Invisible Character Finder

The two bugs this exists for

The first is a diff that shows every line as changed when nothing has changed. One side has Windows line endings and the other has Unix ones, and the difference is a byte per line that no editor renders. No diff tool tells you unless you already suspected it.

The second is two strings that look identical while an equality check returns false. One of them contains a zero-width space, a non-breaking space or a soft hyphen - almost always because it was copied out of a rendered web page, a PDF or a spreadsheet. The comparison is correct and the eye is wrong, which is an unusually frustrating place to be.

Both are five-minute bugs once you know and multi-hour bugs until you do. Closing that gap is the whole point of the tool.

Located, not just counted

Reporting that a string contains one zero-width space leaves you to find it. Every finding here carries a line, a column and a count, because locating it is most of the work - and each one is labelled with what it actually does to your program rather than just its code point.

• Zero-width spaces, joiners and byte-order marks, which break equality and trimming
• Non-breaking and typographic spaces, which look exactly like an ordinary space
• Bidirectional controls, including the overrides used in Trojan Source attacks
• Line-ending style, with the counts when a file is mixed
• Trailing whitespace and tabs, reported by line

Severity is real, not decorative

A zero-width joiner inside an emoji sequence is expected and harmless. The same character in an identifier is a defect. A right-to-left override in source code you did not write is a finding worth escalating, because it can make code display in an order different from the one the compiler reads - which is the basis of the Trojan Source class of attacks.

So findings are ranked rather than listed flat, and each one says what it means in context.

What it deliberately does not do

The cleaned output is offered, never applied. Normalising line endings, stripping zero-width characters and trimming trailing whitespace is what most people want, and rewriting a file on their behalf is not this tool’s job - identifying what is wrong with it is.

Tabs are reported and left alone, because a tab is a real indentation style rather than a defect, and silently converting them would impose a convention nobody chose.

It also cannot tell you whether a character was intended. A zero-width non-joiner is meaningful in Persian and Hindi text and almost never meaningful in a variable name; the tool flags it and you decide.