Advent of Code 2018, Day 2
I'm noticing a theme here: read an input file and build a list of values, then run two operations on that list.
For part 1, I decided to implement this as two pieces:

I started with a core type, the
Inventory
. This keeps a count of the number of twos and threes that have been seen. 
A
void compute_checksum(Inventory &inventory, string boxID)
function that operates on a reference to an Inventory value. Really, this should return a tuple, and there is a std::tuple type, but given this is single threaded and for the sake of simplicity, I decided to operate on that reference. If this was multithreaded, I'd collect all the tuples and sum them later. That adds a little bit of extra complexity, and I felt the tradeoffs were justified given the current scenario.  A
void compute_checksums(vector<string> boxIDs)
that runs through the list updating an Inventory and then returning the checksum from that.
As with yesterday, I thought about making this a streaming solution; the theme (that turned out to be repeated here) from yesterday was being able to operate on a list of values and being able to reuse that list across parts including the ability to cycle multiple times through the list. The tradeoff here is increased memory time and extra CPU time with a pair of loops, but the input sizes are small enough to warrant a more naïve approach.
I also repeated the pattern from yesterday of creating a self test function to run through all the test cases at startup. I've become fond of this because it makes sure that the algorithm is efficient enough to run through several times on simplified cases.
For part 2, I started thinking about what this needed. I made the mistake of not reading the directions well enough, though, and it took me about 15 minutes longer because I assumed the maximum difference between the two characters could be one, not that at most one character could be different. I at first thought about approaching this with sets, but there's nothing that says letters can't repeat, so I quickly disabused myself of that notion.
My implementation is similarly naïve, which I again felt was warranted given the limited input size. To compare two strings, I loop through both of them at the same time, keeping a running count of the characters that match and the number of differences. At the second difference, I return early to avoid doing extra work. Then, I basically do
for i in range(len(boxes)  1:
for j in range(i+1, len(boxes)):
check if the boxes match
if they do, return the common string
If the inputs were larger, this could get to be pretty bad:

There's O(M) time complexity and O(M) storage costs, where M is the length of the strings.

There's O(N log N) time complexity for the loop in the worst case. Again, considering the limited input size, the time available, and a desire for simplicity, I didn't try to optimize.