1 files changed, 99 insertions, 0 deletions
diff --git a/src/csv.rs b/src/csv.rs
new file mode 100644
index 0000000..e434989
--- /dev/null
+++ b/src/csv.rs
@@ -0,0 +1,99 @@
+//! A small RFC 4180 reader for the `/csv` batch endpoint.
+//!
+//! Values may contain commas and are then double-quoted with `""` escaping,
+//! which is why the batch cannot be split on `,`. The reader is deliberately
+//! lenient: blank lines are skipped and a stray quote inside an unquoted field
+//! is kept as data, because a malformed byte in one row must not cost the whole
+//! batch.
+
+/// Split CSV text into records, honouring quoting, `""` escapes, embedded
+/// commas and newlines, CRLF and LF line endings, and a leading BOM.
+pub(crate) fn parse(input: &str) -> Vec<Vec<String>> {
+    let input = input.strip_prefix('\u{feff}').unwrap_or(input);
+
+    let mut records: Vec<Vec<String>> = Vec::new();
+    let mut record: Vec<String> = Vec::new();
+    let mut field = String::new();
+    let mut quoted = false;
+    let mut chars = input.chars().peekable();
+
+    while let Some(c) = chars.next() {
+        if quoted {
+            if c == '"' {
+                if chars.peek() == Some(&'"') {
+                    chars.next();
+                    field.push('"');
+                } else {
+                    quoted = false;
+                }
+            } else {
+                field.push(c);
+            }
+            continue;
+        }
+
+        match c {
+            '"' if field.is_empty() => quoted = true,
+            ',' => record.push(std::mem::take(&mut field)),
+            '\n' | '\r' => {
+                if c == '\r' && chars.peek() == Some(&'\n') {
+                    chars.next();
+                }
+                record.push(std::mem::take(&mut field));
+                records.push(std::mem::take(&mut record));
+            }
+            _ => field.push(c),
+        }
+    }
+
+    if !field.is_empty() || !record.is_empty() {
+        record.push(field);
+        records.push(record);
+    }
+
+    records.retain(|r| r.iter().any(|f| !f.is_empty()));
+    records
+}
+
+#[cfg(test)]
+mod tests {
+    use super::parse;
+
+    #[test]
+    fn splits_plain_rows() {
+        let records = parse("a,b,c\n1,2,3\n");
+        assert_eq!(records, vec![vec!["a", "b", "c"], vec!["1", "2", "3"]]);
+    }
+
+    #[test]
+    fn keeps_empty_fields() {
+        let records = parse("99999999,,,,,,,,,,,,\n");
+        assert_eq!(records.len(), 1);
+        assert_eq!(records[0].len(), 13);
+    }
+
+    #[test]
+    fn reads_quoted_field_with_comma() {
+        let records = parse("cep,logradouro\n01001000,\"Rua Um, Dois\"\n");
+        assert_eq!(records[1][1], "Rua Um, Dois");
+    }
+
+    #[test]
+    fn unescapes_doubled_quotes() {
+        let records = parse("a\n\"say \"\"hi\"\"\"\n");
+        assert_eq!(records[1][0], "say \"hi\"");
+    }
+
+    #[test]
+    fn accepts_embedded_newline_and_crlf() {
+        let records = parse("a,b\r\n\"line\nbreak\",2\r\n");
+        assert_eq!(records[1][0], "line\nbreak");
+        assert_eq!(records[1][1], "2");
+    }
+
+    #[test]
+    fn ignores_bom_and_blank_lines() {
+        let records = parse("\u{feff}a,b\n\n1,2");
+        assert_eq!(records, vec![vec!["a", "b"], vec!["1", "2"]]);
+    }
+}