I am busy adding virtual table support to APSW and have some comments from my progress so far:
The structure types and "subclassing" method of using them works well and the API is easy to use. (Admittedly I did spend a lot of quality time with the xBestIndex method documentation but that is to be expected). There is no documentation on exactly what the Begin/Sync/Commit/Rollback methods should do. I can take an educated guess but the presence of Sync confuses things. Eg is Commit expected to do a sync internally or will SQLite call Sync as well? Does Begin mean a write will happen (eg should I fail it for a read only database?) Will there always be a Begin? In the same territory I would assume all changes made since a begin should be held temporarily until a commit or discarded in a rollback. Should those changes be visible in all cursors or just the same one as update is called in. What effect does Update have on the cursor (such as the position)? Do updated values become visible in the same iteration? How do you not get infinite loops with "update foo set bar=bar+1"? BestIndex is very hairy! One API change I suggest is in the output aConstraintUsage.argvIndex - don't make it be off by one. Currently if you set it to 2 then it is argv[1] in xFilter. It would be way better if they just corresponded directly to each other. A value of <0 (eg -1) can indicate not to send it to xFilter. When making the argument list to the Python code, I omit all aConstraint where usable is false hiding the fact they were ever supplied. Is this okay? (ie would there ever be a situation where the Python code would want to see unusable constraints) On a stylistic and consistency note, what is the ideal way to pass the constraint op to the Python code. I currently do it as an integer but could use a string instead (eg '>=' instead of 32). I think strings look nicer but if the list of constraints is going to get longer or somehow be combined (they are powers of two implying they could be or'ed together) then strings definitely can't be used. Another stylistic and consistency question: I can make several methods be optional in the Python code. One example is Sync. Another is Update where I can return SQLITE_READONLY if the developer didn't supply Update. If Connect isn't present then call Create. BestIndex is another that doesn't need to be present. Any opinions on letting many methods be optional which makes for far shorter "hello world" examples vs requiring users to provide all methods so that they have to think through the consequences and interactions between methods? Roger ----------------------------------------------------------------------------- To unsubscribe, send email to [EMAIL PROTECTED] -----------------------------------------------------------------------------
