John Hawthorn

Marshal is mostly a 32-bit format (presumably matching the 32-bit Ruby it was originally implemented for). This means that large FIXNUMs (> 31 bit) on 64-bit platforms still need to...

New constant caching instruction: opt_getconstant_path

2

_I previously shared an early experimental version of this in #6049, the main change from that PR is finishing the implementation, {Y,M}JIT support, and moving the `ID` segment list into...

Store array of original_insns on iseq

1

This aims to fix https://bugs.ruby-lang.org/issues/18269 as well as making GC iseq marking (and anywhere else we iterate through the instructions) faster by avoiding the hash lookup inside `rb_vm_insn_addr2insn`. This is...

Avoid moving REG_SP (make it a base pointer)

5

This is an idea to improve how we handle stack addresses to make codegen simpler to write and possibly improve performance. Currently, `REG_SP` always has the same value of `reg_cfp->sp`,...

Add support for -F

2

[Suggested by ragepurification on reddit](https://www.reddit.com/r/ruby/comments/4mudm1/curltoruby_convert_curl_commands_into_nethttp/d3z0fnj)

Use find_by_sql instead of select_all and send(:instantiate)

1

These are both private APIs, but _load_from_sql will give us even more standard behaviour (this is what find_by_sql calls) including emitting the instantiation.active_record notification. This may be faster in some...

This adds a fast path for partial lookup on "simple" partial renders. In this case "simple" means: 1. This is a plain partial render, like `render "foo", {...}` or `render...

UnitCancel seems to have a few issues with no simple solutions. We had some PRs with failing specs illustrating those issues: * #1455 Failing test: UnitCancel with loaded associations *...