Chen Liang
Chen Liang
/open This is becoming more necessary as valhalla brings changes to signature format.
Sorry for the force push - Too many commits meaning fixing merge conflicts in every single one of them was a bit complicated.
Keep alive. One day there will be a reviewer...
After a revisit to this PR, I think I might split the work of improving BytecodeDescriptor error message and tests covering failure-case behaviors into a prerequisite PR to this one,...
Will we incorporate the changes for JDK-8 target code as in #19151?
@wenshao As I recall, @cl4es explained that some infrastructures initialized after LambdaMetafactory use explicit anonymous classes instead of lambdas due to startup performance concerns: https://github.com/openjdk/jdk/pull/18690#discussion_r1562661264 Also see #19140: The evaluation...
I think for string concat this in particular is safe; since if a class has indy/condy, that class can still be loaded and initialized in early bootstrap as long as...
I think we might have to close this PR for now, as we seem to be defensive here. Java's early startup sequence indeed isn't well defined and is very susceptible...
Also another question for @cl4es: what would you think of using lambdas in generic accessor methods in reflection? Is it agreed upon that annotation/generics are initialized after invoke?