Julia High Performance - Second Edition

The trouble with globals

In the previous chapter, we saw how Julia achieves its high-performance runtime by compiling specialized versions of functions for particular types of arguments – a process that relies on type inference using data flow techniques. However, global variables can be written to at any time, and by any code. The compiler cannot keep track of all writes to the global variables; this would be akin to solving the halting problem. Therefore, the data-flow technique fails to perform any inference for these types of global variables. As a result, the compiler cannot create specialized functions when using these variables.

To understand the performance hit of using the global variables, let's use a simple function that calculates ...