semantic/weekly/2016-06-06.md

May 31th, 2016

Agenda

1. Retrospective on last week:
   • What went well?
   • What was challenging?
   • What did you learn?

What went well?

@robrix:

• As of this morning we’re pushing data to graphite.
• There's a 2 hour window of when data is pulled in to graphite.
• We have a much more thorough understanding of the shape of the SES problems.

@rewinfrey:

• Came away with a much more thorough understanding of SES.
• Pairing with Rob on the profiling issues.

@joshvera

• Exploring the current productions of tree-sitter output.
• Got an airbnb for the Mini-Summit.

What was challenging?

@robrix:

• SES performance has been a problem. Briefly confused into thinking we had solved the problem. We’re profiling semantic-diff-tool and running the tests on semantic-diff. If we don’t clean semantic-diff-tool then it doesn’t know that semantic-diff has been rebuilt and it doesn’t try to relink it. Happens specifically when changing branches.
• SES performance depends on O(n) cost function.

@rewinfrey:

• Heartbreaking to discover that a huge performance win was a bad build.
• Felt like it was hard to contribute to the deployment and build process.
• Do we have a fallback in place in case S3 fails?

@joshvera:

• Error productions from tree-sitter are difficult to debug and obscure diff summary output.
• Understanding and communicating how our deployment process works to other people. Maybe this means we need better documentation?

What did you learn?

@robrix:

• Learning about parallelism because we have large asymptotic factors in SES.
• Developed a stronger intuition for why cost has to be linear with respect to the size of the diff tree.

@rewinfrey:

• Learned a lot about profiling in Haskell.
• Learned how to use Profiteur to visualize the space and time costs for a given computation.
• Trying to use the Eval monad to parallelize the Minimax algorithm.

@joshvera:

• Learned about designing CRISPR proteins that can be edited into bacteria to defend against viruses and plasmids.
• Read up on GHC's Core language in order to understand some of the optimizations GHC performs.

Other Items

• Mini-Summit plans set for the week of June 20th.
• Rob on vacation starting Tuesday June 7th! 😎