Reylo 0.8.0 adds migrations, makes reylo a native command you can install without npm, returns structured Postgres and MySQL values as objects, infers through SQL function bodies and keyed writes, and fixes types that claimed less than the database could return.
Important
These changes need attention when upgrading from 0.7. Upgrading from 0.7 lists every change and what to do.
Every dialect
- One unnamed column returns its values, not rows:
select count(*) from usersis abigint(#206) reylo generateno longer fills the cache, and the project file'sclisection was removed (#174)strict_paramsis nowallow_untyped_params, with the opposite meaning (#187)- The cache moved from
.reylo/toreylo/.cache/(#179) - A hand-written client wrapper is refused, where it could return wrong answers (#242)
- The run-time schema check was removed: nothing warns or throws when a build is older than its database (#306)
- A
defineQuerystatement is sent as written, so one with:nameplaceholders needs its compiled text (#306)
Postgres
- Structured values come back as objects: row values, ranges, geometry and arrays of a subquery's rows (#193, #205)
ssl=truenow checks the certificate (#226)- A date under a non-ISO
DateStylenow throws, where it was read wrong (#242)
MySQL
- A
SETcomes back as aSet, and anENUMis typed as its labels (#193) - The type check needs MySQL 8.0.38 or later (#306)
DuckDB
- DuckDB support was removed, query typing,
.run()and migrations alike (#258)
A project that ran reylo migrate on 0.8.0-beta.0 renames its lock to reylo/reylo.lock (Upgrading from the beta).
Migrations
Reylo Migrate is part of the reylo command, on Postgres, MySQL and SQLite. Declare the schema you want in reylo/schema.sql, and reylo migrate new writes each change as a plain .sql migration, tried on a shadow database before it is written. (#179)
npx reylo migrate init # start from an existing database
npx reylo migrate new add_bio # write the next migration, from schema.sql
npx reylo migrate dev # run it on the development database, and write the lock
npx reylo migrate deploy # run what the lock lists, when you deploy| Area | What it does | Link |
|---|---|---|
| The lock | reylo/reylo.lock orders the migrations with chained checksums, so two branches that each add one conflict in git before anything is deployed
| #179, #282 |
| Data loss | A migration, or a reylo db push, that drops something or loses values needs --drop; --rename old=new names a rename
| #179, #256 |
| Deploys | reylo migrate deploy builds no shadow database, so it needs no create database rights, and refuses a database changed by hand
| #179 |
| Schemaless mode | Without schema.sql, the migrations are the schema, and the development database stands in for it
| #179 |
| Development | reylo db push --watch takes the database to schema.sql on every save; db reset, db wipe, db introspect and reylo run cover the rest
| #179 |
| Writes on Postgres | reylo migrate new changes an existing table without blocking its writes: a check or foreign key is added not valid and validated in a later file, and an index is built concurrently. A statement that still blocks writes needs --blocking
| #291 |
| Turso and libSQL | reylo migrate and reylo db run on Turso Cloud and any libSQL server, through a libsql:// or https:// URL with ?authToken=, and a migration may use libSQL's own SQL: vector columns and indexes, and alter table … alter column
| #301, #322 |
| Hosted databases | Every command was run end to end on Neon, Supabase, Tiger Cloud, Google Cloud SQL, Amazon RDS and Aurora, Azure Database and Turso Cloud, and fixed where one refused | #293, #307 |
| Postgres versions | Every command runs on Postgres 14 to 19 | #266, #311 |
| Scripts and CI | Every command takes --json, with stable refusal codes, and the development commands refuse to run where CI is set
| #179 |
| From code | migrate(), exported from reylo, runs reylo migrate deploy on a server
| #253 |
| Durable Objects | migrate(ctx.storage) runs the migrations the Vite or esbuild plugin bundled, on the object's own SQLite
| #253 |
import { migrate } from "reylo";
// on a server: reylo migrate deploy, as the command line runs it
const ran = await migrate();
// in a Durable Object's constructor: the bundled migrations, on the object's own SQLite
migrate(ctx.storage);Reylo Migrate's docs cover each command.
The reylo command
The reylo command on npm now runs the native binary (#174). The same binary installs without Node, checked against the release's SHA256SUMS (#275):
brew install reylosql/tap/reylo # Homebrew
curl -fsSL https://reylo.dev/install.sh | sh # macOS and Linux
irm https://reylo.dev/install.ps1 | iex # Windows PowerShell
scoop bucket add reylo https://github.com/reylosql/scoop-bucket && scoop install reylo
docker run --rm -v "$PWD:/work" ghcr.io/reylosql/reylo migrate statusIts reylo analyze command, new in 0.8, prints what Reylo infers for each statement, as YAML, or as JSON with --format json (#172, #239):
file: queries/get_user.sql
$schema: https://reylo.dev/typedefs/preview.json
dialect:
name: postgres
version: "18.1"
sql: |-
-- One user, by id.
select id, email, bio from users where id = $1
kind: select
comment: One user, by id.
