liveSlots consensus mode and better verbosity switches

[michaelfig]

refs: #2519

Description

This properly routes liveSlots output to a disabled console if in consensus mode. Also, updates the use of $DEBUG to be more coherent and not as bizarre (where an empty $DEBUG would log less than a missing $DEBUG).

Also improve a longstanding wart where XS console logging displayed quote's hodgepodge of JSON and deeper inspection, just by replacing the whole thing with an NPM package based on Node.js's util.inspect rendering.

At last, all of agoric start, agoric start -v, and agoric start -vv look right, in order of increasing verbosity. Notably, when running an actual chain, no vat-level errors are displayed.

Security Considerations

Documentation Considerations

Testing Considerations

[dckc]

I looked at all the code and I don't see any problems, but I don't clearly understand the motivation and I'm not confident about the non-determinism guarantees, so I defer to @warner .

[warner]

Ok, that looks good. I would have loved to see a unit test that shows the console being disabled when consensusMode is true, but I certainly couldn't whip one up without a couple of hours, so I can't expect anyone else to.

r+

[warner]

ok, when printAll() causes some random object's toString to be invoked and that does more console.log-ing, this structure should unwind the reentrancy properly, nice

[turadg]

@michaelfig why is this copied from upstream instead of sourced from the object-inspect dependency in package.json?

[michaelfig]

I could have commented that better. It's not a verbatim copy, it's the object-inspect.js with an unnecessary CommonJS require('./util.inspect') removed since it doesn't work under XS.

[turadg]

Cool. What's the reason to prevent prettying it? If it's to make it easier to update from upstream, WDYT of using patch and to make it work under XS?

[michaelfig]

Cool. What's the reason to prevent prettying it? If it's to make it easier to update from upstream, WDYT of using patch and to make it work under XS?

That's fine with me, if you're bored and want something to do. 😄 Tests cover it, so it's a safe thing to try.

[michaelfig]

patch-package can't be used, though, because people might get this code directly via NPM's @agoric/xsnap.

[dckc]

I notice now, after the fact, that object-inspect.js is a significant chunk of code that doesn't follow our style. And it goes right at the bottom of our TCB. I'm surprised it didn't get a mention in Security Considerations like other PRs where I've seen us take on a new dependency. @warner were you conscious of this in your review?

And I see logging tests in test-boot-lockdown.js changed. While the overall issue of console API compatibility for XS (#2146) remains open, we (@kriskowal @erights and I) did decide to use assert.quote from SES in #3856 and I am just now realizing this PR means reconsidering this decision. I better double-check that they're OK with this.