Catfish-Man

Ok, so.

A processor has things called "registers", which are essentially its local variables. Anything it wants to operate on, it loads from memory (RAM) into registers, does the operations, and then stores them back to memory. One of those registers is the "program counter", which holds the location of the next instruction to run. Now, there's a fixed quantity of registers, while you're allowed to have unlimited local variables, so there's also a dedicated region of memory for storing local variables that aren't currently in registers, which is called "the stack" (it's a stack because each time you call a function its variables get added to the top, and each time you return from a function they get taken off the top).

A thread is:
 * The state of all the registers
 * The stack
 * A pointer for the operating system (the kernel scheduler specifically) to keep track of it
 * A few odds and ends of metadata and extra state like priority and such

Catfish-Man

As of iOS 11/macOS High Sierra, and only including ones in Foundation and CoreFoundation

Strings:

_NSCFString - a CFStringRef or CFMutableStringRef. This is the most common type of string object currently.
    - May have 8 bit (ASCII) or 16 bit (UTF-16) backing store
_NSCFConstantString - a compile time constant CFStringRef, like you'd get with @"foo"
    - May also be generated by dynamic string creation if matches a string in a pre-baked table of common strings called the StringROM
NSBigMutableString - an NSString backed by a CFStorage (https://github.com/opensource-apple/CF/blob/master/CFStorage.h) for faster handling of very large strings
NSCheapMutableString - a very limited NSMutableString that allows for zero-copy initialization. Used in NSFileManager for temporarily wrapping stack buffers.

Catfish-Man

import Foundation

func examine(_ str: String) {
    print("\"\(str)\"")
    print("bridges as: \(type(of: str as NSString))")
    str.withCString {
        print("inits an NSString as: \(type(of: NSString(cString: $0, encoding: String.Encoding.ascii.rawValue)!))")
    }
    print("\n")
}

Catfish-Man

/*
This code is completely untested but should be close enough to correct to demonstrate the concepts
*/

struct RegularMonkey {
  var bananaCount: Int
}

struct KlingonMonkey {
  var bananaCount: Int

Catfish-Man

#import <Foundation/Foundation.h>
int main(int argc, const char * argv[]) {
    @autoreleasepool {
        const long count = 10000000;
        const char *str1 = "hellohello"; //fits in a tagged pointer
        const char *str2 = "hellohelQo"; //at this character count, can only fit in a tagged pointer by using <8 bits per character, which requires dropping less frequently used chars like 'Q'
        const char *str3 = "Northern Sami"; //this string is encountered frequently enough that it's baked into a perfect hash table in CoreFoundation
        const char *str4 = "hellohellü"; //this string can't be tagged *and* can't be stored using an ASCII backing store

        NSDate *start = [NSDate date];

Catfish-Man

import Foundation

func test() {
  let object = NSObject()
  let cocoa = NSMutableArray()
  var swift = Array<AnyObject>()

  var start = Date()
  for _ in 1 ... 200_000 {
    cocoa.insert(object, at: 0)

Catfish-Man

#import <Foundation/Foundation.h>
#import <assert.h>

//Compile with `clang -Os -framework Foundation -fno-objc-arc inlinestorage.m -o inline, run with `inline clever` or `inline naive`

/*
 NaiveArray implements a simple immutable NSArray-like interface in probably the most obvious way: store a pointer to a C array of objects
 */
@interface NaiveArray : NSObject {
    NSUInteger count;

Catfish-Man

#import <Foundation/Foundation.h>
#import <time.h>

#define USE_CF_NONOBJC 1

#if USE_CF_NONOBJC
extern CFTypeRef _CFNonObjCRetain(CFTypeRef cf);
extern void _CFNonObjCRelease(CFTypeRef cf);
#endif

Catfish-Man

Let's Reinvent Modern CPU Caches!

In The Beginning, programs were hard-coded, entered directly with switches. Values would be input, and then results would output,
but couldn't really be stored. We'll draw this like so:

Input -> Fixed Calculations -> Output

An early improvement in generality was the addition of storage (ENIAC eventually gained 100 words of magnetic core memory),
leaving us with something along these lines:

Catfish-Man

// Compile with clang -framework Foundation sethack.m

#import <Foundation/Foundation.h>
#import <objc/runtime.h>

/*
 CFHashBytes from http://www.opensource.apple.com/source/CF/CF-1153.18/CFUtilities.c
 */
#define ELF_STEP(B) T1 = (H << 4) + B; T2 = T1 & 0xF0000000; if (T2) T1 ^= (T2 >> 24); T1 &= (~T2); H = T1;