fix(xsnap): supply missing file, line numbers based on sourceURL

[dckc]

fixes #2578

depends on agoric-labs/xsnap-pub#1

@michaelfig @warner I was surprised to find some metering diffs when I grabbed the sourceURL support from the moddable public branch. They're in TypedArray (xsDataView.c), which our "has metering changed?" test is insensitive to. Should I update from xs-meter-10 to 11?

As of #3607 we were at 91b0b8bfdc4c, which is part of Moddable's public branch.

But somehow in chore: publich 7.0.0 we regressed to some of our tweaks (to 6024736).

relevant diffs from 6024736-6d7f33f: https://gist.github.com/dckc/9513e1acc1c263ee962b0e282cf48153

[dckc]

@warner note: this probably enables the debugger network connection in release / production mode.

[dckc]

Also: I thought about testing the interaction with importBundle explicitly but I had enough trouble getting this much working.

Do you want a unit test involving importBundle too?

[erights]

LGTM

[dckc]

Bummer... a test fails in ci but passes locally.

~/projects/agoric/agoric-sdk/packages/zoe
16:53 connolly@jambox$ yarn test test/swingsetTests/zoe-metering/test-zoe-metering.js

[dckc]

@warner you asked how different... see https://gist.github.com/dckc/9513e1acc1c263ee962b0e282cf48153

[dckc]

test-refillMeter was nice enough to fail locally when I showed it to @warner . We have a diagnosis and a crude fix but I'd like to coordinate with @katelynsills on that.

More on that soon...

[dckc]

@katelynsills was there anything more than trial-and-error to the 4.2M computron figure in zoe/test/swingsetTests/refillMeter/test-refillMeter.js?

I experimentally discovered that this change makes the test pass:

-        initial: 4_200_000n, // startInstance seems to take around this much
+        initial: 4_300_000n, // startInstance seems to take around this much

@warner thought about having the test compute this number by measurement in a pre-test phase, since this test isn't so much about the absolute numbers. But we agreed that the absolute numbers are somewhat relevant and they should be tested somewhere, so that's not urgent.

[katelynsills]

@katelynsills was there anything more than trial-and-error to the 4.2M computron figure in zoe/test/swingsetTests/refillMeter/test-refillMeter.js?

It was unfortunately trial and error.

@warner thought about having the test compute this number by measurement in a pre-test phase, since this test isn't so much about the absolute numbers. But we agreed that the absolute numbers are somewhat relevant and they should be tested somewhere, so that's not urgent.

Ah, yeah, that seems like a good idea. This test is intended to confirm that the meter refills when expected and then the vat shuts down when the meter runs out when expected. So, the pre-test data would help me know "when expected" actually is. But you're also right that we need the absolute numbers somewhere too - in other words, I'd like to know when changes to agoric-sdk will result in broken assumptions about computrons for actions. I can make an issue for this, and feel free to tag me and any related issues with a TODO comment if that's helpful.

[katelynsills]

This a review of only the Zoe changes, but those look good to me. I'll work on making the metering test more robust.

[katelynsills]

Added an issue for making the "refillMeter" test more robust: #3804