Today I've found a very nasty problem during development of cdist 1.8.0, related to shell functions and the non-local variables.

Background

cdist, up to version 1.7.x, is implemented in shell scripts. As it turned out, most of the time during a cdist call is spent within the kernel, which seems to be related to some thousands of forks we do for each run (you can use oprofile to verify this yourself). To speedup cdist, the idea was to rewrite cdist to use functions for internal functionality, instead of the shell scripts.

The implementation

So the idea was very simple:

• Take all existing shell scripts
• Add a function header
• Move shell script from bin/ to core/
• Source all functions from core/

But the first problem that came up is that the local keyword had been removed from the POSIX standard. Thus I began to eliminate global variables by either prefixing them with the function name or by removing them completly.

The problem

After a some time I realised that the __cdist_object_run() function calls itself recursively. This leads to subsequent overwritten variables, which in turn leads to unwanted behaviour.

The question

I'm aware that the local keyword is still supported by some shells, but I am wondering, whether you have a good idea, on how to speedup cdist (having fast fork()s would be great!) without violating the POSIX standard?