Part 1
Clearly Immutability minimizes the need for locks in multi-processor programming, but does it eliminate that need, or are there instances where immutability alone is not enough? It seems to me that you can only defer processing and encapsulate state so long before most programs have to actually DO something (update a data store, produce a report, throw an exception, etc.). Can such actions always be done without locks? Does the mere action of throwing out each object and creating a new one instead of changing the original (a crude view of immutability) provide absolute protection from inter-process contention, or are there corner cases which still require locking?
I know a lot of functional programmers and mathematicians like to talk about "no side effects" but in the "real world" everything has a side effect, even if it's the time it takes to execute a machine instruction. I'm interested in both the theoretical/academic answer and the practical/real-world answer.
If immutability is safe, given certain bounds or assumptions, I want to know what the borders of the "safety zone" are exactly. Some examples of possible boundaries:
- I/O
- Exceptions/errors
- Interactions with programs written in other languages
- Interactions with other machines (physical, virtual, or theoretical)
Special thanks to @JimmaHoffa for his comment which started this question!
Part 2
Multi-processor programming is often used as an optimization technique – to make some code run faster. When is it faster to use locks vs. immutable objects?
Given the limits set out in Amdahl's Law, when can you achieve better over-all performance (with or without the garbage collector taken into account) with immutable objects vs. locking mutable ones?
Summary
I'm combining these two questions into one to try to get at where the bounding box is for Immutability as a solution to threading problems.
Best Answer
This is an oddly phrased question that is really, really broad if answered fully. I'm going to focus on clearing up some of the specifics that you're asking about.
Immutability is a design trade off. It makes some operations harder (modifying state in large objects quickly, building objects piecemeal, keeping a running state, etc.) in favor of others (easier debugging, easier reasoning about program behavior, not having to worry about things changing underneath you when working concurrently, etc.). It's this last one we care about with this question, but I want to emphasize that it is a tool. A good tool that often solves more problems than it causes (in most modern programs), but not a silver bullet... Not something that changes the intrinsic behavior of programs.
Now, what does it get you? Immutability gets you one thing: you can read the immutable object freely, without worrying about its state changing underneath you (assuming it is truly deeply immutable... Having an immutable object with mutable members is usually a deal breaker). That's it. It frees you from having to manage concurrency (via locks, snapshots, data partitioning or other mechanisms; the original question's focus on locks is... Incorrect given the scope of the question).
It turns out though that lots of things read objects. IO does, but IO itself tends to not handle concurrent use itself well. Almost all processing does, but other objects may be mutable, or the processing itself might use state that is not friendly to concurrency. Copying an object is a big hidden trouble point in some languages since a full copy is (almost) never an atomic operation. This is where immutable objects help you.
As for performance, it depends on your app. Locks are (usually) heavy. Other concurrency management mechanisms are faster but have a high impact on your design. In general, a highly concurrent design that makes use of immutable objects (and avoids their weaknesses) will perform better than a highly concurrent design that locks mutable objects. If your program is lightly concurrent then it depends and/or doesn't matter.
But performance should not be your highest concern. Writing concurrent programs is hard. Debugging concurrent programs is hard. Immutable objects help improve your program's quality by eliminating opportunities for error implementing concurrency management manually. They make debugging easier because you're not trying to track state in a concurrent program. They make your design simpler and thus remove bugs there.
So to sum up: immutability helps but will not eliminate challenges needed to handle concurrency properly. That help tends to be pervasive, but the biggest gains are from a quality perspective rather than performance. And no, immutability does not magically excuse you from managing concurrency in your app, sorry.