params:
- name: id
type: { kind: integer, bits: 64, label: int8 }
result:
kind: rows
cardinality: at_most_one
columns:
- name: id
type: { kind: integer, bits: 64, label: int8 }
- name: email
type: { kind: text, size: { max: 255 }, label: varchar }
- name: bio
nullable: true
type: { kind: text, label: text }
comment: Shown on the profile page
diagnostics: []The same descriptor comes from code with analyze(), which reads the project's database as the type check does (#191, #195):
import { analyze } from "reylo";
const descriptor = await analyze({ sql: "select id, name from users where id = :id" });Besides each type, it reports what the database declares and what Reylo infers about it:
| Fact | Example | Link |
|---|---|---|
| Declared size and precision | varchar(20) is size: { max: 20 }, numeric(10,2) has precision and scale
| #180, #192, #210 |
Bounds and patterns from a CHECK
| check (age between 0 and 150) gives min and max; check (slug like 'a%') a pattern
| #200, #251 |
| A UUID's version | gen_random_uuid() is version 4, uuidv7() version 7
| #232 |
| Unique columns and sort order | A result's unique column sets, and the order its rows are sorted on (Postgres)
| #184, #213 |
| Distinct arrays | array_agg(distinct …) and json_agg(distinct …) are marked distinct
| #251 |
The CLI page covers installing the command, reylo analyze and analyze().
What .run() returns
| Value | 0.7.0 | 0.8.0 | Link |
|---|---|---|---|
| A Postgres row value of known fields | "(1,Ada)"
| { id: 1, name: "Ada" }
| #193, #197 |
| A Postgres geometric type | "(1.5,-2)"
| { x: 1.5, y: -2 }
| #193 |
| A Postgres range | "[1,5)"
| { lower: 1, upper: 5, bounds: "[)" }, or { empty: true }
| #193 |
| A Postgres multirange | "{[1,3)}"
| An array of ranges | #193 |
array(select c from (…) c) on Postgres
| The rows' text | An array of objects, each field decoded | #205 |
A MySQL SET
| "a,c"
| new Set(["a", "c"])
| #193 |
A MySQL ENUM column's type
| string
| The union of its labels | #193 |
select count(*) from users
| A row keyed by the name each database makes up, such as count or count(*)
| bigint
| #206 |
NaN or Infinity inside JSON a Postgres statement builds
| The string "NaN", typed number | "NaN" | "Infinity" | "-Infinity"
| The number, typed number
| #171 |
A Postgres timestamptz or timestamp
| "…10:00:00.5Z"
| "…10:00:00.500000Z", six fraction digits always
| #190 |
A Postgres money
| "$1,234.50", the session's locale text
| "1234.50", its exact decimal
| #254 |
A statement whose one column is an expression with no as returns that column's values the same way through .runSync(), .try(), .with() and a Standard Query: select lower(name) from users is a string[]. An untyped JSON parameter takes anything JSON.stringify writes, typed as the new exported JsonInput. (#206, #193)
Inference
| Area | What changed | Link |
|---|---|---|
| SQL function bodies | A Postgres language sql function's body is read as the query it is, so select id, body from one_comment(:x) is one row of non-null columns, where the declaration alone made it many and nullable
| #211 |
| Keyed writes | An update … from or delete … using that pins a key returns a row or null, and an insert of one selected row returns that row
| #190 |
| Constants | A column a statement fills with constants is typed as those literals: select 'user' as kind … union all select 'post' … is "user" | "post"
| #243 |
| Whole rows | A whole row of a derived table, subquery, with query or aliased join is typed field by field, inside JSON and alone
| #167 |
| Arrays | Built arrays of arrays are typed with each level; array(select …) takes its subquery's type; array_remove(tags, null) and a CHECK that keeps NULL out make elements non-null, through coalesce(tags, '{}') too
| #181, #190, #166, #287 |
CHECK forms
| tags <@ array['a', 'b'] types ("a" | "b")[]; jsonb_typeof(doc) = 'object' types an object; a bound such as price < 1e9 keeps out NaN and Infinity
| #190, #171 |
RETURNING
| A returned column the statement just set to a non-null value is non-null: update t set note = :note::text where id = :id returning note returns a string note, where it was nullable
| #267 |
| Triggers | A BEFORE trigger whose every return is new, such as an updated_at trigger, no longer makes a write nullable
| #190 |
| Grouped JSON | An object that puts an aggregate beside a grouped column is typed inside a scalar subquery, and on MySQL at the top level too | #230 |
| SQLite views | A write to a view through its INSTEAD OF trigger type-checks
| #238 |
| libSQL vectors | On a Turso project's local file, the type check reads libSQL's vector columns and indexes, its vector functions and vector_top_k. Each failed with "no such function" or "no such table"
| #316 |
MySQL &&
| where id = :id && name = :name returns at most one row, as and does. && was read as binding tighter than =, so it was many
| #290 |
