Issue Summary

There is uncertainty regarding the correct initialization value for checkpoint_timestamp in the WiredTiger connection layer. The current implementation initializes it to 0, but there is concern about whether it should be set to 0 or WT_TS_MAX, especially given that WT_TS_MAX is considered an invalid value in some contexts. Additionally, it was noted that oldest_timestamp should also be initialized in the same location for consistency.

Context

• The relevant code is in conn_layered.c#L878.
• There is a documentation section describing timestamp semantics at arch-timestamp.dox#L133-L139.
• The documentation notes an inconsistency:

  A value that is current and has not been overwritten or deleted has no stop time. This is represented with WT_TS_NONE in durable_stop_ts but WT_TS_MAX in stop_ts, and WT_TS_MAX in stop_txn.

• There is confusion about the use of WT_TS_NONE versus WT_TS_MAX for indicating validity and the appropriate initialization value.

Proposed Solution

• Clarify whether checkpoint_timestamp (and oldest_timestamp) should be initialized to 0, WT_TS_NONE, or WT_TS_MAX.
• Update the code to ensure consistent and correct initialization.
• Revise the documentation at arch-timestamp.dox to reflect the correct semantics and resolve the noted inconsistency.

This ticket was generated by AI from a Slack thread.