[code-reflection] RFR: Improved NormalizeBlocksTransformer and its application in BytecodeGenerator. [v6]

[Paul Sandoz]

On Thu, 15 Jan 2026 08:25:32 GMT, Adam Sotona <asotona at openjdk.org> wrote:

>> Lowering of pattern matching in switch statements and expressions produces an inefficient tree of conditional branches. This PR does not improve the lowering itself; however, it reduces overhead in the generated bytecode. 
>> 
>> `NormalizeBlocksTransformer` is improved to handle redundant conditional branches. It skips intermediate conditional branches with constant boolean arguments and re-targets them directly to the true or false branch, based on the argument's constant value.
>> 
>> The `BytecodeGenerator` prepends `NormalizeBlocksTransformer` to leverage the optimized models for bytecode generation.
>
> Adam Sotona has updated the pull request with a new target base due to a merge or a rebase. The pull request now contains eight commits:
> 
>  - Merge remote-tracking branch 'babylon/code-reflection' into skip-cbranch-transform
>  - fixed javadoc
>    
>  - added test
>    
>  - removed debug prints
>    
>  - Merge of SKIP_CBRANCH_TRANSFORM with NormalizeBlocksTransformer
>    
>  - Merge remote-tracking branch 'babylon/code-reflection' into skip-cbranch-transform
>  - removed debug print
>    
>  - Skip constant conditional branch transformation.

Marked as reviewed by psandoz (Lead).

src/jdk.incubator.code/share/classes/jdk/incubator/code/analysis/NormalizeBlocksTransformer.java line 140:

> 138:                 && cbo.trueBranch().arguments().isEmpty()
> 139:                 && cbo.falseBranch().arguments().isEmpty();
> 140:     }

Eventually i hope we can represent this as a pattern.

-------------

PR Review: https://git.openjdk.org/babylon/pull/829#pullrequestreview-3666740164
PR Review Comment: https://git.openjdk.org/babylon/pull/829#discussion_r2695287930