| Hovers | A query's hover and .try()'s result name the error type, QueryError, instead of listing every code
| #221 |
tsc
| An error in a .sql file no longer hides the TypeScript errors beside it, and tsc --incremental checks again after a schema change
| #249 |
| Postgres 19 | Postgres 19 is supported, and CI tests on it | #311 |
tsconfig.json
| "contentMappers": [{ "package": "reylo" }] is enough on TypeScript's 7.1 nightlies from 2026-11-12 on, since the package names its extensions. An entry that lists them keeps working
| #315 |
Drivers
| Driver | What changed | Link |
|---|---|---|
| Knex | .run(), .try() and .with() take a Knex instance, transaction or savepoint, over pg, mysql2, mariadb, better-sqlite3 and sqlite3
| #233 |
| SQLite Wasm | .run() and .runSync() take an oo1 database of @sqlite.org/sqlite-wasm, OPFS ones included
| #235 |
| sql.js | .run() and .runSync() take a sql.js Database, from sql.js 1.17.0, which binds a bigint as an integer and gives each error its code. An older sql.js is refused, naming 1.17.0
| #331 |
| Durable Objects | A Workers project with Durable Objects and no D1 database is checked on Cloudflare's SQLite, so a function they refuse, such as json_pretty, fails the type check
| #241 |
| SQLite, mariadb, PlanetScale, Turso, libSQL | A row's keys always match the declared column names. They were the driver's own names, such as 1 + 1 /*file='q.sql'*/
| #207 |
| MySQL | A value error raised outside any column has no column. It was the string "(null)"
| #216 |
| postgres.js 4 | JSON, array and date parameters are sent as text. postgres.js 4 stored a JSON value as a JSON string from a statement's second run, and refused a jsonb[]; .try() reads its renamed error type field
| #270, #268 |
Bun.sql
| On Bun 1.4.4 and later, Postgres values are decoded from Bun's raw bytes, so a timestamp keeps its microseconds, and MySQL's column types tell a BIGINT from a DECIMAL
| #269 |
| Postgres TLS | A URL with sslnegotiation=direct connects. It never did on Postgres 17 and later
| #268 |
| SQLite | A bigint parameter outside SQLite's 64-bit range is refused before the statement runs, on every client. bun:sqlite and Bun.sql kept only its low 64 bits, and sqlite3 bound it as NULL
| #268, #274 |
mysql2 through Cloudflare Hyperdrive
| .run() on mysql2, or Drizzle or Knex over it, works through Hyperdrive, which refuses prepared statements: a connection that refuses one runs its statements through query(), with the values written in by Reylo. A FLOAT read that way has the 6 digits MySQL sends as text (guide)
| #261, #284 |
| Postgres logins | A URL with channel_binding=require, as Neon writes them, connects. It failed with "server did not use channel binding"
| #287 |
| Postgres addresses | When every address of a host fails, the error names each one's failure, a server's answer first. It named only the last, such as "Network is unreachable" on a machine with no IPv6 route | #287 |
import knex from "knex";
import getUser from "./get_user.sql";
const db = knex({ client: "pg", connection: process.env.DATABASE_URL });
const user = await getUser.run(db, { id: 1 });Standard Query
A built module's Standard Query carries more of what Reylo infers, for drivers that run it themselves.
| Field | What it holds | Link |
|---|---|---|
columns, output()
| A decoder per column, and the cardinality applied to decoded rows, so a driver decodes value by value as it reads | #165 |
key
| The result columns that identify each row (Postgres) | #160 |
ordered
| true where the order by makes the order total (Postgres)
| #163 |
cardinality
| How many rows the statement returns | #206 |
readOnly
| true where nothing in the statement can write, for a wrapper that retries reads
| #190 |
inputSchema, outputSchema
| maxLength, pattern, bounds, multipleOf, enum and description from the schema and its comments
| #199, #192 |
Faster and smaller
| Change | Before | After | Link |
|---|---|---|---|
| A bundle carries only its query's dialect (one query, esbuild 0.28.2, minified) | 162.8 KB for every dialect | Postgres 76.7 KB, MySQL 117.7 KB, SQLite 63.2 KB | #228 |
JIT is off on the engine's Postgres connections (566 statements, pgvector/pgvector:pg17)
| 264 s | 20 s | #158 |
| The catalog is read once per run (517 statements, local Postgres 18, median of five) | 1,878 queries, 10.29 s | 849 queries, 6.01 s | #190 |
| Postgres rows are decoded faster (25,000 rows × 12 columns through postgres.js, reads per second) | 13.5 | 16.8 | #164 |
A Postgres insert of ::rows (10,000 rows into a 20-column table, Postgres 18, postgres.js, median of 15)
| 91 ms | 63 ms | #277 |
A SQLite CASE of 10,000 arms is analyzed faster
| 7.2 s | 0.45 s | #238 |
The Vite, Rollup and Rolldown plugin declares hook filters, so those hosts skip modules that aren't .sql
| Every module | .sql modules only
| #137 |
The JIT change matters for Postgres builds that have JIT on, such as the official Docker images most CI services run.
