My DataStar Framework Assessment (19th of June, 2025)
Spent about 8 hours of looking into DataStar (DS). My current assessment is “HOLD”, here is my thought process:
2. More than one person committing code, not a scary "one man show". Less risk of abandonment. Good!
a. Documentation does not quite motivate why I would want to absorb this approach. I had to watch outside YouTube videos explaining the rationale and evolution towards signals, some JS implementaions details.
b. Significant learning curve digesting the whole data-xxx alphabet soup. Comparable to the effort I put in learning Hyperscript. (Not a compliment!)
c. It's a DSL (domain specific language) that makes "easy things even easier and hard things - impossible". Interactivity in "Islands of interactivity" using HTMX/Carson terminology can become quite complicated for signals, compared to modern and unlimited JS. Hard to know when you just start a project!
d. Probably you cannot use Chrome Dev tools/debugger. IF SO (I have not tried debugging signals-based UI yet) - BIG MINUS!
e. Probably worth the headache though, if you do not "overcomplicate". Unfortunately most projects start "simple", "complication" comes later, not by choice. See c.
2025 AI code assistants make writing and maintaining imperative, clean and modern JS massively cheaper and "turns the tables" against signals/reactive/declarative "2023 state of the art". That on top of all I mentioned in 3. But AI assistants may eventually master DS and tip the scale in the other direction? Still pondering this one, no hard opinion yet. More on AI code assitance further below ...
a. Lesser known approach, but looks powerful enough!
b. Turns DS from "library" into "framework" though - you also need server SDK (although this is simple wrapping of html/js fragments, etc). The Java SDK is incomplete, might have to contribute to it.
a. No longer a good choice as of 2025.
b. There is a reason why Carson insist on clean minimalist JS with no build steps/packages/cdns/source maps.
c. Modern JavaScript goes a long way these days and innovation is not stopping!
d. Type safety hardly worth the price of build steps and source maps hassle. (and yes, I do use type safe language for my backend!)
a. Some examples are incomplete. Browsing them, I figured out that the backend part is hidden in the GitHub repo in Go files. Then it started making more sense. Often I need to use Chrome Dev tools to get more information.
b. AI crawlers will not put such efforts, not clear how much they understand DS. Gemini Pro 2.5 (current hotness) wildly hallucinates DS code, Grok and ChatGPT seem to know a thing or two about DS (have not probed them deep enough though)
c. In 2025 make AI models your friends, or die in obscurity! This is an unfair advantage React, Vue, Svelte, Angular have against newcomers, often overlooked!
d. Writing docs for humans and AIs is tough and takes time. Consider adding cheap markdown llms.txt in the website/GitHub repo root to help AIs, while you catch up with docs for humans.
a. There seems to be Pro and Open source versions, but only mentioned in Discord channels, with some plugins available in pro only. Fair enough, but the sooner you make up your minds - the better.
b. Latest release (RC) done in Discord channels rather than GitHub. Not good!