SADevToolBox

Guides

Longer write-ups on the problems behind the tools: the ones that are easy to get wrong, hard to notice and rarely explained in one place. Each guide is written from working with the format, not summarised from its specification.

Base64 adds a third to every file: when a data URI still wins Inlining an image as a data URI removes a request and adds a third to the bytes. Whether that trade is worth it depends on four things you can actually measure. September 2026 Characters, code points and bytes: why counts disagree A string's length depends entirely on what you are counting. Getting that wrong is how a 280-character limit rejects a 200-character message. September 2026 Five CSV edge cases that break data imports CSV has a specification, and almost nothing follows it exactly. These are the five deviations that corrupt real imports, with the fix for each. September 2026 Why JSON loses large IDs, and how to keep every digit A 19-digit ID can come back from JSON.parse with the last two digits changed, and nothing reports an error. Here is why, and what to do about it. September 2026 What a decoded JWT proves, and what it does not A decoded token looks authoritative. It is not: anyone can write those claims. Here is what still has to be checked, and the attacks that skipping it enables. September 2026 Writing bulk INSERT scripts that load quickly A million single-row INSERTs and a million rows in batched INSERTs are the same data and wildly different load times. Here is where the difference comes from. September 2026 Unix timestamps, UTC and the bugs in between Most date bugs are not arithmetic errors. They come from storing the wrong thing: an offset instead of a zone, or a local time instead of an instant. September 2026 UUID v4, UUID v7 and SQL Server GUIDs as primary keys Random identifiers are excellent keys and terrible clustered index keys. UUID v7 changes that trade-off, and this is how it plays out per database. September 2026

Why these exist

Every tool page on this site answers the questions that come up while using that one tool. Some questions are bigger than that: they cut across several tools, they have a real answer that takes more than a paragraph, and getting them wrong causes bugs that do not announce themselves. Those are the ones that end up here.

Each guide covers one problem, states what actually happens rather than what the specification permits, and gives the fix. Where a claim can be checked, the guide says how to check it, so you do not have to take it on trust.