From c5666c6e092b1a9d30d205e7cc168b060276123f Mon Sep 17 00:00:00 2001
From: pdet <pedroholanda@gmail.com>
Date: Thu, 19 Dec 2024 14:17:33 +0100
Subject: [PATCH] Fix for issue #14648

---
 data/csv/comments/14648.csv                   |  3 ++
 .../csv_scanner/sniffer/csv_sniffer.cpp       |  2 +-
 .../csv_scanner/sniffer/dialect_detection.cpp | 15 ++++----
 test/sql/copy/csv/test_comment_midline.test   | 36 ++++++++++++++++++-
 4 files changed, 48 insertions(+), 8 deletions(-)
 create mode 100644 data/csv/comments/14648.csv

diff --git a/data/csv/comments/14648.csv b/data/csv/comments/14648.csv
new file mode 100644
index 000000000000..2e8df271055e
--- /dev/null
+++ b/data/csv/comments/14648.csv
@@ -0,0 +1,3 @@
+a,b
+a#b
+x,y
\ No newline at end of file
diff --git a/src/execution/operator/csv_scanner/sniffer/csv_sniffer.cpp b/src/execution/operator/csv_scanner/sniffer/csv_sniffer.cpp
index 74928a5c730c..7d6e2e3ed4bc 100644
--- a/src/execution/operator/csv_scanner/sniffer/csv_sniffer.cpp
+++ b/src/execution/operator/csv_scanner/sniffer/csv_sniffer.cpp
@@ -6,7 +6,7 @@ namespace duckdb {
 CSVSniffer::CSVSniffer(CSVReaderOptions &options_p, shared_ptr<CSVBufferManager> buffer_manager_p,
                        CSVStateMachineCache &state_machine_cache_p, bool default_null_to_varchar_p)
     : state_machine_cache(state_machine_cache_p), options(options_p), buffer_manager(std::move(buffer_manager_p)),
-      default_null_to_varchar(default_null_to_varchar_p) {
+      lines_sniffed(0), default_null_to_varchar(default_null_to_varchar_p) {
 	// Initialize Format Candidates
 	for (const auto &format_template : format_template_candidates) {
 		auto &logical_type = format_template.first;
diff --git a/src/execution/operator/csv_scanner/sniffer/dialect_detection.cpp b/src/execution/operator/csv_scanner/sniffer/dialect_detection.cpp
index ae06d024bad8..672859f8e21a 100644
--- a/src/execution/operator/csv_scanner/sniffer/dialect_detection.cpp
+++ b/src/execution/operator/csv_scanner/sniffer/dialect_detection.cpp
@@ -80,11 +80,11 @@ string DialectCandidates::Print() {
 
 DialectCandidates::DialectCandidates(const CSVStateMachineOptions &options) {
 	// assert that quotes escapes and rules have equal size
-	auto default_quote = GetDefaultQuote();
-	auto default_escape = GetDefaultEscape();
-	auto default_quote_rule = GetDefaultQuoteRule();
-	auto default_delimiter = GetDefaultDelimiter();
-	auto default_comment = GetDefaultComment();
+	const auto default_quote = GetDefaultQuote();
+	const auto default_escape = GetDefaultEscape();
+	const auto default_quote_rule = GetDefaultQuoteRule();
+	const auto default_delimiter = GetDefaultDelimiter();
+	const auto default_comment = GetDefaultComment();
 
 	D_ASSERT(default_quote.size() == default_quote_rule.size() && default_quote_rule.size() == default_escape.size());
 	// fill the escapes
@@ -187,6 +187,9 @@ void CSVSniffer::GenerateStateMachineSearchSpace(vector<unique_ptr<ColumnCountSc
 
 // Returns true if a comment is acceptable
 bool AreCommentsAcceptable(const ColumnCountResult &result, idx_t num_cols, bool comment_set_by_user) {
+	if (comment_set_by_user) {
+		return true;
+	}
 	// For a comment to be acceptable, we want 3/5th's the majority of unmatched in the columns
 	constexpr double min_majority = 0.6;
 	// detected comments, are all lines that started with a comment character.
@@ -208,7 +211,7 @@ bool AreCommentsAcceptable(const ColumnCountResult &result, idx_t num_cols, bool
 		}
 	}
 	// If we do not encounter at least one full line comment, we do not consider this comment option.
-	if (valid_comments == 0 || (!has_full_line_comment && !comment_set_by_user)) {
+	if (valid_comments == 0 || !has_full_line_comment) {
 		// this is only valid if our comment character is \0
 		if (result.state_machine.state_machine_options.comment.GetValue() == '\0') {
 			return true;
diff --git a/test/sql/copy/csv/test_comment_midline.test b/test/sql/copy/csv/test_comment_midline.test
index f0d3e5925c70..e9fab1d1aa2a 100644
--- a/test/sql/copy/csv/test_comment_midline.test
+++ b/test/sql/copy/csv/test_comment_midline.test
@@ -5,6 +5,39 @@
 statement ok
 PRAGMA enable_verification
 
+query II
+from read_csv('data/csv/comments/14648.csv',
+   header=false,
+   sep=',',
+   null_padding=true,
+   comment='#',
+   ignore_errors=true
+   );
+----
+a	b
+a	NULL
+x	y
+
+query II
+from read_csv('data/csv/comments/14648.csv',
+   header=false,
+   sep=',',
+   comment='#',
+   ignore_errors=true
+   );
+----
+a	b
+x	y
+
+statement error
+from read_csv('data/csv/comments/14648.csv',
+   header=false,
+   sep=',',
+   comment='#'
+   );
+----
+Error when sniffing file "data/csv/comments/14648.csv".
+
 # If we only have midline comments, the comment must be explicitly given by the user
 query I
 FROM 'data/csv/comments/only_midline.csv';
@@ -95,4 +128,5 @@ FROM read_csv('data/csv/comments/simple_mid_line.csv', buffer_size = ${buffer_si
 1	3
 6	7
 
-endloop
\ No newline at end of file
+endloop
+