Remove sorting of statements from update queries (#2169)

[kurrestahlberg]

Consistent updates where a new field value depends on an another field that gets updated in the same query are impossible across different databases unless the original update order is maintained.

The problem is with MySQL not following the SQL spec as described here: https://dev.mysql.com/doc/refman/5.7/en/ansi-diff-update.html. This causes the value used for calculating another value sometimes already be updated and sometimes not, depending on which database is used and what the names of the fields are if the statements are sorted alphabetically.

This PR removes sorting of statements when building an UPDATE query. A possible future issue is that the key order of an object is not defined in the spec while all known JS engines keep non-integer-like keys in insertion order so this should work correctly across all engines. The order might thus be different with an engine in the future. More info on ordering here: https://bugs.chromium.org/p/v8/issues/detail?id=164

[elhigu]

Should insert be implemented the same way? Even though I don't see the same kind of problems with it so maybe not...

[kurrestahlberg]

I was thinking the same but as I couldn't think of any reason why there would be an issue when inserting I figured a smaller change is better.

[elhigu]

Yeah, this looks good to me though. Thanks! @wubzz @tgriesser @rhys-vdw anyone, if there are no objections I'll merge this in tomorrow. I marked this as a breaking change even that this was more like undefined behaviour earlier. Maybe documentation could mention that updates are generated in the Object.key() iteration order of used js-engine. Just to make this functionality documented and official.

[wubzz]

No opposition from me, this makes sense. My guess is the .sort() originates from a copy-paste of _prepInsert. Don't see a real use case for sorting columns for insert either, tho..

[elhigu]

I added documentation issue about this. Thanks!