Bail out of Patterns with too long placeables

[stasm]

Fix #388.

[Pike]

I wonder if we should throw to bail out of the attack. But that puts us back into the square of having to try-catch every call? Well, not every, we could do that in formatPattern, and that would only do it for complex patterns.

You had concerns around try-catch and perf?

I'm having second thoughts on this, in particular around question like how this would be implemented with formatToParts where we don't know the length. Or in Rust, where we can't just throw an exception and clean things up at the top-level.

Is this specific implementation practically better than what we have? Maybe it's not worth to tweak the status quo right now, and modify the behavior once we have something spec'ed?

[stasm]

You had concerns around try-catch and perf?

Yes, but I'll need to create a benchmark to verify this. I'll report in #364 or #390, or both.

Is this specific implementation practically better than what we have? Maybe it's not worth to tweak the status quo right now, and modify the behavior once we have something spec'ed?

Yes, my rationale for opening this PR was along the same lines. This isn't perfect, but it is better that what we currently do on master.

[stasm]

I changed the approach to treating the error as a fatal one, which causes the resolver to instantly bail out of the affected pattern. This provides a much better CPU protection in deeply nested scenarios. I think we could even consider increasing MAX_PLACEABLE_LENGTH, but I left it for another time.

[stasm]

Using try...catch has no impact on our benchmarks.

[Pike]

Nice.

[stasm]

Thanks for the review and the suggestion to bail out in this particular case!