I am confused by this loop control. You start with a random garbage on stack for the initial buf[] content.

ptr = buf, means address of buf. buf is filled with a valid string after the getenv_string() call.

The comment is useless.

Ah, I missed it.

Style: declare vars at the top of the function.

Well, almost all of the comments trivially rephrase the code.

This accepts arbitrary garbage after the numeric part.

Now I am confused by this comparision. Valid range for unsigned char on all supported arches is 0..255.

Same note for uint16_t.

And there.

I'll fix.

I think it would be simpler to just add a 'bool signed' parameter instead of overloading type_size. type_size can then always be 'sizeof(foo)' which would be easier to read at the call-site. If you'd rather not use a bool, then define a simple enum:

enum sign {
    SIGNED,
    UNSIGNED,
};

And use that instead so you have something like this for CTLTYPE_UINT:

error = getenv_array(path + rem, data, sizeof(data), &size, sizeof(int), UNSIGNED);

Using a uint64_t array initially threw me off. I wonder if it wouldn't be more accurate to declare this as 'char data[1024] __Alignas(uint64_t)' or the like?

0 should be false, since you use bool as the function argument type.

1 -> true

Wrap the line.