RFR: 8342967: Lambda deduplication fails with non-metafactory BSMs and mismatched local variables names [v5]

[Vicente Romero]

On Wed, 30 Oct 2024 13:52:52 GMT, Aggelos Biboudis <abimpoudis at openjdk.org> wrote:

>> After experimentation and under certain conditions, a few equivalent lambdas (up to variable renaming) fail to deduplicate.
>> 
>> `TreeHasher` and `TreeDiffer` are now aware that other bootstraps may appear in lambda bodies (e.g. `SwitchBootstraps.typeSwitch`) and that variable names do not matter (when they appear in method/lambda definitions). In the first case equivalence is checked based on `bsmKey` and not the `Dynamic{Var, Method}Symbol` itself.
>> 
>> The test was also adjusted since it was assuming BSM with certain structure only (that `.get(1)` was unprotected).
>
> Aggelos Biboudis has updated the pull request incrementally with one additional commit since the last revision:
> 
>   Refactor deduplication test
>   
>   - introduce two kinds of groups
>   - extract assertNotEqualsWithinGroup
>   - extract isMethodWithName
>   - extract addToGroup

corpus jobs finished, mostly same results for both jobs one with and one without this patch. Of a total of ~23940 lambdas, none of them were deduplicated. This is a bit surprising tbh. It could be that most artifacts (all?) are generated with debug information. But still it seems like the benefit of this feature is marginal

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

PR Comment: https://git.openjdk.org/jdk/pull/21687#issuecomment-2448505231