nielspardon opened a new issue, #12843:
URL: https://github.com/apache/gluten/issues/12843
### Description
`ReadFromGlutenStorageKafka` stores its `column_names` as a **reference**
member:
```cpp
// cpp-ch/local-engine/Storages/Kafka/ReadFromGlutenStorageKafka.h:52
const Names & column_names;
```
initialized from the constructor's `const Names & column_names_` parameter
(`ReadFromGlutenStorageKafka.cpp:53`). The sole construction site binds it to a
**local** that goes out of scope before the step runs:
```cpp
// cpp-ch/local-engine/Parser/RelParsers/StreamKafkaRelParser.cpp:102-107
Names names = header->getNames();
auto source = std::make_unique<ReadFromGlutenStorageKafka>(
names, header, getContext(), topics, partition, start_offset,
end_offset, poll_timeout_ms, group_id, brokers);
steps.emplace_back(source.get());
query_plan->addStep(std::move(source)); // the step outlives `names`
```
`names` is destroyed when `parse()` returns, but the step (and its
`column_names` reference) lives on in the query plan and runs later in
`initializePipeline`. The reference dangles.
### Impact
Latent today: `column_names` is never actually dereferenced — it is stored
in the constructor but read nowhere
(`makePipe`/`initializePipeline`/`createKafkaSettings` don't touch it; the
Kafka source derives its schema from `output_header` instead). So it is a
harmless-but-real dangling reference and a footgun: any future read of
`column_names` would touch freed memory.
### Fix
Either make it an owning value member — `Names column_names;` (drop the `&`)
so it copies at construction — or remove the unused member entirely (and drop
the now-unused `column_names_` constructor parameter). Given it has no readers,
removing it is the cleaner option.
### Notes
Pre-existing; not introduced by the Substrait-0.98 rebase (#12597) —
surfaced while reviewing this file for the Kafka `ExtensionTable` remodel
(#12841). By contrast the sibling `topics` member is stored by value (`Names
topics;`), so only `column_names` is affected.
--
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]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