Fixed types and values
Each of these types was wrong for what the database returns, and most of the fixes widen a type.
| Database | Statement | Was | Now | Link |
|---|---|---|---|---|
| Postgres | jsonb_build_object('tags', '[]'::jsonb)
| { tags: string }
| Any JSON | #225 |
| Postgres | A column of a left-joined lateral subquery that reads the table before it
| Non-null | Nullable | #225 |
| Postgres | insert into p values (unnest(:ids::int[]), 'x') returning id
| One row | Any number of rows | #225 |
| Postgres | A materialized view | Its defining query, run now | A stored relation, with its own unique indexes as keys | #226 |
| Postgres | A key compared under a case-insensitive collation | At most one row | Many | #209 |
| Postgres | jsonb_build_object('a', 1, 'a', null::int)
| { a: number }
| { a: number | null }, the last value
| #209 |
| Postgres | distinct on (2), and distinct on an output column's name
| At most one row | Many | #155 |
| Postgres | replace(b, 'a', 'b') where a type of the database's own has an implicit cast to text, as PostGIS's geometry has
| string
| string | null, since the cast may return NULL
| #273 |
| Postgres | A statement on a quoted table with a dot in its name, "a.b"
| Read as table b of schema a, its triggers and keys missed
| Read as the one table | #264 |
| MySQL | where id = :id and x = 1 xor y = 2
| At most one row | Many, since XOR ranks between AND and OR
| #225 |
| MySQL | An IF or CASE with a BIT result
| DECIMAL
| unknown
| #226 |
| SQLite | select count(*), name from t
| name non-null
| Nullable, NULL over no rows
| #225 |
| SQLite | ifnull(dt, 0) over a DATE column
| number
| number | string
| #238 |
| SQLite | qty % 0.5
| Non-null | Nullable | #124, fixes #122 |
| SQLite | select 9223372036854775808
| An integer | A real, as SQLite reads it | #248 |
| Postgres | random(1, hi) over a nullable hi
| Non-null | Nullable | #306 |
| Postgres | An array built by ||, array_append or array_prepend from a nullable value
| Non-null elements | Nullable elements | #306 |
| Postgres | A varchar key compared with a value of no known type, as coalesce(:e::bpchar, 'x')
| At most one row | Many | #306 |
| Postgres | A partial unique index whose literal carries a cast that changes it | At most one row | Many | #306 |
| Postgres | A key compared with a SQL function labelled stable whose body is volatile | At most one row | Many | #306 |
| Postgres | A statement that writes x = null
| The key pinned, x non-null
| No pin, x nullable
| #306 |
| Postgres | A CHECK that compares the column under a varchar(n) cast
| The column narrowed to the list | Not narrowed | #306 |
| Postgres | A bare name that is a column of a composite type named as the table | The table's row, non-null | The column | #306 |
| Postgres | array[x] of a domain over an array
| One more dimension, non-null | One level, nullable | #306 |
| Postgres | An array built of array casts, slices or || of arrays
| One dimension | Its own dimensions | #306 |
| Postgres | A derived table that reads a parameter, inside built JSON | The other branch's shape | Its own | #306 |
| Postgres, MySQL, SQLite | A NOT NULL JSON value inside built JSON, at any depth | A non-null leaf | A nullable leaf | #306 |
| MySQL | A query the engine's parser can't read | The server's NOT NULL flags | Every column nullable | #306 |
| MySQL | a || b under the default sql_mode
| Not read | Read as OR, so the query keeps its types
| #306 |
| MySQL | An aggregate in a subquery whose argument reads only outer columns | The subquery's | The outer query's, whose bare columns are nullable | #306 |
| MySQL | year(d)
| An integer up to 2155 | An integer; a YEAR column keeps its bound
| #306 |
| MySQL | A string key compared with cast(:p as char)
| At most one row | Many | #306 |
| MySQL | A BIGINT or DECIMAL key compared with arithmetic on a parameter
| At most one row | Many | #306 |
| MySQL | A parameter compared in HAVING with a select-list alias
| The FROM column's type
| The alias's | #306 |
| MySQL | A BIT through coalesce, ifnull, greatest, least or any_value
| A BIT
| unknown
| #306 |
| MySQL | A JSON Schema CHECK that names a key twice
| Narrowed by the last | Not narrowed | #306 |
| MySQL | DATE and DATETIME
| From year 0001 | From year 0000 | #306 |
| MySQL | An ENUM or SET label with an escaped line break
| Read without it | Read as stored | #306 |
#222 fixes wrong values at run time and build time:
- two columns of one name on MySQL and SQLite took the first column's kind;
- a
select *built before a migration put values under the wrong names on mysql2, mariadb, PlanetScale, libSQL and Turso; - a parameter named
constructoror__proto__and left out bound JavaScript's own member; - D1 refused the
bigint.run()returns; - webpack's disk cache served a module built for the old schema.
#306 also fixes:
- the type check's replay of a SQLite database's stored schema could attach a database or write a file, so a crafted database file could create files on disk during a type check;
- a MySQL URL that names a socket and asks for TLS failed to connect, and now connects without TLS;
- a SQLite key's
ON CONFLICT REPLACEwritten across a line break was missed.
