Is anyone using AI for coding assistance?

Is anyone using AI assistance for coding?

I’m using ChatGPT 5. Mixed bag of results. It can do simple things pretty well and I’ve used it successfully on some minor additions. It can’t do complicated things very well, especially if the things involve multiple files, or involve interoperability between modules that I’ve developed.

I can send the prompt, and project settings if anyone is interested.

Are there any plans to create a GPT for Solar2D?

Any feedback would be helpful, and perhaps help others, thank you.

I do and it’s amazing! It can absolutely do really complex refactors against multiple classes.

It’s been really helpful suggesting coding patterns to turn O(n2) into O(n) which is a massive speed boost in real tight enterframe code on tens of thousands of objects.

I uploaded my codebase and it said it was about 4.5m tokens! so not practical yet considering the 128k token limit for sessions.

If it’s not working for you the problem isn’t the AI - more how you are using it. I am using a paid tier with full memory so YMMV.

1 Like

I use it with good result (also claude.ai). I had to figure out how to add a backend with google cloud and cloudflare while solar2d being the client. It provided coding for all ends and settings and suggestions what I should do and what I should not do and so on. It is like having a tech lead with vast knowledge in your hand.

If you know how to communicate with it, it will give outstanding results … I haven’t written a line of code since i started using it … not only in solar2d but in .NET as well … i believe it can write in all languages, perhaps much better than 80% of developers worldwide including me … and keeps getting better and better … many developers still don’t know this!! it increased my productivity by 10000%

I’m talking about business apps, dealing with backend webservices … it will code the solar2d part for front end, calling the backend, coding the backend itself, and designing your database, as well as desktop and web interfaces … it came at the right time, because I’m too old those days to write code :slight_smile:

1 Like

I’ve given it a 500mb query log and it’s optimized indexes based on real world usage - not what I assumed would be the most efficient. I also don’t have to worry about spending hours looking up “how to do things in PHP” when I can tell it what the input is, what tables it needs to hit and it knocks up bullet proof PHP and stored procedures to match!

It really is a productivity multiplier! But as always… GIGO

1 Like

Thank you for the replies!
I’m using “projects” when I work, and I give the project guidelines.
E.g.
Use Solar2D (Corona SDK).
Never break anything already working.
Structure in RTS pattern logic:

  • Separation of display, logic, input, control, and data layers.
  • Display objects must be owned by display groups, not logic modules.

Then I use as specific task requests as is possible.
Follow these instructions:

  1. line of battle, is the unit display default, rectangle is wider than tall, type indicator is drawn, and the unit is facing where it was last pointed by a face, or move to waypoint.
  2. line of march, when a unit STARTS ANY new waypoint movement, then it should morph along with its type indicator, such that the rectangle is taller than wide, and the rectangle’s display angle should point to where the unit is moving.

Is this optimal? Similar?

I can see where you are going wrong and why… you’ve basically given it a word salad of thoughts without anything meaningful to action. This is what chat replied when I gave it your prompt and asked for feedback.

Why it’s meaningless (for code)

  • No inputs/outputs. “Line of battle / line of march” aren’t variables—what function gets called with what params?
  • No events. When does it switch modes? “Starts any new waypoint” is hand-wavy (touch? pathfinder callback? velocity threshold?).
  • No data model. Where do facing, waypoint, rect, and “type indicator” live? What are their types?
  • No rendering contract. “Rectangle wider than tall” = how? Swap sprite? Scale X/Y? Tween? Rotation origin?
  • No persistence rules. If a face command happens mid-move, does it override movement heading or get deferred?
  • No edge cases. Zero-length waypoints, jitter near arrival, cancelled moves, chained waypoints, pauses.
  • No testability. No acceptance criteria → you can’t assert “works”.

How to make it implementable (minimal spec)

  • State: mode ∈ {LOB, LOM}; facing (radians); pos, vel; target (waypoint or nil).
  • Transitions:
    • onMoveStart(wp) → LOM
    • onMoveEnd() or onCancel() or noActiveWaypoint → LOB
    • onFace(a): if LOB apply immediately; if LOM store for later.
  • Rendering rules:
    • LOB: aspect = wide>tall; angle = facing.
    • LOM: aspect = tall>wide; angle = atan2(target - pos).
    • “Type indicator” is a child; follows parent scale/rotation.
  • Hysteresis (optional): enter LOM if speed > 0.1 for 200ms; exit if speed < 0.05 for 200ms.
  • Acceptance tests:
    1. Idle + face east ⇒ wide rect rotated east.
    2. Start move north ⇒ morph to tall, angle north.
    3. Mid-path change west ⇒ stays tall, rotates west.
    4. Issue face south while moving ⇒ no visual change; upon arrival ⇒ wide, facing south.

Example of a good prompt

I have a unit UI with a rectangular marker and a small type icon (child).
Data: unit.mode (“LOB”/“LOM”), unit.facing (rads), unit.pos, unit.vel, unit.target (x,y or nil).
Hooks I can call: setAspectWideTall(), setAspectTallWide(), setAngle(theta).
Implement:

  • onMoveStart(wp): set LOM, tall>wide, angle toward wp.
  • onMoveEnd/onCancel: set LOB, wide>tall, angle=facing.
  • onFace(a): store facing=a; if LOB update angle now, if LOM defer.
  • onTick(dt): if in LOM, keep angle toward current target.
    Include 200ms hysteresis on speed to avoid flapping. Provide Lua.

That’s precise, bounded, and testable. Now an AI can actually help—suggesting a tiny state machine, edge-case guards, or a clean tween for the morph.

2 Likes

Holy #$%, this is awesome!
Thank you very much for the example.
As you point out, I may not be being specific enough when asking for complex tasks.

I hope that this thread stays alive, and that it gets added to, to help others. I looked for information on Solar2D dev using AI and what you saw was the best that I came up with. BTW I asked ChatGPT to describe how to work with it in this environment, and my inadequate sample is what it came up with. Your example is much more thorough, and it’s almost like a structured english program, no? There is a lot of room for information distribution here…

Thanks again for the great example!

Taking advantage of the tips, I saw on a channel about SwiftUI a guy customizing the Cursor to create apps without needing XCode (Only when creating the project), just using extensions and configurations, does anyone here use the Cursor to create Solar2D apps?