Parse fast

[fr-an-k]

Greatly accellerates parsing by memoizing toksuper per depth level.
Some tests were also fixed due to conflicts.
I can't spend anymore time on this.

[rurban]

This is missing the return

[pt300]

So instead of parent_links in the tokens themselves, we have a separate side-array used only during parsing to store the "parents stack". Cool idea. There's probably a cache argument to be made, but I don't think it's a concern for JSMN.

[fr-an-k]

If you profile the previous code (with a larger JSON) almost 100% of the time is spent on looking for the parent, but the parent is already known so that's unnecessary work.
A linked list could be used instead of an array on the stack or heap, I didn't think of that in the moment.
If you utilize the token pool for that, then there is no need for a separate API function, so the the final PR would be very minimal.
I guess you just need to link to the last encountered opening { or [ token and they should link to their parent.

[pt300]

Did you try #define JSMN_PARENT_LINKS? How does your solution compare to that? That is the same thing, but done in the token pool itself. Its existence is not really well documented if at all.