Don't step in to clojure.core functions

[grammati]

The new step-in command makes it easy to step in (often accidentally) to
a function in clojure.core. But instrumenting clojure core functions is
dangerous, and can result in a stack overflow that crashes the nrepl
process. In fact, just instrumenting clojure.core/symbol? (without
even debugging it) causes a stack overflow. I'm not entirely sure yet
how this happens, but it seems safest to just prevent stepping in to all
core functions.

[grammati]

Oops, I just realized I changed a debug test to step in to a clojure.core function just this morning! It's going to fail... stand by.

[bbatsov]

I'd suggest making this configurable - we can either use a function that would return true if some ns should be skipped or a list of namespaces. At any rate - a tiny bit of extra code will buy us a lot of flexibility. The advantage of using a function would be that you could check for namespaces starting with some common prefix (e.g. clojure.tools).

P.S. By this I mean this should configurable on the Emacs side and the config should be passed when the debugger is initialized.

[bbatsov]

Even if we hardcode this temporarily, it probably makes sense to check for all built-in namespaces (or at least all the commonly used ones).

[Malabarba]

P.S. Buy this I mean this should configurable on the Emacs side and the config should be passed when the debugger is initialized.

I think the only risk of overflow here comes from the fact that the debugger is instrumenting functions used by the debugger. So one calls the other in an endless recursion.

If that's the case, we don't need to make this customizable, we only need to exclude all namespaces that debug.clj depends on (be it directly or indeirectly).

[grammati]

OK, I already pushed a commit that makes it at least possible (if not easy) to configure this, before I saw your comment @Malabarba.

I think in the short term, we should just prevent people from accidentally stepping into a clojure.core function that's going to crash their process. Then we can refine the list of excluded namespaces as we find troublesome ones.

[Malabarba]

Should this be a regexp?

[grammati]

I made it so either a string (exact namespace name) or a regexp works. But in light of your earlier comment, do you think I should just switch back to a simpler implementation that hard-codes a list of namespaces to skip (initially containing only "clojure.core")?

[Malabarba]

Oh sorry, I see you have a string? clause below.

[Malabarba]

But in light of your earlier comment, do you think I should just switch back to a simpler implementation that hard-codes a list of namespaces to skip (initially containing only "clojure.core")?

I think so. And I think breakpoint-if-interesting should also avoid instrumenting if the current *ns* is in this list.

[grammati]

OK, I went back to a simpler, non-configurable implementation.

It still only prevents step-in, but I'll add code to prevent manually instrumenting functions in these namespaces if you think that's the right thing to do, @Malabarba .

[Malabarba]

Let's keep it like this for now. I might block it for manual instrumentation soon, but there's no hury.

[bbatsov]

If that's the case, we don't need to make this customizable, we only need to exclude all namespaces that debug.clj depends on (be it directly or indeirectly).

Sure. I originally thought the purpose of this PR was to just skip stepping into boring functions, which definitely seems useful to me.

Guess we should mention those peculiarities in the manual, as some people might wonder why they can't step into certain functions.