CSV to JSON Converter

CSV parsing edge cases that trip up developers

CSV looks trivially simple — split by commas, split by newlines, done. But that mental model breaks down quickly on real-world data. Here are the edge cases that cause silent bugs in naive parsers.

Embedded commas in quoted fields

RFC 4180 allows a field to contain commas if it's wrapped in double-quote characters. This is extremely common — addresses, descriptions, and any freeform text field will often contain commas:

name,address,city
Alice,"123 Main St, Apt 4",Vancouver
Bob,"456 Oak Ave, Suite 200",Toronto

Quoted newlines

A quoted field can also contain literal newline characters. This means a single logical CSV row can span multiple physical lines — a split-on-newline parser will break it into two rows incorrectly:

name,bio
Alice,"Software engineer.
Loves open source."
Bob,"Designer and illustrator."

This converter handles quoted newlines correctly. Note: the live tool above processes single-page input — for files with quoted newlines, the parser correctly tracks quote state across line boundaries.

BOM characters

Windows Excel adds a UTF-8 BOM (EF BB BF in hex, or  in Unicode) to CSV files it saves in UTF-8 mode. This invisible character prepends itself to the first column name. Without stripping it, a lookup on row['name'] silently returns undefined because the actual key is 'name'. This converter strips it automatically.

Type inference: should "30" be a string or a number?

This is one of the most debated questions in CSV-to-JSON conversion, and the answer is: it depends on what you're doing with the data. This tool takes a conservative approach — all values come out as strings. Here's why.

Automatic type coercion is lossy. The string "007" is a valid product code that must stay a string — coercing to a number drops the leading zeros. The string "true" might be a status field value, not a boolean flag. The string "1.0" might be a version string, not a float.

If you need typed output, cast fields after parsing. In JavaScript: Number(row.age), row.active === 'true'. In Python with pandas, use pd.read_json() with an explicit dtype map, or call df.convert_dtypes() after reading. Explicit is safer than implicit here.

Using CSV-to-JSON in ETL pipelines

In production ETL pipelines, CSV-to-JSON conversion is usually a transformation step between a raw data source and a downstream system. A typical flow: S3 bucket receives CSV exports from an operational database nightly → AWS Lambda or a Glue job converts them to JSON → JSON gets written to a DynamoDB table or Elasticsearch index.

The most important production concern is schema drift. If the upstream CSV gains or loses columns, your pipeline needs to handle it gracefully. Rather than hardcoding column names, derive them from the CSV header at runtime, validate against an expected schema, and raise an alert if required columns are missing rather than silently producing malformed JSON objects.

A second concern is character encoding. CSV files from legacy enterprise systems are often encoded in Windows-1252 or Latin-1, not UTF-8. If your pipeline reads files as UTF-8 without checking, you'll get mojibake on any accented characters or special symbols. Always detect or explicitly specify encoding at the file-read stage.

Streaming vs batch CSV conversion

For small files (under a few thousand rows), batch conversion — read the whole file, parse it, output JSON — is fine. This tool uses the batch approach. For large files (millions of rows), streaming is necessary to avoid loading the entire file into memory.

Streaming CSV parsers process the file row by row. Node.js libraries like fast-csv and csv-parse support streaming via Node.js Readable streams. Python's built-in csv.reader() is already an iterator — it doesn't load the whole file. For multi-gigabyte files, streaming is non-negotiable; a 2GB CSV loaded into memory as JSON objects can easily hit 6–8GB RAM before the garbage collector cleans up.

Frequently Asked Questions