Xianqing He created IMPALA-15547:
------------------------------------
Summary: Fix LIKE pattern truncation at embedded NUL bytes
Key: IMPALA-15547
URL: https://issues.apache.org/jira/browse/IMPALA-15547
Project: IMPALA
Issue Type: Bug
Components: Backend
Affects Versions: Impala 5.0.0
Reporter: Xianqing He
h3. Problem
IMPALA-12374 optimized constant LIKE/ILIKE patterns with leading/trailing {{%}}
wildcards by trimming them and switching to anchored partial regex matching.
The pattern conversions introduced there used the null-terminated
{{StringVal(const char*)}} constructor, which computes the length with
{{strlen()}}. As a result, a constant pattern containing an embedded NUL byte
(valid in STRING values) is silently truncated at the first NUL before
{{ConvertLikePattern()}} compiles the regex, producing wrong results.
Example: {{LIKE 'a%<NUL>b'}} is truncated to {{a%}}, so the value {{'ax'}}
incorrectly matches.
h3. Fix
Use the length-aware {{StringVal(uint8_t*, int)}} constructor in
{{LikePredicate::OptimizeConstantPatternMatch()}} for both the raw and the
trimmed pattern, preserving every byte.
h3. Testing
Added regression tests in {{be/src/exprs/like-predicate-test.cc}} covering an
untrimmed pattern ({{a%<NUL>b}}) and a pattern trimmed on both sides
({{%a<NUL>_%}}), each with matching and non-matching values, in case-sensitive
and case-insensitive (ILIKE) modes. Verified the tests fail before the fix and
pass after; {{TestLikePredicate.RegexMatchOptimization}} still passes.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
