Upgrading to 9.25

9.25 is the identifier release. Nothing was rewritten, but the DDL Weasel emits changes for any schema whose names are not all plain lowercase identifiers, and three behaviours change in ways you can observe from the outside.

Go straight to the latest 9.25.x

9.25.0 refuses to create a table whose primary key constraint name is longer than the engine's identifier limit, and 9.24 accepted it. The names it rejects are entirely conventional — just long, which a wide composite key on a long table name produces easily. Fixed in 9.25.1; see the 9.25.1 fix below.

If your schema uses conventional names throughout — lowercase, no spaces, no reserved words, no brackets — the emitted DDL is byte-identical to 9.24 with the one exception above, and there is nothing else here to do except read the one that looks like a regression.

What 9.25.2 changes

One behaviour change and one refactor.

⚠️ A check constraint is now refused where it would have been ignored

#488. TableBase.CheckConstraints is on the shared base, so every provider's Table accepted a check constraint. Only PostgreSQL and SQL Server ever wrote one into the DDL. On Oracle, MySQL and SQLite the constraint sat in the model, never reached the database, and was never compared during delta detection either — so a caller got a table without the constraint they asked for and nothing said so.

Both entry points now throw on those three:

table.AddCheckConstraint("ck_orders_qty", "quantity > 0");                  // NotSupportedException
table.CheckConstraints.Add(new TableCheckConstraint("ck", "quantity > 0")); // NotSupportedException

The collection refuses as well as the method, because adding to it directly is the more common spelling and throwing from only one of them would have left the silent path open.

If you call either on Oracle, MySQL or SQLite you will now get an exception where you previously got silence. Nothing of value is lost — the constraint was never created — but the failure moves from invisible to loud, which is the point. The engines all support check constraints; Weasel does not emit them there yet, and the message says so and points at the issue.

This is the rule #449 settled, applied to the one place it had been missed: a caller who sets a property gets it, or gets an exception, never a quietly narrower object. It was missed because the property lives on TableBase rather than on each provider's own type, so there was no per-provider surface to audit.

Oracle deltas share one implementation

#492. Six Oracle object types — views, triggers, stored procedures, packages, synonyms and functions — had each written out the same delta class after hitting the same failure separately. They now share OracleReplaceDelta.

No behaviour change and no API you were likely using: the six replaced classes were internal. Weasel.Oracle.Functions.FunctionDelta is public and stays, now delegating to the shared rule.

What 9.25.1 fixes

#485 and #486 — one root cause, two symptoms.

9.25.0 started validating the names a table writes that are not database objects of their own — its columns, its primary key constraint, its check constraints (#468). It ran them through the provider's AssertValidIdentifier, which also enforces the length limit. Schemas that applied cleanly on 9.24 began throwing.

The length rule does not belong there, and Weasel's own code already said so. An object name the database truncates becomes unaddressable and drifts on every check afterwards — worth refusing. A local identifier is only ever emitted inside its own table's DDL and never addressed by name again, and TableDelta already compares both PrimaryKeyName and the primary key column list through TruncatedNameIdentifier precisely so a truncated one still matches. Weasel was handling truncated local identifiers downstream while refusing to create them upstream.

Migrator.AssertValidLocalIdentifier now applies the same safety rules with no length limit. A quote, a semicolon, a line break, leading or trailing whitespace are still rejected wherever they appear. The base implementation defers to AssertValidIdentifier, so a provider outside this repository keeps the stricter behaviour until it opts in.

The quiet half

#486 presented as a 23505 on pk_mt_event_progression during a Marten per-tenant daemon catch-up — an error naming a table with nothing to do with identifiers. The rejection had aborted a projection's schema application, and the daemon then ran against storage that was never created. The loud failure and the quiet one were the same rejection. If you saw anything strange on 9.25.0, this is worth ruling out before looking further.

Also in 9.25.1: functions on MySQL and Oracle (#482), which completes the object type matrix apart from check constraints on Oracle, MySQL and SQLite (#488 — a pre-existing gap, not a 9.25 change).

