revert
D25302218
Summary:
Now that I've fixed perf of List.length in Typing_check_service.next, I want to see if
D25302218 (https://github.com/facebook/hhvm/commit/
8c8c8e78d43e6b9172d6d24b257a54bcb3490959) was worth it. Here are some representative numbers:
* Full init, 56cores:
* With GC and SharedMem.Invalidate: 1294s
* With GC and Ast_provider.Invalidate: 1462s
* Without any GC attempt: 1307s
Here's a wider summary based off more scenarios and more samples:
* 56core: GC attempts in workers bring 1% improvement on full.init, and 5% regression on ss.change
* 80core: GC attempts in workers brings 2-3% improvement on full.init, and 3-4% regression on ss.change
Reviewed By: andrewjkennedy
Differential Revision:
D25523992
fbshipit-source-id:
6bc5d1f2c77f1073242ff001d8957af8e1fa7f16