Cosmos: Fixup to Regex.IsMatch translation

[AndriySvyryd]

Fixup to #28121

• Throw a detailed exception for unsupported options when translating Regex.IsMatch (see discussion below)
• "Constantize" regex options via EvaluatableFilter (Cosmos: Add translator for Regex.IsMatch method #28121 (comment))
• Avoid generating empty options parameter if RegexOptions.None is specified (Cosmos: Add translator for Regex.IsMatch method #28121 (comment))

[roji]

Depend on #26410, right?

[smitpatel]

Not really blocked, the method translator should throw exception from itself.

[roji]

Don't we avoid doing that since it would block client eval?

[smitpatel]

Depends if we want to do client eval of certain things.

[roji]

OK, I'll bring this up in design, possibly with a proposal for #26410.

[smitpatel]

If you think we would need #26410 for this, we can wait till we put that one in plan.

[roji]

Just to figure out how we see things... If we're OK with throwing in method translators, #26410 becomes much less important, maybe even closeable (note it's already in the milestone).

I still think that ideally we just do #26410, e.g. by having translator return some NotTranslatableExpression (which wraps details); and then we don't have to discuss throwing because that always allows client eval and provides nicer exceptions. But we can discuss all that.

[Marusyk]

I'll prepare a PR for this

[roji]

@Marusyk sure, but we'll need to decide what we want to do with unsupported options (design discussion pending).

[roji]

Design decision: we'll do #26410 at some point, but for now it should be fine to just return null for unsupported options, so the user will just get the generic "cannot translate" exception. @Marusyk this should unblock you for working on this issue.

[Marusyk]

So, should I return the previous version with HashSet?

"Constantize" regex options via EvaluatableFilter

I'm not sure that understand this task. Do I need to implement CosmosEvaluatableExpressionFilter?

[roji]

So, should I return the previous version with HashSet?

Nope, I don't see any reason for doing so. The current code does the right thing (for now) - if there's any unsupported option, it just returns null.

Do I need to implement CosmosEvaluatableExpressionFilter?

Yeah, @smitpatel can probably offer some more assistance if needed.

[roji]

@Marusyk do you intend to work on this?

[Marusyk]

Unfortunately, I don't have time to investigate EvaluatableExpressionFilter.

[roji]

Discussed with @smitpatel and the constantization of the RegexOptions parameters would require some infra work, so we'll leave that for now.