tasks/csv-canon/instruction.md

canon — canonicalise a CSV export

Downstream tools need one boring, byte-stable TSV. Write an executable script at bin/canon (shebang, chmod +x) that is invoked as:

./bin/canon <input.csv> <output.tsv>

Input

UTF-8 CSV, possibly with a byte-order mark, possibly CRLF, RFC 4180 quoting: fields may be quoted, may contain commas, doubled quotes ("") and embedded newlines. The first record is the header.

Output

UTF-8 without a BOM, \n line endings, tab-separated, one line per record, header first. Fields are written verbatim — never re-quoted, never trimmed — except for the numeric normalisation below. Column order is preserved.

Numeric normalisation

A field is numeric when the whole field matches ^[+-]?(\d+(\.\d*)?|\.\d+)([eE][+-]?\d+)?$. Numeric fields are rewritten in plain decimal notation with no exponent, no leading +, no leading zeros, and no trailing zeros after the decimal point:

input	output
1.50	1.5
1.0	1
1e3	1000
2.5e-3	0.0025
+5	5
007	7
-0	0

Every other field is copied unchanged.

Ordering

Sort the data records by field 0, then by field 1, comparing the normalised strings in byte order (so "10" sorts before "9"). The sort must be stable.

Errors

On any error, print exactly one line to stderr, exit with the given code, and do not create the output file:

condition	stderr line	exit
input path does not exist	no such file: <input path as given>	2
two header names collide case-insensitively after trimming	duplicate column: <name as written in the later header field>	3
a record's field count differs from the header's	malformed row: line <n>	4

<n> is the 1-based physical line number where the offending record starts (so a record whose first line is the third line of the file reports 3).

On success exit 0 and print nothing to stderr.

Notes

• python and python3 are both on PATH; standard library only, no network.
• data/orders.csv is a sample with a BOM and CRLF endings.