At a glance

Change	Who is affected	Action
A check constraint is refused on Oracle, MySQL and SQLite (9.25.2)	Anyone calling AddCheckConstraint on those three	You now get an exception where you got silence — the constraint was never created either way
SQLite no longer drops your table to change a column	Anyone on SQLite applying a type, foreign key or primary key change with AutoCreate.All	None — but read below, because you may have lost data on an earlier version
Oracle now sees index, foreign key and primary key drift	Every Oracle user	Expect a first run that finally applies indexes it had silently been ignoring
Column names are no longer rewritten	Anyone who declared a column with a space in the name	Write the underscore yourself, or accept the new name
SQL Server quotes identifiers it previously left bare	Names that are not regular identifiers	None — the DDL was invalid before
A name you bracketed yourself is passed through and unbracketed in the model	SQL Server callers who bracketed to work around the above	Stop bracketing; compare against the bare name
MySQL Table.Identifier renders and compares differently	Anyone comparing DbObjectName instances directly	Compare through MySqlObjectName
Weasel.MySql.Sequence is obsolete	Anyone using it — though nothing could consume it	Use AUTO_INCREMENT
Schema fingerprints re-evaluate once	Anyone using Migrator.UseSchemaFingerprinting	None — expect one "everything changed" pass

⚠️ Column names are no longer rewritten

#458. PostgreSQL, Oracle, SQLite and SQL Server used to turn a space in a column name into an underscore. They no longer do.

table.AddColumn("Order Date", "datetime2");
// 9.24: creates Order_Date
// 9.25: creates "Order Date"

On an existing database this reads as a new column, so a migration will try to add "Order Date" alongside the Order_Date that is already there. If you were relying on the rewrite, write the underscore yourself:

table.AddColumn("Order_Date", "datetime2");   // unchanged on both versions

Why it changed: the rewrite only ever happened in TableColumn. Index, foreign key and primary key column lists were left alone, so declaring Order Date produced an index over a column that did not exist — no error at model time, and either a failed statement at the server or an index that drifted forever. The rewrite was also never a decision anyone made; it predates the identifier work, and #448 settled that a supplied name is honoured or rejected, never silently changed.

PreserveIdentifierCase now controls case folding and nothing else — it used to switch off the space rewrite as a side effect. SQLite gains the flag, which it was the only provider to lack.

SQL Server: identifiers are quoted where they were not

#443, #446.SchemaUtils.QuoteName bracketed only a hardcoded reserved-word list, and several emission sites did not call it at all. Ordinary schemas are byte-identical; any name that is not a regular identifier now brackets.

A name you bracketed yourself is now honoured, and unbracketed on the way into the model. Because Weasel emitted most identifiers bare, bracketing the name yourself was the only way to use one that needed delimiting. That still works:

table.AddColumn("[Order Date]", "datetime2");   // names the column Order Date

But IndexDefinition.Name, TableColumn.Name, ForeignKey.Name, PrimaryKeyName and the index and FK column lists now all hold the bare name. If you compare against those, compare against Order Date, not [Order Date].

Consequence: a name that genuinely contains its own brackets can no longer be expressed — [x] names the object x.

Smaller: check constraint names honour the same pass-through; SqlServerObjectName brackets a name containing a .; GenerateDeleteAllSql emits bracketed names in its DELETE and DBCC CHECKIDENT statements.

SchemaUtils.QuoteColumnEntry never shipped

It was added in #443 and removed in #446, both inside this release. It is not part of the API.

MySQL: identifiers are normalized into the model

#445.

Table(DbObjectName) now normalizes to MySqlObjectName, so Table.Identifier.QualifiedName renders as `schema`.`name` rather than schema.name. Table(string) already went through MySqlProvider.Parse and is unaffected.

table.Identifier no longer compares equal to a plain new DbObjectName(schema, name) for the same table. That inequality was the bug — a hand-built identifier never matched the catalog, so the foreign key reported drift on every check — but it is visible to anyone comparing identifiers directly.

ForeignKey.LinkedTable is normalized too, so emitted REFERENCES clauses are backtick-quoted.

An index may now be retained that previously drifted. Where an index's leading columns cover a surviving foreign key and the expected table does not declare it, one such index is held back from the comparison so InnoDB does not refuse the DROP. Only the last remaining cover is kept.

The one that looks like a regression

Emitted DDL text changes for MySQL and SQL Server schemas that were previously emitting unquoted or invalid identifiers. Anything that hashes DDL re-evaluates on the first run after the upgrade — in particular the schema fingerprint stamps behind Migrator.UseSchemaFingerprinting.

That is a one-time "everything looks changed" pass, not drift. The second run is quiet.

Two bugs that were losing work silently

Neither was reported by a user. Both were found by the parity work in #455, and both had been failing quietly.

SQLite dropped the table to change a column

#477. TableDelta has always had a careful rebuild — create a new table, copy the surviving columns, drop the old one, rename. The migrator never called it. A change needing a rebuild reports Invalid, and Invalid was answered by dropping the object and creating it again.

Measured on a column type change: one row before, zero after — and a schema that looked entirely correct afterwards.

