[PARQUET-2131] Number values decoded DCHECKs should be exceptions

• Type: Bug
• Status: Resolved
• Priority: Minor
• Resolution: Fixed
• Fix Version/s: cpp-8.0.0
• Component/s: parquet-cpp
• Labels: pull-request-available
• Assignee: William Butler
• Reporter: William Butler
• Created: 22/Feb/22 22:16
• Updated: 05/Mar/22 19:31
• Resolved: 04/Mar/22 23:48

Description

As discussed on some other bugs, there are some parquet-cpp DCHECKs on the number of values decoded that really should be exceptions. When invalid Parquet files are read, it is possible for the decoders to return less values than expected and this should be signaled back to the user even in non-debug mode and it should not be a crash in debug mode. A fuzzer I have been running managed to create two examples that crashed parquet-cpp due to the DCHECKs. These don't currently crash the non-arrow cpp interfaces for reading parquet but we might consider making those methods also throw exceptions in these situations.

Issue Links

• links to: GitHub Pull Request #78
• links to: GitHub Pull Request #12490