repl: don't crash if self is not in replMap

[evanlucas]

Checklist

• make -j4 test (UNIX), or vcbuild test nosign (Windows) passes
• tests and/or benchmarks are included
• commit message follows commit guidelines

Affected core subsystem(s)

repl

Description of change

Fixes: #7716

This fixes an undefined referenced. I am still trying to figure out why the undefined reference is occurring though, so marking as WIP.

[evanlucas]

/cc @Trott since he seems to be familiar with the code (b354be7)

[Trott]

Paging REPL folks: @addaleax @Fishrock123

[Fishrock123]

Sorry, no idea what's up here.

[Trott]

@evanlucas Alternatively, this will fix it too:

diff --git a/lib/repl.js b/lib/repl.js
index ed89655..ab61624 100644
--- a/lib/repl.js
+++ b/lib/repl.js
@@ -487,6 +487,7 @@ function REPLServer(prompt,
     }

     debug('eval %j', evalCmd);
+    replMap.set(self, self);
     self.eval(evalCmd, self.context, 'repl', finish);

     function finish(e, ret) {

[evanlucas]

@Trott that does work for me as well. Want to submit a PR for it? Or should I just change it here?

[addaleax]

This patch seems to be the Right Way™ to stop the undefined reference from occurring in the first place:

diff --git a/lib/repl.js b/lib/repl.js
index ed896552656c..82f532f42e5f 100644
--- a/lib/repl.js
+++ b/lib/repl.js
@@ -727,12 +727,12 @@ function complete(line, callback) {
     });
     var flat = new ArrayStream();         // make a new "input" stream
     var magic = new REPLServer('', flat); // make a nested REPL
+    replMap.set(magic, replMap.get(this));
     magic.context = magic.createContext();
     flat.run(tmp);                        // eval the flattened code
     // all this is only profitable if the nested REPL
     // does not have a bufferedCommand
     if (!magic.bufferedCommand) {
-      replMap.set(magic, replMap.get(this));
       return magic.complete(line, callback);
     }
   }

(magic.bufferedCommand is empty in the scenario from #7716 because TAB is pressed at the beginning of a line, I think.)

I’m inclined to say that some larger refactoring of the entire REPL completion implementation might be a good idea in the long run.

[Trott]

I'm not sure which is better--I can think of hypothetical pros and cons for both, although they might not hold up to closer scrutiny of the code. And it's possible that there's a third fix lurking in there. (I have an idea half-formed.)

If this isn't super-critical to fix RIGHT AWAY, I think what I'd like to do is get the code coverage up a bit by adding other tests first, just to get a few more wonky edge cases in the mix. There's about 54 statements in repl.js that aren't covered by tests right now. https://node-core-coverage.addaleax.net/coverage-4220c24b17955f7f/lib/repl.js.html

Then, once the tests are in place, try out both fixes, and maybe a third option if I can put it together, and see how they all do. Maybe try to break one or the other to see if we can.

Alternatively, if you want to get this fixed ASAP, by all means, add a test and land either fix.

[Trott]

Oh, I should add that there's an excellent chance what's going on is that I or someone else failed to account for something or other about domain behavior. It's been a while since I've had to be aware of that stuff...

[evanlucas]

What's really weird is that the test from test/parallel/test-repl-tab-complete-crash.js should not be passing. The exit event is never fired on process. The completion handler should take a callback...still digging into this though

[addaleax]

@Trott That comment was written before I posted mine, right?

The exit event is never fired on process.

Huh. I’d guess there’s a good chance I broke that test in 3ee68f7…

[Trott]

@addaleax wrote:

@Trott That comment was written before I posted mine, right?

Indeed! You have the better grasp on the totality of issues here.

[princejwesley]

@Trott REPLServer.prototype.convertToContext is unused function.

[addaleax]

@evanlucas Do you want to use my patch for this PR?

The reason test/parallel/test-repl-tab-complete-crash.js passes is that the process.on('exit', …) line is never reached, the completion fails silently without any error…

[evanlucas]

@addaleax yea, I've been working on fixing that test as well. That test should still call the event handler, IMO. I have traced the reason back to the domain exiting. So, not really sure if that is another bug in itself or not...

I'll be able to update tonight/tomorrow hopefully

[evanlucas]

Ok, updated. PTAL

[addaleax]

LGTM with CI (which I would start if loading Jenkins didn’t take, like, ages rn)

[evanlucas]

CI: https://ci.nodejs.org/job/node-test-pull-request/3394/

[evanlucas]

Landed in 392c70a. Thanks!