James Iry

Useless but also somehow vital fact: 2/22/2022 is going to be a Tuesday

@nickhoward Worth every penny and you can often pick them up used. I got mine from madisonseating.com/?subcats=Y&pco…

Should I buy a Herman Miller Aeron?

Dunning-Kruger? I don't know what that is, but I bet I'd be really good at it.

My team at Stripe is working on an experimental ahead-of-time compiler for Ruby and we’ve moved our development to open source. sorbet.org/blog/2021/07/3…

Schrödinger's Dumpster!

My wife caught a neighbor peeping through the window at her!

@nikhilbarthwal @pvsukale @patrickc As for for creating a new language that would mean replacing not only the VM but also all the development tools we use on our Ruby, like linters and editors. That would greatly extend to our “time to value”

@nikhilbarthwal @pvsukale @patrickc It also allows us to release our compiler gradually. If there’s some Ruby we don’t compile correctly or efficiently we can just let the interpreter run it until we do.

We're big believers in multi-year infrastructure bets. After a few years of Ruby infra work, our in-house Ruby compiler is now 22–170% faster than Ruby's default implementation for Stripe's production API traffic. If interested in working on such problems, we're hiring!

@nikhilbarthwal @pvsukale @patrickc Not quite. We’re not replacing the MRI VM. What we’re doing is compiling Ruby code into native code and registering it with the VM. The VM and our compiled code interact via the MRI’s FFI mechanism

@jamesiry @pvsukale @patrickc So would I be right in saying you are doing something similar to Facebook’s hip-hop virtual machine for Hack?

@cperciva @_ColeMurray @patrickc Sorry, I can't say publicly what percent of latency is attributable to what cause, but I can say given how critical purchase flows are to our customers, every aspect of latency is worth our attention. Want to know more? We're hiring :)

@jamesiry @_ColeMurray @patrickc This surprises me even more. You're saying that customer-observed request latency isn't overwhelmingly dominated by network latency?

Mathematical Bumper Sticker

@JorgeO We're still a less than 1 pizza team working on the compiler. Want some pizza, Jorge?

Also, Ruby’s inputs+outputs are very programmatic, fairly well defined, easy for computers to compare. A two-pizza team can work on Ruby. Stripe’s inputs+outputs are highly complex, evolve with products+users+partners+regulators, etc. A team of 1000+ engineers work on Stripe.

This is non-obvious! All of Ruby is X lines of code, and steady/growing slowly. All of Stripe is 10X-100X lines of code, and growing quickly.

@dbreunig @patrickc You got it! For two main reasons. The first is that Sorbet is the "front end" of our compiler (the parser and semantic analyzer+typer). The second is that many of the compiler's optimizations depend on the static type information that Sorbet provides.

@DavidjMarkey @patrickc Keep your eye on sorbet.org . We plan to open source the compiler as an extension to our Ruby type checker.

@patrickc Gonna open source it then?

@baedert @patrickc Keep your eye on sorbet.org . We plan to open source the compiler as an extension to our Ruby type checker.

@patrickc Upstream when?

@iamwhoisme @patrickc Keep your eye on sorbet.org . We plan to open source the compiler as an extension to our Ruby type checker.

@patrickc Yeah, you gonna open that up?

@dbreunig @patrickc Keep your eye on sorbet.org . We plan to open source the compiler as an extension to our Ruby type checker.

@ilaguiar @patrickc Keep your eye on sorbet.org . We plan to open source the compiler as an extension to our Ruby type checker.