Hard Things in Computer Science

The post pushes back on the claim that only cache invalidation and naming things are hard in computer science. It says many other areas are difficult, including time-related edge cases, project estimates, distributed coordination, and proving correctness.

Reading notes

#

• Cache invalidation depends on choosing a TTL that balances stale reads against unnecessary reloads.
• Naming is hard because code must be precise, and different terms can hide different meanings between business and developers.
• Dates, times, and timezones are complicated by calendar changes, DST, timezone shifts, and uneven offsets.
• Estimates are difficult because software work differs from house building, unexpected problems appear, and estimates are often treated as deadlines.
• Distributed systems are easy to get wrong because network assumptions fail, and coordination problems like dual writes and leader election are hard to solve.
• Dual writes may require 2PC, compensating transactions, or CDC, but consistency across stores is still only eventual in practice.
• Leader election depends on consensus, and algorithms like Paxos and Raft are meant to handle partitions and re-election.
• Bug-free code cannot be guaranteed by testing alone; the reliable route is formal proof, which still remains mostly in academia.