Skip to content

TODO.md

Latest commit

 

History

History
52 lines (35 loc) · 2.98 KB

TODO.md

File metadata and controls

52 lines (35 loc) · 2.98 KB
Raw

TODO

Known gaps, in rough priority order. Everything here is a deliberate omission rather than an unknown, so each entry says what is missing and what it would take.

1. Nothing compiles the MySQL or Oracle generated output

Tester.Integration.EFCore8/9/10 compile the SQL Server output, which is what turns a bad mapping into a build failure. MySQL and Oracle have goldens (Generator.Tests.Integration/TestComparison/*_MySql_*, *_Oracle_*) but nothing compiles them, so a method the provider does not define only shows up if somebody reads the diff.

That is exactly how Settings.ColumnIdentity came to emit .UseIdentityColumn() for MySQL, which Pomelo spells UseMySqlIdentityColumn - the generated code would not have compiled, and no test said so.

Fix: a tester project per dialect referencing Pomelo.EntityFrameworkCore.MySql and Oracle.EntityFrameworkCore. This is the largest remaining hole.

2. No stored procedure return models for MySQL or Oracle

Neither reader implements ReadStoredProcReturnObjects, so procedures generate without result classes. PostgreSQL builds them from the catalogue; SQL Server executes with SET FMTONLY. Oracle can be done from ALL_ARGUMENTS; MySQL has no catalogue for it, so it would need a prepared-statement round trip.

3. Oracle reads exactly one schema per run

OracleDatabaseReader scopes every query to SYS_CONTEXT('USERENV','CURRENT_SCHEMA'). SQL Server and PostgreSQL span schemas in a single context; Oracle cannot. The change is contained - take a schema list from a setting and replace the CURRENT_SCHEMA predicate with an IN list - but nobody has asked for it.

4. MySQL cannot see across databases

MySQL treats a database as a schema, so connecting to EfrpgTest makes the EfrpgTestOther tables invisible. TestDatabases/MySQL/EfrpgTest.sql creates that second database for the cross-schema foreign key case, so the script contains the scenario but the test does not actually exercise it. Fixing it means teaching the MySQL reader to read across databases.

5. Oracle synonyms are not read

SynonymsSQL deliberately returns an empty string. ALL_SYNONYMS is there whenever someone asks for it.

7. MySqlToCSharp maps tinyint to SByte

The BCL name, where every other entry in every mapping uses the C# keyword. It compiles and DatabaseReader.MapType accepts both spellings, so this is cosmetic. Left alone deliberately: changing it churns both copies of the mapping, DatabaseReaderSqlTests and a golden, for no behavioural gain.

8. Language mapping keys are no longer compared between the two repositories

WireContractTests replaced ParallelSourceTests when the tool moved to its own repo, and it deliberately does not compare mapping keys: an unknown key legitimately falls through to each mapping's default - that is how every SQL Server character type is typed - so a missing key cannot be asserted mechanically. Only the integration goldens catch it, and only for a dialect they cover. Item 1 would close most of this.