RFR: 8342967: Lambda deduplication fails with non-metafactory BSMs and mismatched local variables names [v2]

[Liam Miller-Cushon]

On Fri, 25 Oct 2024 15:11:50 GMT, Maurizio Cimadamore <mcimadamore at openjdk.org> wrote:

> > side: not sure if the dedup effort is worthy at the end in terms of performance / class file reduction / maintenance
> 
> Good point - these deduplication tests are rather painful to maintain every time the compiler is changed (I had to fight with them as well when writing the revamped backend support for flexible constructor bodies). I share the general feeling of a slightly negative cost vs. benefit ratio -- but I don't have any backing evidence (maybe we should gather that up).

I don't have good data on what the remaining benefit is either, but for what it's worth I share the feeling that this feature may not be worth keeping around. The deduplication isn't performed if debug info is enabled to avoid incorrect line numbers in stack traces, so there's a large fraction of code that enables debug info and isn't benefiting from it.

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

PR Comment: https://git.openjdk.org/jdk/pull/21687#issuecomment-2438144066