[anilkir]

Problem

In StorageConfig::Header mode, DataContainer never records subscriptions because the header is marked complete on the first AddLoggedMessage. As a result:

• isValidDataMessage() returns false for every data message so valid data near a corrupted region is skipped until a different message type comes in
• data() skips the payload size check that FullLog mode runs
• Invalid AddLoggedMessages inside a corrupted region are ignored instead of raising an error, so the reader keeps reading out of sync

Change

In Header mode, keep a msg_id -> MessageFormat map. No Subscription is created and no samples are stored so memory use stays the same as before. With the map:

• addLoggedMessage() throws on a duplicate msg_id or a missing format (same as FullLog)
• data() throws on an unknown msg_id or a payload size outside [minWireSizeBytes(), sizeBytes()]
• isValidDataMessage() uses the map so recovery can resync on data messages

Testing

• Added two test cases:
  • Header mode recovers on the next valid data message after injected corruption
  • Header mode rejects a wrong-size data message for a known msg_id
• Both fail without the fix. All 8 tests pass.
• On two real flight logs with corrupted regions, Header mode now delivers exactly the same data messages as FullLog mode. Before, one log had about 540 fewer messages in Header mode.

Behaviour change

Header-mode users now get a ParsingException (reported through error()) for invalid data or subscriptions, the same as FullLog mode, instead of having them silently ignored.

[bkueng]

Can you check CI as well? For clang-tidy you can add the pragma once to the ignore list.