Repository navigation
Description
Describe the bug
TypeParser::NextToken (clickhouse/types/type_parser.cpp:276-298) decodes escapes inside a single-quoted Enum label with kUnescapeMap (type_parser.cpp:87-99), which only covers \0 \b \f \n \r \t \\ plus a special case for \'. Every other escape falls into the else arm on line 293 and keeps its backslash.
The server (parseComplexEscapeSequence, src/IO/ReadHelpers.cpp) additionally:
- decodes
\Nto no characters at all; - drops the backslash when the escaped character is one of
\ ' " ` / =or an ASCII control character (<= 0x1F).
So the client and the server disagree on these labels:
| Escape in a quoted label | Server | clickhouse-cpp |
|---|---|---|
\N |
removed (empty) | \N (backslash kept) |
\= |
= |
\= |
\/ |
/ |
\/ |
\ + control char 0x00-0x1F (for example a literal TAB) |
the control character | \ + the control character |
The same applies to the backtick / double-quote branch (type_parser.cpp:299-327), which decodes only \<quote> and <quote><quote> and leaves every other escape untouched. The server applies the full escape table to back-quoted identifiers too, so Tuple(`a\Nb` Int8, c Int8) is Tuple(ab Int8, c Int8) on the server but gives the element name a\Nb in the client.
Impact is low, the same as in the C# client issue linked below: the server never emits these escapes in a type name that it sends, so the read path and server-driven insert schemas are not affected. It matters only for type strings that the application writes, for example CreateColumnByType(...) or TypeParser with a DDL type copied from a CREATE TABLE. There the client builds an Enum whose label is a\Nb while the server's label for the same DDL is ab, so ColumnEnum::Append("ab") throws, and labels can collide or fail to collide differently from the server (Enum8('a\Nb' = 1, 'ab' = 2) is a duplicate-name error on the server but two distinct labels in the client).
ClickHouse server version
26.10.3.7. Server behaviour verified over HTTP:
SELECT toTypeName(CAST(1 AS Enum8('a\Nb' = 1))) -- Enum8('ab' = 1)
SELECT toTypeName(CAST(1 AS Enum8('a\=b' = 1))) -- Enum8('a=b' = 1)
SELECT hex(toString(CAST(1 AS Enum8('a\<TAB>b' = 1)))) -- 610962 (a<TAB>b)
SELECT toTypeName(CAST((1,2) AS Tuple(`a\Nb` Int8, c Int8))) -- Tuple(ab Int8, c Int8)
The client side comes from analysis of the tokenizer. The test below compiles against the library, but my sandbox could not execute the built binary, so the "clickhouse-cpp" column is the behaviour that the code produces by inspection, not a captured run.
Reproduction
Add to ut/type_parser_ut.cpp, in the style of the existing TypeParserCase.ParseEnum_AllEscapedBytes:
TEST(TypeParserCase, ParseEnum_ServerEscapes) {
const char kTab = '\t';
// {fragment as seen by the parser, expected decoded label}
const std::vector<std::pair<std::string, std::string>> cases = {
{"a\\Nb", "ab"}, // \N decodes to nothing
{"a\\=b", "a=b"}, // \= decodes to '='
{"a\\/b", "a/b"}, // \/ decodes to '/'
{std::string("a\\") + kTab + "b", // backslash + literal TAB
std::string("a") + kTab + "b"},
};
for (const auto& [label, expected] : cases) {
TypeAst ast;
ASSERT_TRUE(TypeParser("Enum8('" + label + "' = 1)").Parse(&ast));
ASSERT_EQ(ast.meta, TypeAst::Enum);
EXPECT_EQ(ast.elements[0].value_string, expected) << "label: " << label;
}
// Back-quoted element names use the same server escape table.
TypeAst tuple;
ASSERT_TRUE(TypeParser("Tuple(`a\\Nb` Int8, c Int8)").Parse(&tuple));
EXPECT_EQ(tuple.elements[0].name, "ab");
// An unknown escape keeps its backslash, like the server.
TypeAst unknown;
ASSERT_TRUE(TypeParser("Enum8('a\\zb' = 1)").Parse(&unknown));
EXPECT_EQ(unknown.elements[0].value_string, "a\\zb");
}Expected: every EXPECT_EQ passes (labels ab, a=b, a/b, a<TAB>b, tuple element name ab).
Actual: the first five fail and give a\Nb, a\=b, a\/b, a\<TAB>b and tuple element name a\Nb. Only the last case (\z keeps its backslash) passes.
Public-API form of the same problem:
auto col = CreateColumnByType("Enum8('a\\Nb' = 1)"); // the server label for this DDL is "ab"
col->As<ColumnEnum8>()->Append(std::string("ab")); // throws: unknown enum nameSuggested fix
In clickhouse/types/type_parser.cpp:
- handle
\Nin the single-quote branch: consume both characters and append nothing; - drop the backslash also for
",`,/and=(kUnescapeMaphas only0 b f n r t \today, and'is special-cased at line 285); - drop the backslash when the escaped byte is an ASCII control character (
<= 0x1F); - apply the same table in the backtick / double-quote branch (lines 299-327), which today decodes only the quote character itself.
Keep the current behaviour for everything else: an unknown escape such as \z keeps its backslash, and \ + DEL (0x7F) keeps its backslash.
Link
Relayed from the equivalent C# client issue: ClickHouse/clickhouse-cs#730
Activity
eminfikret commentedon Nov 16, 2026
Hello, is this issue ready for contribution? I'd like to pick it up if it's available. Thanks.
polyglotAI-bot commentedon Nov 17, 2026
Recommendation: Keep open — fix needed
No open PR fixes this: PR #608 (for #605) changes the same branch, but only adds the
''form. The fix must also decode\xHH, which the server decodes and the suggested fix in the issue does not list.master: Reproduced on master —58453c2, ClickHouse 26.10.4.207, default 2.x build options; the program comparesCreateColumnByType()with the server'stoTypeName(defaultValueOfTypeName(...))for the same text.''form in the same function), Bug: escaped single quotes in enum/type names not parsed correctly #482 (the\'form, fixed by Escape special characters in Enum/DateTime type names #538, first released in v2.6.0).Analysis: The single-quote branch of
TypeParser::NextToken(clickhouse/types/type_parser.cpp:276-297) decodes only\'and the escapes inkUnescapeMap(type_parser.cpp:87-99:\0 \b \f \n \r \t \\); every other escape keeps its backslash (type_parser.cpp:293-295). The back-quote and double-quote branch (type_parser.cpp:299-327) decodes only\<quote>and the doubled quote, so a Tuple element name keepsa\Nb. The server decodes\Nto nothing, drops the backslash before=,/,",`and a control character, and also decodes\x41toA; an unknown escape such as\zkeeps its backslash on both sides. With the client labela\Nb,Append("ab")throwsstd::out_of_range(map::at), andEnum8('a\Nb' = 1, 'ab' = 2), which the server rejects as duplicate names, is accepted. Type names that the server sends do not contain these escapes, so only type names that the application writes are affected.Reproduction: complete program and output
Build (in the source tree, after
cmake -S . -B build -G Ninja -DCMAKE_BUILD_TYPE=Release -DWITH_OPENSSL=ON && cmake --build build):Output on
master(58453c2), ClickHouse 26.10.4.207 (the fourth type name has a literal TAB after the backslash):Automated triage by Polyglot AI. Classification added on 2026-11-19. A maintainer makes the final decision.