Tarik Al Hadethi Download CV

Writing

Where the bugs were hiding.

Each of these cost me real time, and each one turned out to be something other than what it looked like. The measurement that settled it is in every post.

  1. Nine emails to a live product

    Twenty-three days from a cold email to a live booking system for the UOWD Media Studio. The questions, the conflicting brief, the scope agreed in writing, and the bug a rehearsal on a clean server found when every test was green.

  2. 170 Supabase apps were leaking their whole database. Here is how to check yours.

    A five-minute check that your Supabase app is not exposing its database through the public key, and the five ways it goes wrong.

  3. Three green tests that tested nothing

    Fixing a word list made three of my own regression tests unable to fail. They stayed green. I could have deleted the code they guarded and nothing would have told me.

  4. A 41% error rate I cannot price

    A kernel picker chooses badly on twelve of twenty-nine real shapes, wasting sixteen of twenty-four rows in the worst case. That headline is unusable, because a rate counts occurrences and nothing yet says which of those shapes are hot.

  5. Frozen everywhere except the game

    My app stopped answering and, separately, my mouse stopped working in Chromium apps but not in Overwatch. One bug: a drag-and-drop that never ended, holding the whole desktop. Two earlier fixes explained the symptom perfectly and were both wrong.

  6. The benchmark I nearly published

    My first result said one inference library was 21 times slower than another. It was reproducible to within a few percent, and it was measuring the wrong thing. The check that caught it took thirty seconds and I nearly skipped it.

  7. The workaround I never questioned

    A note in my own repo said to strip a field out of the model file. It worked, so I never asked why. Months later the real answer turned out to be a bug in someone else’s library: its parser reads 1/2 as zero.

  8. Chromium doesn’t fail when it distrusts your GPU

    It switches WebGL to software rendering and still reports a working context, so the try/catch around the renderer catches nothing. Measured at 5.1 fps, and then my first fix broke a top-end GPU because its owner was playing a game.

  9. Windows unhooks you silently

    A low-level keyboard hook that runs late is removed by the OS with no error, no callback and no way to ask whether yours is still alive. One log line turned an unfalsifiable bug report into a measurement: five of eight keypresses never arrived.

  10. A race my tests could not reproduce

    Two people accepting the last seat at the same instant should collide. Every test I wrote passed whether the lock was there or not, twice for different reasons. Proving it took real concurrency, and the attempt found a security hole beside it.

  11. Verified on the wrong engine

    Six rounds fixing the same layout bug, because every measurement said it was already fixed. The measurements were honest. They were all taken on the one browser that implements the property correctly.

  12. A comment that was wrong in both halves

    One line said a setting was chosen for speed at a small cost in size. It was 21 times larger, and the option I replaced it with was smaller and faster. Nobody had measured it since the day it was written.