JSON Formatter

Pretty-printing looks like a purely cosmetic operation, and mostly it is. A few of the decisions a formatter makes do carry further than appearance, particularly once the output lands in version control or gets handed to another system.

Format for the diff, not just the eye

The strongest argument for formatted JSON is reviewability. A minified document sits on one line, so changing a single value produces a diff that reports the entire file as modified and shows one enormous line replaced by another. Nobody can review that. Formatted with one value per line, the same change shows as a one-line diff.

Indentation width matters less than consistency. Two spaces is the most common convention and keeps deeply nested documents from running off the right margin. What causes real friction is a repository where different tools format the same file differently, because every save produces spurious changes. Agreeing one setting and applying it in a build step is worth more than the specific width chosen.

Key order is preserved, and sorting it is a real decision

The JSON specification treats objects as unordered, but essentially every parser preserves the order keys appeared in, and code sometimes depends on that without anyone intending it. A formatter that alphabetizes keys makes files far easier to compare and to spot missing entries in, at the cost of destroying whatever grouping the author intended, such as keeping an id next to its name rather than scattered alphabetically. Neither choice is wrong, but it should be a choice.

Formatting will break a JSON Lines file

This one catches people out regularly. JSON Lines, also seen as NDJSON, stores one complete JSON object per line and is common in logs, data exports, and streaming pipelines. The file as a whole is deliberately not a single valid JSON document. So running it through a formatter either fails outright or, worse, reformats the first line and discards the rest.

The tell is a file where every line starts with an opening brace and there are no commas between records. Those files should be processed line by line, and the line structure is load-bearing rather than incidental formatting.