We have the following use case:

• document contains a "version" field that, for legacy reasons, is a string containing an int
• we want to ensure that we are not overriding a newer version of the document
• if no document is present, create it

if "version" was an int, it would be simple: updateOne({ id: XXXX, version: { $lt: newVersion }}, {$set: {...}}, { upsert: true })

as "version" is a string, we have to use: updateOne({ id: XXXX, $where: "parseInt(this.version) < newVersion", {$set: {...}}, { upsert: true })

This has pretty dramatic performance implications, though.

In this use case it would be useful to have support for $expr in the query predicate of an upsert as we could use aggregation expressions such as $toInt instead of $where. The semantics would ideally be the same as with $where - ie the $expr would be discarded when upserting rather than evaluated to figure out if there are fields to be set.