JSON Library

[Liberty Lover]

Please see a JSON library in Eiffel produced in 4 days flat.

ljr1981/simple_json: A simple JSON parsing library for Eiffel

[Eric Bezault]

Hi Larry,

Nice to hear from you.

> Please see a JSON library in Eiffel produced in 4 days flat.
>

> ljr1981/simple_json: A simple JSON parsing library for Eiffel <https://
> github.com/ljr1981/simple_json>

That's impressive, a lot of material to review.

I just started to read the Readme file and came across this code:

~~~~~~
schema_text: STRING_32
do
Result := "{%
%"type%": %"object%",%
%"properties%": {%
%"name%": {%"type%": %"string%"},%
%"age%": {%"type%": %"integer%", %"minimum%": 0}%
%},%
%"required%": [%"name%"]%
%}"
end
~~~~

This is not valid Eiffel. A string starting with "{ is a verbatim
string, no need to escape new-lines nor quotes. The correct
(and more readable) code should be:

~~~~
schema_text: STRING_32
do
Result := "[
{
"type": "object",
"properties": {
"name": {"type": "string"},
"age": {"type": "integer", "minimum": 0}
},
"required": ["name"]
}
]"
end
~~~~~


--
Eric Bezault
mailto:er...@gobosoft.com
http://www.gobosoft.com

[Liberty Lover]

Yeah -- AI is pretty dumb that way. It produced a ton of material (as you see). It'll take time to review it. I've noted that it tends to "forget" even though I have given it very detailed instructions and guidance (it's in the docs directory). If one doesn't give it everything it needs, the hallucinations come on strong. This might be that. Code that you're seeing in the Readme.MD is not taken directly from the code itself, where the compiler has vetted and tests have been run. This is Claude making junk up for the Readme.MD.

What I am truly interested in is getting interactive with Claude. Your comment about Zig really intrigues me. Behind the scenes, the AI is creating a small Linux environment to work in. What would be really fun is to see if it could be given an Eiffel compiler to where instead of just writing code from loose Eiffel specs it could actually attempt to compile that code, see the issues of its own mistakes and then just keep brute-force trying. That's a little scary. AI is not "AGI" ... it's still really REALLY stupid. However, it can do things at impressive speeds and with some patience and getting into a dev-cycle groove, it does perform well. That entire library was only 4 days in the making. So far, to my bleary eyes, it looks really solid. I'd love someone to pull it down, put it in an Eiffel project and give it a go as well as spot check like you've done with the Readme. 

I've also wondered if I can write some local Python code (quick-n-dirty) that allows Claude to participate with me on the EiffelStudio IDE as a pair-programmer (watching, suggesting, perhaps even writing, et al). I've got to play with that and see if it can be created. Not sure.

Anyway ... thank you ALL for taking a peek and giving feedback.