I am curious how it round-trips. For the output languages that are also supported input languages, can you transpile back and get something close to or identical to your input?
must match the original identically
(formatting aside -- and/or sent through a deterministic code formatter on both sides, pre and post...).
Otherwise there is something wrong (or potentially wrong!) -- in the transpiler and/or target language and/or the data model of the target language... and the tiniest error could potentially turn into something way bigger later if not addressed at that stage...
I am curious how it round-trips, too!
Anyway, excellent question!
advThanksance
* Galois connexions subsume, among other things, Dijkstra's weakest preconditions/strongest postconditions.
Also, I think that some of the next breakthroughs for AI might be as follows:
1) Given any Turing-complete computer programming language A, construct a translator to a higher-order intermediate representation, then from that higher-order intermediate representation, construct a translator to programming language B...
2) Do this for all known Turing-complete computer programming languages;
3) Do this for all known Mathematical categories, groups, lemmas, theories, etc., and other Mathematical "objects" (for lack of a better term!) -- in other words, create an intermediate data model "map" (analagous to an intermediate higher-order language used by a transpiler) -- between all of these things in all of Mathematics...
Category Theory -- seems like it would be "The next level up"...
(Note that I am speculating about most of this(!) -- I claim to be an idiot in [some/many/all] of these areas! <g>)