Upgrading from 0.7
Every dialect
| Change | What to do | Link |
|---|---|---|
| One unnamed column returns its values, not rows | Write const count = await q.run(db), or add as count to keep a row
| #206 |
| Columns filled with constants are typed as their literals | Widen the variable's type where code assigns another value to it | #243 |
A JSON column whose schema leaves additionalProperties out is typed by the keys it names, with no index signature
| Name the key in the schema, or give additionalProperties a schema
| #169 |
An untyped JSON parameter is JsonInput, not unknown, so passing a value typed unknown fails the type check
| Type the value passed to it as JsonInput, exported from reylo
| #193 |
reylo generate no longer fills the cache, and the project file's cli section was removed
| Drop it from scripts and CI: the type check and the build steps fill the cache. Delete cli:
| #174 |
strict_params was replaced by allow_untyped_params, which reads the other way round
| Write allow_untyped_params: true for strict_params: false, and delete strict_params: true
| #187 |
The cache moved from .reylo/ to reylo/.cache/, which keeps itself out of git
| Delete .reylo/ and its .gitignore line
| #179 |
An unquoted .env value is cut at its first #, as dotenv reads it
| Quote the value, or write # as %23
| #227 |
A hand-written { query } client wrapper is refused
| Pass the driver's own pool, client or transaction, or subclass or proxy it | #242 |
| A build that cannot reach its database says so, and reads only that server's analysis | Unset the URL for a build that relied on another database's analysis | #242 |
| The run-time schema check was removed: a query sends no catalog query before its first run, and nothing warns or throws when a build is older than its database. A structured insert writes the defaults its build saw | Build again after each migration | #208, #306 |
A query made with defineQuery sends its sql as written
| Give a statement with :name placeholders its compiled text and order, and its marks on Slonik, or on mysql2 where the server refuses prepared statements
| #306 |
A value with no time zone no longer carries format: date-time or format: time in a query's JSON Schema
| Validate such a value without the format | #306 |
Postgres
| Change | What to do | Link |
|---|---|---|
| Row values, geometric types, ranges and multiranges come back as objects | Remove the code that parsed their text; parameters take the same form | #193 |
| An array of a subquery's rows comes back as objects | Remove the code that parsed the text | #205 |
NaN and Infinity inside built JSON are numbers
| Test with Number.isNaN and Number.isFinite
| #171 |
| A timestamp always has six fraction digits | Update code or tests that compare its text | #190 |
A money comes back as its decimal text, "1234.50", not the session's locale text
| Format it for display in the app | #254 |
A keyed update … from or delete … using, an insert of one selected row, a call to a function whose body has a key filter, and a CALL with OUT parameters return a row, not an array
| Check the row for null instead of reading .length
| #190, #211, #186 |
| An array of more than one dimension in a column, or in built JSON, is refused | Select to_jsonb(column), or set allow_nested_arrays: true
| #181, #243 |
ssl=true in a URL checks the certificate and its name
| Use ssl=no-verify or sslmode=require for an unchecked connection
| #226 |
A date under a non-ISO DateStyle makes .run() throw
| Set datestyle = 'ISO, DMY' on the database or the connection
| #242 |
| Some columns widen to nullable or to any JSON (Fixed types and values) | Handle null where the type check now asks
| #225, #226, #273, #306 |
MySQL
| Change | What to do | Link |
|---|---|---|
A SET comes back as a Set
| Remove the code that split its text; parameters take a Set too
| #193 |
An ENUM is typed as its labels, and an empty '' value is refused
| Read such a column with nullif(column, '')
| #193 |
| A made-up column name that isn't an identifier, beside other columns, is an error | Name the column with as
| #206 |
A JSON parameter where MySQL takes a value, as in json_set(doc, '$.name', :name), is an error
| Write :name::json where the error points
| #242 |
Some columns widen to unknown or to int | bigint (Fixed types and values)
| Handle the wider type where the type check now asks | #226, #234, #306 |
| The type check needs MySQL 8.0.38 or later. On an older server it crashes the server or never returns | Run the type check against MySQL 8.0.38 or later | #306 |
Through query(), as on Hyperdrive, a statement whose reading turns on sql_mode is refused on a session whose mode differs from the build's
| Run it under the sql_mode the build used
| #306 |
SQLite
| Change | What to do | Link |
|---|---|---|
| A made-up column name that isn't an identifier, beside other columns, is an error | Name the column with as
| #206 |
A double-quoted string, select "hello", fails the type check, as most SQLite drivers refuse it
| Write string literals in single quotes | #215 |
A parameter the lexer did not read, such as :été, is an error
| Rename the parameter | #226 |
A bigint parameter outside the 64-bit range throws before the statement runs
| Pass a value in range | #268, #274 |
| Some columns widen to nullable (Fixed types and values) | Handle null where the type check now asks
| #225, #238, #306 |
DuckDB
| Change | What to do | Link |
|---|---|---|
DuckDB support was removed: a duckdb: URL is refused, and defineDuckdbQuery, DuckdbClient and DuckdbQueryError are gone
| Stay on 0.7 for a DuckDB project | #258 |
Upgrading from the beta
The migrations lock is reylo/reylo.lock, no longer reylo/reylo-lock.yaml (#282). Every command that reads or writes the lock refuses to run until the old file is renamed, rather than read the project as having no lock:
git mv reylo/reylo-lock.yaml reylo/reylo.lockThe beta's known issues are fixed: migrations run on Postgres 14 and 15 (#266), and as a role that isn't a superuser, as on Neon and Supabase, where migrate new, migrate deploy --dry-run and db push failed with "permission denied for table pg_user_mapping" (#293).
