TopolearnUser uploads arrive as commas, semicolons or tabs, and you can't always ask which.
csv.Sniffer inspects a sample of text and guesses the format.
Sniffer().sniff(sample, delimiters=None)returns aDialectclass with the detecteddelimiter,quotechar,doublequoteandskipinitialspace. Restrict the candidates withdelimiters=",;\t|"for more reliable results. It raisescsv.Error("Could not determine delimiter") when it can't decide.Sniffer().has_header(sample)guesses whether the first row is a header, using heuristics such as "the first row's values differ in type or length from the rest".
Typical use:
sample = f.read(64 * 1024)
f.seek(0)
dialect = csv.Sniffer().sniff(sample, delimiters=",;\t")
reader = csv.reader(f, dialect)Read a sample of a few kilobytes that ends near a line boundary, then rewind with seek(0).
These are heuristics. They fail on single-column files, tiny samples, or data where every column is text (then has_header can't tell a header from data). Use them for user-facing imports with a fallback (default to excel) and, ideally, a preview the user can correct. For files you control, specify the dialect explicitly instead.
Example
import csv, io
files = {
"commas": "id,name,score\n1,Ada,9.5\n2,Bob,7.0\n",
"semis": "id;name;score\n1;Ada;9,5\n2;Bob;7,0\n",
"tabs": "1\tAda\t9.5\n2\tBob\t7.0\n",
}
sniffer = csv.Sniffer()
for label, text in files.items():
d = sniffer.sniff(text, delimiters=",;\t")
print(label, repr(d.delimiter), sniffer.has_header(text),
next(csv.reader(io.StringIO(text), d)))
# commas ',' True ['id', 'name', 'score']
# semis ';' True ['id', 'name', 'score']
# tabs '\t' False ['1', 'Ada', '9.5']
try:
sniffer.sniff("just one column\nno delimiters\n", delimiters=",;")
except csv.Error as e:
print("csv.Error:", e) # csv.Error: Could not determine delimiterRemember
- Sniffer().sniff(sample) returns a dialect with the detected delimiter
- has_header() guesses whether the first row is a header
- Restrict delimiters and keep a fallback; sniffing is heuristic
Watch out for
- Sniffing a sample and forgetting f.seek(0), so the reader starts mid-file.