Chapter 10:

1. The most important constraint is that the result (or, more strictly, the observable behavior) of the program must not change. The bar here is high: the compiler is allowed to optimize only when it can be proven that the results are correct for all possible inputs. The second consideration is practicality: the compiler has to make tradeoffs between compilation time and efficiency of the optimized code. Even with the highest optimization enabled, it may be too expensive to prove that some code transformations do not break the program.
2. In addition to the obvious effect (elimination of the function call), inlining enables the compiler to analyze a larger fragment of code. Without inlining, the compiler generally has to assume that "anything is possible" inside a function body. With inlining, the compiler can see, for example, whether the call to the function produces any observable behavior, such as I/O. The inlining is beneficial only up to a point: when...