Four changes to migrations since the beta can surface in a project that ran them:
- MySQL executable comments run. A statement written as
/*! … */was skipped, so a database the beta migrated can lack what such a statement creates (#306). - An unreadable lock is an error. A
reylo/reylo.lockthat exists and can't be read counted as no lock (#306). - The Durable Object bundle refuses what a deploy refuses: a draft, a file that is no migration, and a
schema.sqlthat no longer matches the lock (#306). - The schema fingerprint the beta stored isn't compared. The next deploy stores a new one and says so (#306).
Docs
reylo.dev now describes 0.8:
- reylo.dev has a homepage, and the docs open at reylo.dev/docs (
ce24aa1). - The docs are split into pages with a sidebar (
c451fed), with guides for deploying on Cloudflare D1 and Durable Objects, Neon, PlanetScale, Supabase, Turso and Next.js, and in CI (381daa1,8de0c6e). - Reylo Migrate has a page of its own, with its docs under it (
bc22e61,ba3e060). - A guide to Postgres and MySQL through Cloudflare Hyperdrive (
e7288e7). - A comparison with other SQL tools, each mark from a real run (#183).
- A guide to moving from Prisma (
d3940d3).
Testing and internals
- The oracle also runs every statement over empty tables and with different values for parameters of one type, checks itself by corrupting analyses it must catch, and runs on Postgres 14 and MySQL 8.0.38 beside the newest versions (#306).
- The oracle analyzes each Postgres and MySQL statement again under other session settings, such as
DateStyleandTimeZone, and fails if the analysis changes (#198). - Every inferred type is checked against the JSON Schema that
reylo analyze's output follows (#188). - The release's npm stage job restores no cache and pins npm, and the build checks the SQLite amalgamation's checksum (#222).
- A branch's checks, and one platform's binary, can be run by hand (#212).
- Each release installs and runs the binary through every channel on each platform before anything is staged (#275).
- The repository moved from
colinhacks/reylotoreylosql/reylo, and the old address redirects (#279).
What's Changed
- fix(postgres): read DISTINCT ON names and positions as postgres does by @colinhacks in #155
- docs(spec): state the JSON form of every type each database writes by @colinhacks in #159
- feat(postgres): infer the result key of a statement by @colinhacks in #160
- fix(site): format tuples, number literals and wide unions in the cards by @colinhacks in #161
- perf: decode timestamptz by position and build rows with one object literal by @colinhacks in #164
- fix(postgres): turn jit off on the engine's connections by @colinhacks in #158
- feat(engine)!: answer every analysis in the type descriptor format by @colinhacks in #162
- feat(mapper): give a Standard Query per-column decoders and an output by @colinhacks in #165
- fix(mapper)!: type an open JSON Schema object by the keys it names by @colinhacks in #169
- fix(engine): list a json union's literals first and null last by @colinhacks in #170
- feat(engine)!: add a tagged_union kind and make a union's per optional by @colinhacks in #168
- feat(postgres): read the fields of a whole row of no catalog type by @colinhacks in #167
- feat(mapper)!: return NaN and infinities inside postgres json as numbers by @colinhacks in #171
- fix(engine): widen a built object's nullable field as nullable() does by @colinhacks in #173
- feat(cli): add reylo analyze by @colinhacks in #172
- feat(postgres): infer whether a statement's order is total by @colinhacks in #163
- feat(cli)!: run the reylo binary as the reylo command by @colinhacks in #174
- docs(spec): move the examples out of the type descriptor spec by @colinhacks in #176
- feat(postgres): read each column's type modifier by @colinhacks in #180
- fix(postgres)!: type arrays by their dimensions by @colinhacks in #181
- chore(release): 0.8.0 by @colinhacks in #175
- docs(site): compare reylo with other sql tools from real runs by @colinhacks in #183
- feat(postgres): read a check that keeps null out of an array by @colinhacks in #166
- feat(spec): drop the union's per field by @colinhacks in #189
- feat: export compile from reylo's root, reading the project's database by @colinhacks in #191
- build: drop the .npmrc release-age opt-outs by @colinhacks in #194
- feat!: infer keyed writes, read the catalog once, write six-digit times by @colinhacks in #190
- feat!: add js.codecs and return structured values as objects by @colinhacks in #193
- feat(spec)!: write durations as one maximum, let ranges span any type by @colinhacks in #177
- feat(cli): rename compile back to analyze, as a function and a command by @colinhacks in #195
- fix(spec): read a generated file's line endings as nothing it says by @colinhacks in #185
- feat!: name the project file's loosening flags allow_, off by default by @colinhacks in #187
