Incomplete-model IntelliSense

functional

Get help from rough DDL before it is valid enough to deploy; SQLly keeps that uncertainty visible.

A DDL sketch does not need to be deployment-ready before it can be useful. SQLly reads a clearly unfinished definition, recovers what it can, and overlays its best interpretation on the real schema, without presenting that interpretation as fact.

Recovered, and labelled as recovered

Every object in the model carries a fidelity: either exact, it parsed cleanly, or recovered, in which case it also carries the list of assumptions the parser had to make and where in the text it made them. Nothing is silently guessed. A half-written CREATE TABLE still contributes its columns to completion, and the model remembers that it did so on a recovery rather than a clean read.

Incomplete regions

When a whole area cannot be collected, a file that will not parse at all, a database or schema that failed to load, the model marks that region as incomplete rather than pretending it is empty. This is the distinction that matters in practice:

“I do not know about this” is not the same as “this does not exist.” Inside an incomplete region, SQLly does not tell you a table is unknown, it cannot know that. Validation holds its tongue where its information is missing, which is what keeps the squiggles trustworthy everywhere else.

Edit mode and exemptions

While you are actively shaping a schema, the engine can run in edit mode with a set of exemptions: name prefixes or file paths excused from validation, each with a written reason. It is the difference between an editor that helps you build something new and one that shouts at you for every reference that does not exist yet, and because each exemption carries its reason, the list does not quietly become permanent.

Validation also runs at two scopes: against the statement tree you are typing, and against the whole model. So a query can be syntactically fine and still flag a reference that no source, live or local - knows anything about.