SQL is a lot more intellectually coherent than its syntax belies. It's easy to be attracted to things like SPARQL, which has a well-written simple public spec, or Datalog, which doesn't really have a spec but does CTE in a way that doesn't look so painfully half-baked, or to feel really queasy looking at triggers and other features... but SQL isn't that bad. Dialects like postgres actually have pretty good support for JSON.
SPARQL is cool, but fairly limited. For example, one cannot implement common graph operations like a BFS or a DFS in SPARQL, and SQL of course can (even though I wouldn't necessarily recommend it). I was fairly puzzled when I figured this out. This means SQL is strictly more expressive than SPARQL!
The semweb badly needs a SPARQL 2.0 but we won’t get it, not least because the community was burnt out by the standards process. Myself I am keen on facilities common in document databases to represent ordered collections because the culmination of my work on the semantic web was figuring out how to use (mostly) OWL to represent complex document schemas such as the ISO 20022 financial messaging standard. And order matters for that!