[PATCH] D82598: [analyzer][Liveness][NFC] Get rid of statement liveness, because such a thing doesn't exist

[Artem Dergachev via Phabricator via cfe-commits]

NoQ added a comment.

In D82598#2116373 <https://reviews.llvm.org/D82598#2116373>, @xazax.hun wrote:

> > I suspect that it would be pretty bad if you, say, kill the condition of the `if`-statement before picking the branch. Or kill the initializer in the `DeclStmt` before putting it into the variable (same for `CXXCtorInitializer` which isn't even a `Stmt`!).
>
> I would argue that the end of the full expression is AFTER the `if` was evaluated in this case. But I do see what you mean, thanks :)


Well, formally it isn't. The notion of full-expression is defined pretty strictly because that's the moment of time when temporary destructors are invoked. So the language pays a lot of attention to what exactly constitutes an expression, and as of now if-statements don't. In particular, temporary destructors will run before the branch is chosen.

I guess your point is that live expressions analysis could be replaced with imperative cleanup requests from, say, `ExprEngine` to the `Environment`. I.e., "We've just finished evaluating the if-statement, now we should actively tell the Environment to remove the condition expression". I guess it could totally work that way, but it'd be pretty hard to not forget such cleanups. Given that we'd also barely ever notice that we forgot one of those, i'm very much in favor of having liveness analysis instead, that would declaratively describe which expressions are live when, so that to automatically guarantee that we always only track what's necessary and never snowball our state with dead expressions.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D82598/new/

https://reviews.llvm.org/D82598