Conversation
danyadev
changed the title
danyadev
force-pushed
the
java-asm-transformer
branch
from
0edf80f to
2c6124f
Compare
saves ~50-60ms in the full pack
it cut the transforming time from 200ms to 110ms
stabilized on ~95-100ms
…ters' own classes
danyadev
force-pushed
the
java-asm-transformer
branch
from
2c6124f to
252a6ba
Compare
|
Huh, this looks cool, it's also impressive how bad it was before, lol... Didn't know about ClassConstantPoolParser in gtnhlib, I need to remember that, cheaper than parsing things each time and I bet this is where the majority of the performance improvement comes from. The recursive check in isTileEntity before wasn't great either, iteration is better here. The copyMethodNode I really like here, previously it... looks like it was putting MethodNode objects from a template being inserted directly into multiple classes? Which I'm pretty sure was violating some invariants... o.O Also looks like it checks class implementations now for interfaces instead just dumb stripping, much nicer! The more conversion from scala to java is simple enough for this, probably more maintainable by people into the future, lol. (Really though, that prior code was... not great? Good PR!) |
Alexdoru
left a comment
Alexdoru
left a comment
There was a problem hiding this comment.
If you are adding GTNHlib as a dep can you also make each transformer dump the classes when a debug flag is set ? like it does in hodgepodge, gtnhlib and angelica
4 participants
Optimized transform time in the full pack from 850ms to 100ms
Tested in the full pack, confirmed that every transformer works