> Sent: Thursday, October 29, 2015 at 1:34 PM > From: "Scott Hess" <shess at google.com> > To: "SQLite mailing list" <sqlite-users at mailinglists.sqlite.org> > Subject: Re: [sqlite] How would sqlite read this from disk? > > On Thu, Oct 29, 2015 at 10:20 AM, Jason H <jhihn at gmx.com> wrote: > > > > If I could ask a followup question. You made the statement "SQLite reads > > that row of the table from storage, from the first column to the last > > column needed by the SELECT, but perhaps not all the way to the end of the > > columns in the row.", Given an example select that requires columns 2,3,5, > > does this mean that column 4 is also read, or is column 4 be skipped? I > > guess the question is I am assuming that a row is serialization of one or > > more strings that are a serialization of a string preceded by their lengths > > (i.e. [L|data] ), would SQLite do something akin to (using my 2,3,5 > > example): > > begin row read, read L1, seek L1, read L2, read L2*byte, read L3, read > > L3*byte, read L4, seek L4, read L5, read L5*byte > > or would the 'read L4, seek L4' be changed to 'read L4, read L4*byte' ? > > > You should consider reading https://www.sqlite.org/fileformat2.html , > especially sections 2.3 "Representation Of SQL Tables" and 1.5 "B-tree > Pages". If your project _really_ needs to know this level of detail, then > you really should read up on the underlying system. Also maybe throw in > https://www.sqlite.org/arch.html to get a broad feel of how things break > down.
Thanks Scott. I had actually already linked to the fileformat2 url, but I did not see a discussion to the level of detail for row reading, for which I am after. I did however re-read in the context of your previous statement and gained some insight. I should a have stated it more like this: begin row read header_varint, serial_types*N, seek, read, seek, read or begin row read header_varint, serial_types*N, seek, read, read, read The documentation does not go into the detail of the engine is able to skip the reading of unneeded interior rows. In theory, it can because the length is contained in the header. So instead of read() on every column in the row, it can call seek() if it knows it doesn't need that column. My question is now simply: does it seek past unneeded columns, or does everything get send through read once data has started being read? It's a minor detail with bug performance implications. Many thanks again.
