YuMingLiao

 I don't understand how on-disk representation would reduce memory usage. But it's working. The memory usage is down from ~240M to ~160M. Nice punch!

I couldn't find benchmark/OnDiskClient.hs for reproducing. I guess it's in your .gitignore or what.

At the first version in that branch, my server already started to be able to work! Thank you! I'll try the second one and report back.

 8 fields x 80000 tuples Now it's 13MB on disk. 90MB on mem. It's pretty amazing. Thanks!

I encounter this issue too when I want to record tree-like comments. I found an article talking about tree structure in sql (in Chinese). [Tree structure in SQL](http://blog.rangilin.idv.tw/tree-structure-in-sql/) Basically, it...

I tried adding `StepIO` in `Syn`. But `orr` is left-biased. In my implementation attempt, any IO action in `orr` will definitely get finished and then the most left result will...

> an event can both be caused and perceived at the same time `advanceIO Or` run `advanceIO p` then pass eid and ios to `advanceIO q`. If I implement no-bias...

After reviewing the concepts of synchronous and asynchronous, I guess that Syn.hs is for synchronous programming and unbiased orr belongs to Concur.hs.

Still not sure where a thread-cancellable orr should belong to. Concur.hs here uses STM to step, which leaves IO actions excluded.

It seems that logical time `orr` (events) and real time (IO actions) `orr` should be separated for single responsibility principle. Otherwise this [threading or not](https://github.com/ajnsit/concur/pull/16#discussion_r297401831) issue will happen again. Is...