jank repl clojure.core loads clojure.core twice

[frenchy64]

$ jank repl clojure.core
Bottom of clojure.core
Bottom of clojure.core
clojure.core=>

[elken]

We can actually go up to 3 it seems

After this, any subsequent calls don't show the log message.

This implies that clojure.core is not being cached properly, I assume what's happening here is:

• jank start up and loads clojure.core, doesn't add to the cache
• It passes over to the repl, which gets told to load clojure.core, doesn't add to the cache
• When we get into the REPL, we finally require it and add it to the module cache

Is this right @jeaye ? If so it seems we should always cache namespaces when we require, or is there some AOT/JIT shenanigans going on?

[jeaye]

Nope, it's an issue with the module loader. Should be marking this as loaded and then not loading it again.

I said the same thing over on the PR: #175

[Samy-33]

the native loader::load() doesn't yet mark the ns loaded. It's the job of the clojure.core/load to include it in the clojure.core/*loaded-libs*.

Two ways we can go:

1. Update *loaded-libs* from loader::load. And, skip modules already loaded. ; I don't like this because we are essentially doing the same thing at two places.
2. Natively load clojure.core in the beginning and initialize the *loaded-libs* with clojure.core ; I dislike this idea more, because of the spatial distance between initialization and loading clojure.core. It's not at all obvious.

Perhaps, we can make do with the first one. But I am gonna check how clojure does it first.

[jeaye]

Ok, that's a good breakdown. Choosing one of those is necessary, over the change we have in this PR, though. We'll need to ensure that modules don't get loaded again.

To me, I think that the first option is the only viable one, since this will apply to any module loaded natively. Given that jank is a C++ library as well as a language, people will be using jank to load modules natively. Those need to be properly handled.