Fix codegen for binary functions and add fuzz test for math functions
(we need to compile `rusqlite` with `-DSQLITE_ENABLE_MATH_FUNCTIONS` in
order to bundle sqlite with math functions compiled)
Reviewed-by: Jussi Saurio (@jussisaurio)
Closes#1015
Align `substr` implementation with SQLite spec
(https://www.sqlite.org/lang_corefunc.html#substr):
> The substr(X,Y,Z) function returns a substring of input string X that
begins with the Y-th character and which is Z characters long. If Z is
omitted then substr(X,Y) returns all characters through the end of the
string X beginning with the Y-th. The left-most character of X is number
1. If Y is negative then the first character of the substring is found
by counting from the right rather than the left. If Z is negative then
the abs(Z) characters preceding the Y-th character are returned. If X is
a string then characters indices refer to actual UTF-8 characters. If X
is a BLOB then the indices refer to bytes.
Reviewed-by: Jussi Saurio (@jussisaurio)
Closes#1013
According to SQLite documentation, the way to use these instructions
is to compare the seek key to the index key as you would with the
Compare opcode. The compare opcode states:
"Compare two vectors of registers in reg(P1)..reg(P1+P3-1)
(call this vector "A") and in reg(P2)..reg(P2+P3-1) ("B")."
In other words, we should compare the same number of columns from each,
not compare the entire keys.
This fixes a few Clickbench queries returning incorrect results, and
so closes#1009
---
Future work: support index seek keys that use multiple columns. Our
index seek is many times slower than SQLite because we're not utilizing
all the possible columns -- instead we just use the first index column
to seek.
I added the two opcodes as an initial step. They are pretty easy to
implement since we already have the counterparts i.e., IdxGE and IdxGT
Is there a design reason behind their omission @penberg @PThorpe92?
I noticed the same for SeekLE and SeekLT.
Reviewed-by: Jussi Saurio <jussi.saurio@gmail.com>
Closes#1010
easy implementation, sqlite claims it is a noop now
"This pragma no longer functions. It has become a no-op. The
capabilities formerly provided by PRAGMA legacy_file_format are now
available using the SQLITE_DBCONFIG_LEGACY_FILE_FORMAT option to the
sqlite3_db_config() C-language interface."
Closes#1007
easy implementation, sqlite claims it is a noop now
"This pragma no longer functions. It has become a no-op. The capabilities
formerly provided by PRAGMA legacy_file_format are now available using
the SQLITE_DBCONFIG_LEGACY_FILE_FORMAT option to the sqlite3_db_config()
C-language interface."
After reading the fine print, SQLite documentation explains that `BEGIN
IMMEDIATE` and `BEGIN EXCLUSIVE` are the same thing in WAL mode:
https://www.sqlite.org/lang_transaction.html
As that's the only mode we support, let's just add code generation for
`BEGIN EXCLUSIVE`.
Fixes#1002Closes#1003
After reading the fine print, SQLite documentation explains that `BEGIN
IMMEDIATE` and `BEGIN EXCLUSIVE` are the same thing in WAL mode:
https://www.sqlite.org/lang_transaction.html
As that's the only mode we support, let's just add code generation for
`BEGIN EXCLUSIVE`.
Fixes#1002
Emit the following code sequence for `BEGIN IMMEDIATE`:
```
limbo> EXPLAIN BEGIN IMMEDIATE;
addr opcode p1 p2 p3 p4 p5 comment
---- ----------------- ---- ---- ---- ------------- -- -------
0 Init 0 4 0 0 Start at 4
1 Transaction 0 1 0 0
2 AutoCommit 0 0 0 0 auto_commit=false, rollback=false
3 Halt 0 0 0 0
4 Goto 0 1 0 0
```
Please note that SQLite emits *two* transaction instructions -- one for
main database and one for temporary tables. However, since we don't
support the latter, we only emit one transaction instruction.
This PR reworks the unix I/O backend, removing runtime reference
counting/borrow checking and optimizing away the hashmap in favor of a
static array, with an unlikely fallback vec.
The only reason the fallback vec is there is because unlike the
`io_uring` module, we cannot simply index into the array with the fd as
the OS could theoretically give us a fd up to I believe 1024 so keeping
an array of that size for a few elements is unnecessary.
Closes#940
In order [experimentally
compile](https://github.com/DougAnderson444/wit-limbo) `limbo_core` to a
[wasm component](https://component-model.bytecodealliance.org/), limbo
needed to have no reliance on `js`, `js-sys`, `wasm-bindgen`, et al.
(for those who aren't familiar, there are many `wasm` runtimes and not
all of them play nice with `wasm-bindgen`)
This PR simply cleans up the dependencies, and puts them behind optional
flags and whatnot in order to enable this. Both `log` and `tracing` were
being used, so I reduced this only to `tracing`.
End result is limbo can be used like this:
https://github.com/DougAnderson444/wit-limbo
We can open a discussion on the possibilities that running limbo as a
wasm component can offer, including potentially using composable
components to implement the sqlite runtime extensions, as well as giving
us a clean interface for PlatformIO operations -- define them once,
implement many ways on various platforms. I'm new to limbo, but it looks
like current extension are Rust based deps and features flags, whereas
sqlite is runtime, right? What if limbo was runtime extensible too?
The WIT interface is largely sync (though I believe wasmtime has an
async feature), but in my limited exposure to limbo so far a lot of the
wasm seems sync already anyway. Again, topic for further discussion.
Suffice to say, aligning these deps in this way paves the road for
further experiments and possibilities.
Related: https://github.com/neilg63/julian_day_converter/pull/2
Related: https://github.com/tursodatabase/limbo/issues/950
Closes: https://github.com/tursodatabase/limbo/issues/950Closes#983
