simon, >This is normally resolved by logging the time each command was issued
Not "when it was issued" but when exactly it was finally commited. When one or more update statements are part of a transcation, you now have to use sqlite3_profile to log when exactly the transaction ended (i.e. can be considered actually stored). That implies both sqlite3_trace and sqlite3_profile need to be monitored to keep track of what constitutes an update transaction and when precisely it was commited. If machines A and B (or more) run independantly, I don't believe you need a central "untouched" copy: you can very well carefully merge logs and determine for each machine what change took place after every of its own updates for every row in every table. Another tough issue is how to re-sync machine times down to a very small time window in a common time frame. Whatever small "vulnerable" time window you achieve, you can be assured that it will cause in a problem someday, resulting in updates merged out of real-world order. Note that merging can also violate constraints in the merged database. Say a rule (enforced by constraint or at applicative level) limits the number, volume or nature of the updates permitted in a given timeframe, you have no guarantee that this rule will be enforced if independant (albeit individually conformal) changes are made but only later merged. Several changes can very well have taken place independantly without breaking local rules (to the knowledge of the local machine), but discover only at merge time that some of the last operations would have been rejected if they have been submitted to a centralized machine/DB. For instance you have an export restriction (available stock or manufacturing capability) which allows no more than N given items XYZ to be exported (or sold) per day/week/month. Even if independant machines+DBs don't violate the rule from their own point of view, the merged dataset may be in violation of the limit(s) or simply rejected by SQL constraints at merge time. While I agree that this is not exactly in the realm of raw data merge in the right order, this still is a serious problem to deal with, not solvable with such simple merging operations.
