Talk 2: Ryan Newton

[rrnewton]

22 minutes -- too long!

[rrnewton]

Sam: throw away part 1.

[samth]

• Cut all of Part 1.
• Have some benchmarks that show CDS/pure-in-a-box being bad at the beginning.
• Have a more detailed walkthru of the hybrid part.
• Move thms to after the data structure stuff, before benchmarks.
• Put times in charts in a sensible unit
• Cut the Java benchmarks
• Say something about the IU pictures.

[cgswords]

Conference Question: Both 50/50 have "unexpected" 2-thread cases. Why?

---

I didn't like the slides that had gray-wash background. I know they were meant to "break up" the talk, but it was difficult to read them.

Also, quite a few bullets (I'm sure Sam will say something...)

The java test slide is rough. Why not more slides, just faster?

[jasonhemann]

Wonderfully quick tour of the problem space to precisely the topic of the talk. I liked the "here's what we're not talking about".

[eholk]

I like that you mentioned the incorrect approach from earlier GHC versions.