad visual representation (if the css don’t match at least 90%), it automatically deducts some credibility

• I thought it would be a good idea to have agents screenshot every single page of the original codebase and map it to the new codebase, with sixty pages in place. but the agent is not, to my surprise, quite as good at this as I thought

  • because from the original draft generated by coding agents is just way too much different from the original css.
  • I sent a screenshot and asked the agent to recreate, and it only shows fix it pieces by pieces, i have to keep prompting it with directions and very detailed instructions for certain cases for it to work. a screenshot with a “recreate the css” is not enough
  • at this point, it genuinely feels like i just ran a few-hour long sessions with a few follow-up sessions to generate some codes that seem to work but actual don’t, aka, slop

up

down

reference