LuciferYang opened a new pull request, #9620:
URL: https://github.com/apache/paimon/pull/9620
### Purpose
close #9619
`StringToMapCastRule.splitMapEntries` used the escape marker to stop a
separator from splitting an entry, then threw away the character it was
protecting:
```java
if (escaped) {
escaped = false;
continue; // the escaped character
} else if (c == '\\') {
escaped = true;
continue; // and its backslash
}
```
So `{a\,b -> v}` produced the key `ab` rather than `a,b`. A MAP column
default value written that way is silently stored as something other than what
the DDL said, with no error anywhere.
Appending the escaped character is all this needs. It also lines the
backslash up with the quote handling a few lines below it, which already keeps
the content and drops only the marker, so `{"a,b" -> v}` has always produced
`a,b`. After this both escape mechanisms in this rule behave the same way.
Two things I deliberately left out. `ENTRY_PATTERN` splits key from value
after the entry has been unescaped, so `\->` inside a key still cannot be
protected; making that work means moving the key/value split into the
escape-aware scan, which is a larger change than this bug needs. And
`StringToArrayCastRule` and `StringToRowCastRule` run the same loop but append
every character, so backslashes and quotes survive into their elements while
the map rule strips both. Bringing the three into agreement is worth doing, but
it changes parsed output for array and row, so it does not belong in a bug fix.
### Tests
`CastExecutorTest.testSplitMapEntriesWithEscapes` covers the three cases the
loop distinguishes: an escaped separator stays inside the entry as a literal
comma, `\\` collapses to one backslash, and an escaped quote is a literal that
does not toggle quote state (so a comma inside the quoted section still does
not split).
`CastExecutorTest.testStringToMapPreservesEscapedCharacters` runs the whole
STRING to MAP cast and asserts the parsed key and value, which is the level a
default value goes through.
Both fail against the unfixed rule, with the escaped characters missing from
the expected strings.
`mvn -pl paimon-common -Dtest=CastExecutorTest test` on JDK 8: 40 tests, 0
failures. `spotless:check` and `checkstyle:check` on paimon-common are clean.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
