Insert or Update #782
Merged
Insert or Update #782
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
- Introduce a new CompiledInsert abstraction in BasicProfile. Drivers can implement it with a compiled AST or a more abstract object that can compile to different kinds of insert statements and cache them. - JdbcProfile provides its own JdbcCompiledInsert class caches statements on first use. - Introduce StreamableCompiled (extends RunnableCompiled) and StreamingExecutable (extends Executable) to distinguish compiled and runnable collection-valued queries from other (record-valued and scalar) queries, and provide the element types. StreamableCompiled is needed here for inserts (you can only insert into collections) and can also be used later for a streaming Executor that can fully replace Invoker. - Provide a StreamingExecutable for TableQuery and BaseJoinQuery (#746) - Allow inserts from StreamableCompiled queries. Since the actual cached statement is contained in the JdbcCompiledInsert produced on demand by the StreamableCompiled, this removes the need to cache the InsertInvoker for efficient reuse of insert statements, thus providing the same kind of abstraction for cached inserts that was already used for query, update and delete operations since Slick 2.0. Tests in TemplateTest.testCompiled (inserting into a Compiled query, compiling TableQuery and BaseJoinQuery). Fixes issue #746.
- Split the code generator and mapping compiler off from InsertCompiler, using a standard CodeGen phase instead. CodeGen also gains a skeleton implementation that can be used by all existing profiles. (There is no need to inherit from CodeGen at all if it doesn't fit.) - Allow efficient building of insert statements from source queries in InsertBuilderResult. - Fix a bug in FullInsertInvoker.insert(query) that caused the query to be compiled twice. - Add an extra `baseIdentity` TableIdentitySymbol to TableNode which is initialized with the same value as `identity` but keeps that value throughout the compilation. This TableIdentitySymbol can be used to check if two TableNodes point to the same database table. So far we used the unqualified table name for that (which was wrong). - Clean up type-checking and the use of symbols in Insert nodes. - Allow inserts from pre-compiled queries (as the data source) in addition to the ability to use pre-compiled queries for the target (as added in the previous commit). - InsertColumns can now have zero or more paths. For skipped columns, an empty InsertColumn is generated, indicating that the value is part of the result type (and thus the ResultConverter) but is gets discarded. InsertColumns with zero paths or more than one path are compiled to a CompoundResultConverter. - Remove special handling for setting full and skipping data from ResultSetConverter and compile different converters instead. Ironically, that's what I tried first a couple of months ago and then aborted because it seemed like a dead end. Now the approach I took instead turned out to be the dead end. Test in InsertTest.testSimple (inserting from a compiled query).
- Refactor InsertInvokers: Split them off from InvokerComponent into a separate InsertInvokerComponent because in the JDBC implementations they contain a lot more code than all other Invokers combined. A separation into traits and protected implementation classes makes them more readable and allows InsertInvokers for "return ... into" to always work out of the box without extra customization in drivers. - Remove the unnecessary "forced" parameter from ResultConverter.set. - Add an UpsertBuilder to JdbcStatementBuilderComponent which builds a standard SQL MERGE statement by default. - Add CheckInsertBuilder and UpdateInsertBuilder for use in emulated client-side insertOrUpdate operations. - Add JdbcProfile.capabilities.insertOrUpdate to indicate native insertOrUpdate support in a driver. - Move scalarFrom out of QueryBuilder to JdbcProfile (UpsertBuilder needs it, too). - Make ResultConverterCompiler stateless and use the index from the path into the ResultSetMapping instead of enumerating columns there. In the future, this could also be used for query operations to allow transformations of client-side mappings. Driver support (which makes you wonder why we even bother with native MERGE support, given how limited it is): - Hsqldb uses the standard MERGE syntax but cannot return generated keys from a MERGE, so we have to use the client-side emulation when returning keys. - MySQL requires a custom upsert command with good semantics. We can use it in all cases. - H2 and SQLite require custom upsert commands and can only perform upserts with forced insert semantics, so we have to use the client-side emulation for tables with AutoInc columns. - PostgreSQL uses a multi-statement server-side emulation. - The client-side emulation is used for Access and Derby. Fixes issue #6. Tests in InsertTest.testInsertOrUpdatePlain and InsertTest.testInsertOrUpdateAutoInc.
|
rebased on master |
|
What happy if pass None to a option id and that is AutoInc? |
|
That should work the same way as the AutoInc test case. It'll compare to null instead of 0, which will also fail every time. The rest is the same (since there is no difference between Option[T] and T in the generated code) |
|
LGTM |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
A ton of work to close our longest-standing feature request. Includes lots of refactoring of the insert-related code, including a new way to compile mappings from InsertCompiler which allows client-side transformations and gets rid of the requirement to use all columns in the order in which they appear in the lifted value. We should be able to reuse that later for all queries (might be useful for nested options and nested collections).
Fixes #6 and #746. Review by @cvogt.