Hi @kutsurak. I actually implemented this using a hash table which should fix the performance issues. I took the liberty to get a couple of lines of your unit tests, the other line was not necessary because since we are now using a hash table we won't be able to guarantee the order. Let me know what you think. Thank you!!!! 🙏

Oooops, forgot to mention #86.

I tested this PR with a large file and got these results:

scheme@(guile-user)> ,t (->bool (call-with-input-file "large-file.json" (lambda (port) (json->scm port))))
$8 = #t
;; 1.261434s real time, 1.657068s run time.  0.465849s spent in GC.

So it doesn't have to have a big impact, which is nice. Actually, it seems to behave better than the hash table approach (see #86 (comment)). I need to test bigger files since the one I tested was an array with objects in it and the objects were not too big. So, the hash table approach would probably be better for really big objects, but I don't think that's usually the case.

Hello @aconchillo

Yes I agree that for large objects a hash map will have better performance, but for small objects the setup overhead becomes expensive. I am fine with either solution, or even with a combination of the two where there is a keyword argument to allow users to switch between the recursive and the hash table solutions depending on their data.

Let me know if you want me to do some more work on this.

Hello @aconchillo

Yes I agree that for large objects a hash map will have better performance, but for small objects the setup overhead becomes expensive. I am fine with either solution, or even with a combination of the two where there is a keyword argument to allow users to switch between the recursive and the hash table solutions depending on their data.

Let me know if you want me to do some more work on this.

Just added one tiny comment. After that I will merge it. Thank you!