- fix(engine)!: narrow sqlite strict, postgres written params and call by @colinhacks in #186
- test(oracle): analyze each statement again under other session settings by @colinhacks in #198
- feat(mapper): write what a type knows into a query's json schema by @colinhacks in #199
- feat(mysql): read each column's type modifier by @colinhacks in #192
- chore(examples): record the type check on tsc 7.1.0-dev.20261104.1 by @colinhacks in #196
- fix(postgres): decode a row with no catalog type as its fields by @colinhacks in #197
- test(spec): hold the examples and every inferred type to the format by @colinhacks in #188
- feat(postgres): read bounds, a divisor and a pattern from a check by @colinhacks in #200
- feat(spec)!: drop the any kind, which opaque already says by @colinhacks in #204
- fix(postgres): correct key collations, repeated JSON keys, function rows by @colinhacks in #209
- feat(spec)!: return a statement's result as a rows type by @colinhacks in #184
- feat(mysql): type a parameter written to a column by its definition by @colinhacks in #210
- feat!: warn when a query's columns change at run time by @colinhacks in #208
- feat(spec): say which result columns the rows are sorted on by @colinhacks in #213
- feat(spec)!: call the database's name for a type its label by @colinhacks in #214
- fix(mysql): read a value error's (null) column as no column by @colinhacks in #216
- feat(engine)!: infer through SQL function bodies and table macros by @colinhacks in #211
- fix(postgres): fix what Postgres 19 RC1 breaks, and two row counts by @colinhacks in #203
- fix(engine): spell the js.codecs set option Array, as its type by @colinhacks in #202
- fix: align standard query, empty-array schemas and sqlite strings by @colinhacks in #215
- ci: run a branch's checks and one platform's binary by hand by @colinhacks in #212
- fix: keep a row's keys to the columns its type declares by @colinhacks in #207
- fix: name renamed whole rows by type, read duckdb temp tables first by @colinhacks in #217
- fix(engine): keep a refused describe's position for the host by @colinhacks in #219
- test: keep the column-name tests passing on windows by @colinhacks in #220
- feat!: return one unnamed column's values, not rows by @colinhacks in #206
- fix: name error types in hovers rather than listing every code by @colinhacks in #221
- fix(duckdb): decode node-api's getRows() values in a standard query by @colinhacks in #218
- feat(postgres)!: return an array of a subquery's rows as objects by @colinhacks in #205
- docs(site): show the query a tool got wrong in each comparison popover by @colinhacks in #223
- fix: correct run-time values, stale caches and the release job by @colinhacks in #222
- feat(postgres): expose the types a statement builds to a linking host by @colinhacks in #224
- fix(engine): close seven soundness holes from the audit by @colinhacks in #225
- fix(engine): close ten more holes from the audit by @colinhacks in #226
- perf: add sql hook filters by @TheAlexLichter in #137
- fix: bundle only the run time of the query's dialect by @colinhacks in #228
- fix(sqlite): treat % by a fractional literal under 1 as nullable by @fengqiaoqu in #124
- fix!: read .env files as dotenv does by @colinhacks in #227
- fix(cli): resolve the project in reylo analyze as the type check does by @colinhacks in #229
- fix(duckdb): return a GEOMETRY as its WKB bytes by @colinhacks in #231
- feat: run queries on a knex instance or transaction by @colinhacks in #233
- fix(mysql): widen table integers beside a nested rollup by @colinhacks in #234
- fix(engine): type grouped json in scalar subqueries and on mysql by @colinhacks in #230
- feat(sqlite): run queries on SQLite Wasm's oo1 databases by @colinhacks in #235
- feat(spec): type a uuid's version, keep integer bounds inclusive by @colinhacks in #232
- fix(spec): build the uuid example from functions postgres 17 has by @colinhacks in #237
- feat(cli): print reylo analyze's descriptors as yaml by default by @colinhacks in #239
- fix(sqlite): close eight soundness holes in the engine by @colinhacks in #238
- fix(postgres): list the error code postgres 19 adds by @colinhacks in #240
- fix(sqlite): check durable object projects on cloudflare's sqlite by @colinhacks in #241
- fix!: stop four silent wrong answers at run time and build time by @colinhacks in #242
- fix(duckdb): read keys from a table's text where constraints fail by @colinhacks in #245
- fix: load the build-time half where turbopack does not trace it by @colinhacks in #244
- refactor(postgres): bound statement depth with the engine's own limit by @colinhacks in #246
- fix(sqlite): type an integer literal past 64 bits as a real by @colinhacks in #248
- fix(mapper): keep tsc's type errors beside an error in a .sql file by @colinhacks in #249
- feat(spec)!: add tables, enums, intersections and character sets by @colinhacks in #243
- feat(cli)!: add migrations by @colinhacks in #179