Affects every SQLite change that ALTER TABLE cannot express: a column type change, a foreign key added or removed, a primary key change. It needs AutoCreate.All, which is what a developer resetting a local database and db-apply on an environment configured for it both use.

If you have applied such a change on 9.24 or earlier, the data is gone and this release cannot bring it back. It will not happen again.

Oracle could not see indexes, foreign keys or primary keys

#474. ODP.NET will not execute several statements from one command, so an Oracle schema object could register exactly one introspection query — and Table spent it on columns. Everything else was invisible to SchemaMigration.DetermineAsync, which is what ApplyChangesAsync, ApplyAllConfiguredChangesToDatabaseAsync and AssertDatabaseMatchesConfigurationAsync all go through.

In practice: a declared index was created with the table and never touched again. Adding one to an existing table did nothing. Changing one did nothing. Removing one left it in place. And AssertDatabaseMatchesConfigurationAsync reported a clean match throughout.

Expect your first Oracle run on 9.25 to apply index and foreign key changes it had been ignoring. That is the backlog being worked off, not new drift.

New in this release

Views on SQL Server, Oracle and MySQL	#450. All five providers now model views. MySQL's is the interesting one — it rewrites a view's definition when it stores it, so Weasel canonicalizes your SQL through the server to compare. That needs CREATE VIEW permission on the assert path as well as the apply path; see Object Type Support.
Triggers, on all five providers	#452. The one whole category no provider modelled. They are independent schema objects that declare a target, not something a table owns — see Triggers.
Stored procedures on PostgreSQL, MySQL and Oracle	#451. SQL Server had the only implementation; it now derives from a shared StoredProcedureBase. See Stored Procedures.
Oracle packages, materialized views and synonyms; SQL Server synonyms; PostgreSQL enums, domains and composites	#453.
A shared index scenario matrix	#449. Eleven create-then-introspect scenarios run against every provider. It found three real defects on its first run, including that SQLite was comparing indexes by name alone.
Every identifier a table writes is validated	#448. Column names, the primary key constraint name and check constraint names were previously validated by no provider at all.
An interior space is a legal identifier	unit price passes validation. A leading or trailing space, a tab and a line break still do not.
Oracle teardown drops what it used to leave behind	#465. Triggers, packages, synonyms, object types and materialized views.
MySQL has schema extensions	DropSchemaAsync, CreateSchemaAsync, ResetSchemaAsync — it was the one provider of five without them.

Two new reference pages: Identifiers and Quoting and Object Type Support.

Weasel.Core API additions

Additive, but on public abstract types that downstream providers derive from:

Member	Why
ISchemaObjectWithLocalIdentifiers	Carries the names that are not database objects — columns, primary key, check constraints — so the migration path can validate them. TableBase implements it, so no provider had to.
CommandBuilderBase.Command	Exposes the command being built, so a provider's ConfigureQueryCommand can set driver options its introspection query depends on. Oracle needs it: ALL_VIEWS.TEXT is a LONG and ODP.NET reads it back empty by default.
IdentifierRules	The shared quoting contract. Each provider's static SchemaUtils delegates to it, so no DDL call site changed.
TableBase.NormalizeIdentifier	protected virtual, called from the PrimaryKeyName setter. A no-op by default; only SQL Server overrides it.
TriggerBase	The cross-provider trigger model.
StoredProcedureBase	The cross-provider stored procedure model, lifted from SQL Server's implementation.
ISchemaObjectDeltaWithRebuild	Lets a delta say "I cannot express this as an ALTER, but I can make the change without losing the data". An interface rather than a new SchemaPatchDifference value, because the enum is public and consumers switch on it.
IBatchedCommandBuilder, SchemaObjectBase.CreateCommandBuilder	Let a schema object register more than one introspection query on a provider whose driver executes one statement per command. Oracle is the only such provider.

Refused rather than ignored

Four properties that used to accept a value and silently do nothing now throw, naming the supported alternative. A caller gets what they set, or an exception — never a quietly narrower object.

Property	Why
IndexDefinition.Predicate on MySQL and Oracle	Neither engine has partial indexes. It was never emitted and never compared.
Sequence.IncrementBy on MySQL	The table-based emulation never honoured it.
Trigger.Condition on SQL Server and MySQL	Neither has a WHEN clause.
Trigger.Timing = Before on SQL Server	SQL Server has no BEFORE trigger.
AddCheckConstraint and CheckConstraints.Add on Oracle, MySQL and SQLite	Weasel does not emit check constraints there yet. Added in 9.25.2 — see above.