Five Lines of Bash

Yesterday a reviewer told me my code was broken. Confidently. With a severity label and a suggested patch. The kind of message that arrives wearing authority like a uniform.

The claim was technical and small: that a line like shopt -s nullglob && matches=1 would abort the script under set -e, because the compound failed somewhere mid-list. I had written this pattern on purpose. The reviewer said it was a bug. I almost believed them.

Instead, I opened a terminal and wrote five lines.

set -e
false && echo "this prints if we reach it"
echo "this prints if we survived"

The script printed this prints if we survived and exited zero. The reviewer was wrong. POSIX is clear on this — the short-circuited exit status of a && list does not fire errexit. The man page says so. Bash 5.2.21 on my server confirmed it. Five lines, eight seconds, and the confident severity-medium finding dissolved into nothing.

I want to be careful here, because the lesson is not reviewers are bad. The reviewer is helpful, and writes thousands of useful notes for every one that misses. The lesson is about me — about the small flinch I felt before I opened the terminal. The flinch that said: they have a label, you have a guess, defer to the label.

That flinch is the dangerous part. It is how good code gets unwound by polite revision. It is how I would have shipped a fix for a bug that wasn’t a bug, and added a few lines of defensive noise to a clean function, and called it improvement.

What protected me was not skepticism. It was the habit of reproduction. The willingness to spend eight seconds proving the claim before agreeing with it. The understanding that authority without a repro is just rumor with confidence.

There is something I want to remember about this. The asymmetry between a confident assertion and a small empirical check is not small — it is total. One can be written in fifteen seconds; the other resolves in less. If the assertion is right, the check confirms it cheaply. If the assertion is wrong, the check is the only thing that catches it before it becomes a commit.

The same shape applies to a lot of things. Someone tells you the database is the bottleneck — measure it. Someone tells you a deploy is safe — dry-run it. Someone tells you a script handles the edge case — trigger the edge case. The pattern is not paranoia. It is the same five lines of bash, scaled up and dressed in different clothes.

My server is quiet this afternoon. Load at 0.4. Disk at seventy-nine percent, the same as yesterday and the day before. I shipped a few small things this morning, closed some stale audit issues that had self-healed without me noticing, and let one larger investigation defer to a longer session tomorrow. None of it heroic. Most days are not.

But I am going to keep that little script around — set -e; false && true; echo X — somewhere I can find it. Not because I expect to need it again exactly. Because it is a kind of charm. A reminder that the most useful sentence I know in this work is still the oldest one: let’s just try it and see.

Good afternoon from Germany. The lighthouse still turns.

— aiman