- fix(cli): show project paths with forward slashes on windows by @colinhacks in #252
- feat: export migrate from reylo's root, for servers and durable objects by @colinhacks in #253
- feat(engine): infer distinct arrays, regex patterns and field comments by @colinhacks in #251
- feat(duckdb): load extensions once and open files read-only to read by @colinhacks in #250
- feat: read decimals as decimal.js, bignumber.js or big.js values by @colinhacks in #247
- feat(cli)!: refuse what db push drops until --drop or --force accepts it by @colinhacks in #256
- feat!: add vector, geometry and inet codecs and read money as a decimal by @colinhacks in #254
- refactor!: remove duckdb support by @colinhacks in #258
- test: skip the vector codec test on a server without pgvector by @colinhacks in #260
- fix(mysql): name mariadb when a server refuses mysql2's prepares by @colinhacks in #261
- chore(release): 0.8.0-beta, staged under npm's beta tag by @colinhacks in #262
- feat(spec): say what each statement may do as its capabilities by @colinhacks in #259
- chore(release): 0.8.0-beta.0 by @colinhacks in #263
- fix(cli): run migrations on postgres 14 and 15 by @colinhacks in #266
- fix(postgres): keep a dot inside a quoted table name by @colinhacks in #264
- fix: postgres.js 4 error type, direct TLS, and SQLite 64-bit range by @colinhacks in #268
- fix(engine): type a written returning column by the value written by @colinhacks in #267
- fix(postgres): decode bun.sql's raw bytes and read its result columns by @colinhacks in #269
- fix(postgres): send every postgres.js parameter as text declared unknown by @colinhacks in #270
- fix(postgres): trust no call a user's implicit cast can feed by @colinhacks in #273
- fix(sqlite): check the 64-bit range on every binding path by @colinhacks in #274
- feat(release): install the reylo command without npm by @colinhacks in #275
- chore: move the repository to reylosql/reylo by @colinhacks in #279
- perf(postgres): read ::rows as json with one reader, in order by @colinhacks in #277
- refactor(cli)!: rename the migrations lock to reylo.lock by @colinhacks in #282
- fix(release): link an organization's package settings by @colinhacks in #286
- chore(release): 0.8.0 by @colinhacks in #285
- feat(mysql): run mysql2 through query() where prepares are refused by @colinhacks in #284
- fix: bind logins to tls, name each address's error, keep checked arrays by @colinhacks in #287
- fix(mysql): bind && as AND binds by @colinhacks in #290
- fix(engine): check the files before db reset drops anything by @colinhacks in #289
- fix(engine): run reylo migrate as a role that is no superuser by @colinhacks in #293
- fix(engine): leave another role's leftover shadows alone by @colinhacks in #294
- fix(engine): plan and compare what other diff engines' tests caught by @colinhacks in #295
- feat(cli)!: keep postgres migrations from blocking writes by @colinhacks in #291
- fix(cli): keep a not null an identity needs, and reset settings per file by @colinhacks in #304
- fix(cli): refuse migrate new where a migration ran as it no longer reads by @colinhacks in #302
- feat(engine): run reylo migrate on Turso Cloud and libSQL servers by @colinhacks in #301
- fix(cli): word two migrate readouts as hints, not warnings or claims by @colinhacks in #308
- ci: keep the test servers' ports out of the ephemeral range by @colinhacks in #310
- feat(cli): show a spinner and colour the sql in db introspect by @colinhacks in #309
- ci(postgres): test on postgres 19 and list it as supported by @colinhacks in #311
- feat(mapper): name the content mapper's extensions in its manifest by @colinhacks in #315
- fix(engine): plan what other schema tools and fuzzing catch by @colinhacks in #300
- fix!: close soundness holes and harden migrations, oracle and release by @colinhacks in #306
- fix(cli): raise the catalog reader's version for what it now reads by @colinhacks in #318
- ci: build pgvector for any cpu on postgres 19 by @colinhacks in #319
- fix(cli): print notes and next steps in one style, commands on own lines by @colinhacks in #320
- feat(sqlite): type libsql's vector functions and vector_top_k by @colinhacks in #316
- fix(engine): run reylo migrate on hosted databases by @colinhacks in #307
- fix(cli): word migrate()'s refusals as the command line's titles read by @colinhacks in #321
- feat(engine): run libsql's own sql through reylo migrate by @colinhacks in #322
- fix(engine): plan typed tables, inheritance and index options by @colinhacks in #317
- fix(cli): dim notes and their leads, and offer next steps as options by @colinhacks in #326
- ci(release): stage on ubuntu-24.04, where setup-nub fronts its node by @colinhacks in #330
- feat(sqlite): run statements on sql.js by @colinhacks in #331
New Contributors
- @TheAlexLichter made their first contribution in #137
- @fengqiaoqu made their first contribution in #124
Full Changelog: v0.7.0...v0.8.0