Week 7 Session: What does it need from a filesystem?

Establish what your program actually requires of a filesystem, then break a filesystem that looks correct

12 April 2027

Two halves this week.

First, your target. Does it need durability, or only a working directory for the length of a session? Does it mmap? Does it rename over a live file? Does it create symlinks, or only follow them? Does anything else — a build tool, an editor, another runtime — need to see the same bytes at the same time?

Most programs need far less than their authors assume, and the useful output is a short list of requirements you can defend.

Second, a conformance test. Write a small program that exercises one filesystem behaviour precisely — directory enumeration while deleting is the recommended one — and run it in two places: a real POSIX system first, then a browser-hosted filesystem.

The native run is not a formality. It is the oracle. Without it you cannot tell a wrong answer from an answer you did not understand, and you will spend the hour debugging your own test.

Suggested behaviours, if you want a different one:

• enumerate a directory while unlinking from it
• rename over an open file, then read the old descriptor
• two writers appending to one file
• stat after a write, from a different runtime

Every one of those has a scale you can turn up: entries in the directory, bytes in the file, concurrent writers, calls before you check. Turn it up until the two hosts disagree, then turn it back down until they agree again.

That threshold is the finding. Not the bug — the number of entries, or kilobytes, or writers below which everything looks correct. It tells you how large a test would have had to be to catch this, and therefore why nobody did.