fix(errors): Restore support for SES 0.18.3 and prior

[kriskowal]

Description

Agoric’s chain bootstrap vat runs with a version of SES that predates the bare function. Consequently, any library including @endo/errors running in a contract bundle in that vat must be prepared to shim any features absent on the global assert.

Security Considerations

I do not believe this comes with direct security implications, but the situation emphasizes the need for extra scrutiny for programs admitted thru governance to run in the Agoric bootstrap vat given that it runs on a fixed version of SES.

Scaling Considerations

None.

Documentation Considerations

None.

Testing Considerations

This includes a test that fails before the fix and passes after the fix. To review this PR, rebase incrementally and run tests before and after the fix.

Compatibility Considerations

The bare method falls through to the behavior of quote. This is consistent with the behavior of programs were written before the addition of bare, so I expect them to converge if that is indeed important.

Upgrade Considerations

• [ ] Includes *BREAKING*: in the commit message with migration instructions for any breaking change.
• [ ] Updates NEWS.md for user-facing changes.

[mhofman]

LGTM but I'll let Mark decide if this is the best fallback to take.

[erights]

If this were a long lived anomaly, I might desire a more accurate bare emulation. By I don't buy the "in perpetuity". We cannot ever get stuck needing to support an infinite number of prior versions, because some of those will have fatal problems.

So this PR LGTM, thanks!

[gibson042]

Is this 2024 a year reference?

[kriskowal]

Yes, to indicate when we thought this because, as @erights points out, it cannot be permanent. But, as @michaelfig indicated, it’s true for the lifetime of Agoric’s current chain. My understanding is that we can only upgrade the chain bootstrap vat at genesis.

[mhofman]

To clarify, we are planning on replacing the environment in which core evals run, and retire the bootstrap vat, at which point old agoric chains will be able to evaluate core evals without the need for this backwards compatibility. Technically replacement is not an upgrade (something something ship of Theseus)

[gibson042]

Alternative possibility:

Suggested change

	bareOrQuote as bare,
	makeError,
	note,
	quote,
	redacted,
	throwRedacted,
	makeError,
	note,
	quote,
	bare = quote,
	redacted,
	throwRedacted,

[kriskowal]

As noted with the failing test, our module transform does not support this pattern. Per the convergence of great minds, this was the shape of my first attempt!

[kriskowal]

If this were a long lived anomaly, I might desire a more accurate bare emulation. By I don't buy the "in perpetuity". We cannot ever get stuck needing to support an infinite number of prior versions, because some of those will have fatal problems.

For sure. Hopefully, the unstoppable force can move the immovable object. I suspect the only way out of this situation is for the chain bootstrap vat to eval a final contract that bequeaths its authority to a replacement chain bootstrap vat. We would presumably prevent any further core evals to execute in the original chain bootstrap vat via governance.