Skip to content

SQL Aide

These are the core interfaces:

  • SqlTextSupplier is a block of partial SQL, almost always a statement terminatable by ’;’
  • SqlEmitContext is the shared context which can carry state between blocks of SQL

Complex SQL generation requires many different generators working together. To help coordinate state management across SqlTextSupplier instances you use an instance of SqlEmitContext. The simplest context is an empty one which inherits the base SqlEmitContext interface.

// Type Option 1: if you do not have any custom properties
type SyntheticTmplContext = SQLa.SqlEmitContext;
// Instance Option 1: if you do not have any custom properties
const syntheticTmplContext = () =>
SQLa.typicalSqlEmitContext() as SyntheticTmplContext;
// Type Option 2: if you have any custom properties
interface SyntheticTmplContext extends SQLa.SqlEmitContext {
// your custom properties and functions go in here
myCustomStateProp1: string;
myCustomStateFunc1: (ctx: SyntheticTmplContext) => string;
// Instance Option 2: if you have any custom properties
const syntheticTmplContext = () => {
const result: SyntheticTmplContext = {
// initialize your custom properties
myCustomStateProp1: "something",
myCustomStateFunc1: (ctx) => `computed-value with ctx`,
return result;
// whichever option you use, prepare your ctx this way
const ctx = syntheticTmplContext();

Once you have your ctx prepared, you’re ready to use the Typescript string template literals as your templating engine.

The best way to start is to create each SQL statement generator as an instance of SqlTextSupplier, minimally defining an object which returns a SQL function which prepares a single SQL statement without a trailing ’;’:

const syntheticTable1Defn: SQL.SqlTextSupplier<SyntheticTmplContext> = {
SQL: (ctx) =>
CREATE TABLE "synthetic_table1" (
"column_one_text" TEXT NOT NULL,
"column_two_text_nullable" TEXT,
"column_unique" TEXT NOT NULL,
"column_linted_optional" TEXT,

If you don’t need ctx in SQL: (ctx) =>, you can leave it out — the shared state context will be provided so you can use it to store stateful information about what you’ve generated.

It’s best for templates and generators to be stateless but in complex cases you might not want to generate copies of text so you might need to track whether you’ve generated something already or not. For example, every time you put ${syntheticTable1Defn} into a template as an expression its SQL(ctx) function will be called. If you it generate each time it’s referenced, you can skip state management; however, if, for some reason, some parts or some expressions in that SQL should only be generated in specific circumstances you would utilize the typed ctx instance to do your state management appropriately.

Once you have prepared your SqlEmitContext and SqlTextSupplier instances you can use the template generator:

const stsOptions = SQLa.typicalSqlTextSupplierOptions<SyntheticTmplContext>();
const templateDefn = SQLa.SQL<SyntheticTmplContext>(stsOptions)`
-- this is a minimal SQL generator template

stsOptions creates a typical (default) set of options. templateDefn will be a new instance of SQL.SqlTextSupplier<SyntheticTmplContext>.

The template definition does not generate anything and the shared context properties will not be populated until the template is executed using templateDefn.SQL(ctx). Calling const x = templateDefn.SQL(ctx) will “run” the template and return the text while populating ctx with any shared state.

Inside the SQLa.SQL<SyntheticTmplContext>(stsOptions)\foo`string template you can have arbitrary text and a variety of${expr}instances whereexpr` can be one or more of:

  • A string
  • A number
  • A single SQL.SqlTextSupplier<SyntheticTmplContext> instance
  • An array of SQL.SqlTextSupplier<SyntheticTmplContext> instances
  • A function which accepts a SyntheticTmplContext as a parameter and returns a single or array of SQL.SqlTextSupplier<SyntheticTmplContext> instances.
  • The full list of typed ${expr}s allowed are in render/emit/sql.ts and fully described by the SqlPartialExpression algebraic type. If a type is not a component of SqlPartialExpression then it’s not a valid ${expr}.

Using the mix above, you can prepare all your SQL statements as properly typed object instances through literal SQL statements or use any SQL generator to compose your SQL in text.

There are many other features available in generated templates, including:

  • Lint state to add warnings/errors, etc. in the emitted SQL
  • Events are generated to allow cataloging of what’s being written
  • Requests to persist portions of the output in separate files
  • Behaviors can be defined so an expression at the bottom of generated text may safely influence content at the top of generated text and vice-versa
  • Properly preserves indentation and whitespace whenever possible to create reproducible SQL text (to ease tracking in Git as generated code)