Motive:
I am just learning the fundamentals of multithreading, not close to finishing them, but I'd like to ask a question this early in my learning journey to guide me toward the topics most relevant to my project I 'm working on.
Main:
a. If a process has two threads, one that edits a set of variables, the other only reads said variables and never edits their values; Then do we need any sort of synchronization for guaranteeing the validity of the read values by the reading thread?
b. Is it possible for the OS scheduling these two threads to cause the reading-thread to read a variable in a memory location in the exact same moment while the writing-thread is writing into the same memory location, or that's just a hardware/bus situation will never be allowed happen and a software designer should never care about that? What if the variable is a large struct instead of a little int or char?
a. If a process has two threads, one that edits a set of variables, the other only reads said variables and never edits their values; Then do we need any sort of synchronization for guaranteeing the validity of the read values by the reading thread?
In general, yes. Otherwise, the thread editing the value could change the value only locally so that the other thread will never see the value change. This can happens because of compilers (that could use registers to read/store variables) but also because of the hardware (regarding the cache coherence mechanism used on the target platform). Generally, locks, atomic variables and memory barriers are used to perform such synchronizations.
b. Is it possible for the OS scheduling these two threads to cause the reading-thread to read a variable in a memory location in the exact same moment while the writing-thread is writing into the same memory location, or that's just a hardware/bus situation will never be allowed happen and a software designer should never care about that? What if the variable is a large struct instead of a little int or char?
In general, there is no guarantee that accesses are done atomically. Theoretically, two cores executing each one a thread can load/store the same variable at the same time (but often not in practice). It is very dependent of the target platform.
For processor having (coherent) caches (ie. all modern mainstream processors) cache lines (ie. chunks of typically 64 or 128 bytes) have a huge impact on the implicit synchronization between threads. This is a complex topic, but you can first read more about cache coherence in order to understand how the memory hierarchy works on modern platforms. The cache coherence protocol prevent two load/store being done exactly at the same time in the same cache line. If the variable cross multiple cache lines, then there is no protection.
On widespread x86/x86-64 platforms, variables having primitive types of <= 8 bytes can be modified atomically (because the bus support that as well as the DRAM and the cache) assuming the address is correctly aligned (it does not cross cache lines). However, this does not means all such accesses are atomic. You need to specify this to the compiler/interpreter/etc. so it produces/executes the correct instructions. Note that there is also an extension for 16-bytes atomics. There is also an instruction set extension for the support of transactional memory. For wider types (or possibly composite ones) you likely need a lock or an atomic state to control the atomicity of the access to the target variable.