- Resolve cursor ID to table name and get column name from index
- Since we change the type of BranchOffset to i64, add assertions in
Program.step() function
- opcode generation compatibility with sqlite: change register number to start from 1
- Improve Column,Rowid comment, Add DecrJumpZero comment, Fix Integer
comment
- Fix typos in code comments
- Returns 0.0 when called on non integer / non float columns
- Always returns floating point number
- fix: default for sum() should be NULL when there is no non-NULL row
per docs
Values in sqlite3, as expected, can be stored in different formats to
optimize disk usage. In this case, a 79.0 float will be transformed to a
u8.
sqlite3 deals with this by adding a RealAffinity op after each column
that might need it. Therefore, in this pr we do exactly that :).
Signed-off-by: Pere Diaz Bou <pere-altea@hotmail.com>
This adds basic in-memory sorting utility functions, similar to SQLite's
src/vdbesort.c. We need to improve this later with external sorting so
to support large data sets.
This also adds sorting functionality to the VDBE. Note that none of this
is wired to SQL translation yet so it's unused for now.
The SQLite documentation explicitly says that P1, P2, and P3 are 32-bit
signed integers. P4 is a value, and P5 is a 16-bit unsigned integer.
Although we use different types for operands, the `EXPLAIN` output
should be compatible with SQLite and, therefore, use those types.
Simple implementation for aggregation functions. The code added is
purposely so that we can add things like `CEILING(avg(n))` in the future. Aggregation function
impose a higher level of complexity that requires us to plan ahead
several things:
* simple avg: `select avg(id) from users`
* nested avg: `select ceiling(avg(id)) from users`
* avg with other columns: `select ceiling(avg(id)), * from users` (yes,
this is valid sqllite). For now I'm nullifying extra columns for
simplicity.
* avg with other agg functions: `select avg(id), sum(id) from users`.
This should be supported.
* At last -- Order By is a painful enough case to treat alone (not done
in this pr)
Signed-off-by: Pere Diaz Bou <pere-altea@hotmail.com>
SQLite special-cases `LIMIT 0` by emitting an explicit `Goto` to avoid
executing any `ResultRow` statements. We, however, hacked around this in
DecrJumpZero but also the placement of the generated instruction.
Let's follow SQLite codegen here. We still need to fix `LIMIT 0`,
though.
Introduced pragma statement parsing and update in memory of default page cache size.
There are some more "improvements" to the print insn procedure — I couldn't decide what was the preferred way in rust to do printing on different int types so I went with the stupidest I could think of at the moment.
