Implementation Details

I'm not going to cover every line of code. But I want to point out a few things. First notice that I often prefix private functions, functions that are not accessible as part of the library, with two underscores. To be consistent, I should have done this with every private function (but this was a learning project for me too).

If you look real hard, you'll probably see some gotos. I'm not writing spaghetti code. I always use them in the same way to escape from the function when something goes wrong.

Take a look at ast_strip_response2. First notice that it's invoked as a one liner from ast_strip_response. You should see a lot of these short front-end functions, and I'll explain them in a moment. Next notice how the code chooses between stripping __headers and your custom headers. Originally I wrote this so you had to specify an array of pointers and an array length. But using the NULL terminator simplifies the interface and makes changes easier. It also allowed me to simplify the implementation and add in additional functionality all at once.

The reason for all those front-end functions is to reduce duplicate code. Frequently, two or more functions, like those ending in _r and _f, will do nearly the same thing. Instead of writing two versions of the same function, I create a back-end function that has most of the logic and then invoke that function in a different way from the front-end function. The cost of this is an extra function call (which hopefully isn't much). Another reason for those functions is threading. Sometimes I need to lock a mutex inside a function. But then I need to call that function from another function, and I need to lock a mutex there also. You can't do this (think about why). One solution is to remove the mutex from the inner function. But this would require whoever is using the API to lock a mutex around the function. This is just asking for trouble. Instead, create another back-end function. Then the inner function now locks a mutex, calls the new function, and unlocks the mutex. Structuring your source so that you can reduce duplicate code while keeping the interface simple requires you to be thinking at multiple levels of abstraction simultaneously.

prev next


home