Compilers and Interpreters
Compiler & Interpreter
What are compilers and interpreters?
- Compiler = translates everything up front
- Interpreter = live interpreting, one line at a time
- Speed and approach by language
Whether a language is compiled or interpreted is not a matter of developer taste: it changes 'when you discover errors' and 'what you deploy'. A compiled language catches mismatched types or misspelled function names at the build step, before anything runs. An interpreted language fails only when execution reaches that line. Incidents where an overnight batch runs fine for five hours and then stops at the final aggregation because of a single typo mostly come from this. Deployment differs too: with one, you just move a single executable; with the other, you must align the source, runtime version, and libraries on every server.
Today’s languages don’t fall neatly into one camp or the other. Java and C# compile source code into bytecode that a virtual machine runs, and a JIT recompiles frequently executed sections into machine code at runtime. Python also creates a .pyc cache, and JavaScript’s V8 engine uses a JIT too. TypeScript is closer to a transpiler that converts code to JavaScript. So the distinction that matters in practice is 'does the deployment pipeline have a build step?' With a build, you wait a few minutes in CI but the release artifact is fixed; without one, changes take effect immediately but each server’s environment may differ, leaving room for failures.
The most common misconception is 'Python is slow, so let's rewrite it in C or Java.' In practice, after a rewrite, the bottleneck usually turns out to be DB queries and network round trips, not the language, and numerical libraries are already written in C internally. It is not unusual for a team to spend six months porting, see little noticeable improvement, and roll it back. There is a misconception in the opposite direction too: 'it compiled, so it is safe.' A compiler checks only syntax and types, not calculation logic or data quality. The same goes for 'if we don't hand over the source, our technology is protected': executables can be decompiled and reverse-engineered.
Jaryong KooMarketing PhD and leading authority on AI, data analytics, and personal branding
Seungtae ParkAX practitioner who builds AI business development and marketing automation tools and teaches from that hands-on experience| Category | Compiler | Interpreter |
|---|---|---|
| Method | Translated all at once, ahead of time | Interpreted live, line by line |
| Execution Speed | Fast | Relatively slower |
| Yes | C · Java | Python · JavaScript |
A foundational computer science concept: the two main ways programming languages are processed.
Let’s look at a hypothetical case: Company A, a chemicals distributor with 320 employees and KRW 180 billion (≈ US$134,670,100) in annual revenue. It runs its month-end settlement with 40+ Python scripts, starting at 2 a.m. and not finishing until 8 a.m. Last March the run failed at the final step, and the sales team’s close slipped by half a day.
- Kim, a deputy general manager on the dev team, dug through the logs and found the job had run normally for 5 hours 40 minutes, then stopped on a single typo in an argument name in the final aggregation function. The team lead's first words were, "Was there really no way to know this before we ran it?"
- The CTO put the full Java rewrite proposal on hold and instead added type hints and static analysis tools to CI, running a two-minute pre-check on every commit. Their framing: "Let's add a step that skims the script before the interpreting begins."
- Park, a data engineer, profiled the 6-hour stretch: actual computation took 40 minutes, and the other 5 hours 20 minutes was spent waiting on queries that made 30,000 round trips to the DB. The language wasn’t slow; there were simply too many round trips.
- After the team replaced per-record queries with batch queries and handed only the heavy aggregation to SQL and a numerical computing library, total batch time dropped from 6 hours to 1 hour 50 minutes. The language was still Python.
- That is also when the team found that the three production servers were running a mix of Python 3.9 and 3.11. They pinned the version with a container image, and over the next six months batch failures fell from 7 to 1, while the settlement close moved up from 8 a.m. to 4 a.m.