Here are my cryptic notes from last time:

Victor: Intended mostly to discourage large user functions.

Make it clear that we are not saying whether STL is in same translation unit.

Should we ban atomics in transactions?

M. Spear: Ban. Behaviors for data-race-free programs vary

M Scott: All instances of a common semantics?

M. Scott: Users will communIcate within transaction

Consensus: Ban it for now.

Ban in the same way as exceptions: Undefined Behavior.

Require static checks for e.g. atomics in transaction?

Michael Spear: All within compilation unit

M Scott: Need undefined if we will weaken restrictions later

Hans: Already unclear which piece of code will be dynamically called in transaction.

Add discussion in alternatives section that we decided against requiring static detection of disallowed operations.