Return out_of_memory error tuple if row alloc fails

[ruslandoga]

Attempt at #301 (comment)

This PR makes multi_step allocate cells array once, and return an error tuple if that allocation fails.

Some benchmarks (the "10 rows" one might need to be re-run, it seems to be an outlier):

This approach

Mix.install [{:exqlite, github: "ruslandoga/exqlite", branch: "make-row"}, :benchee], force: true

sql = """
with recursive cte(i) as (
  values(0)
  union all
  select i + 1 from cte where i < ?
)
select i, i / 1.0, null, x'000000', 'hello' || i from cte
"""

alias Exqlite.Sqlite3

Benchee.run(
  %{"fetch_all" => fn %{conn: conn, stmt: stmt} -> Sqlite3.fetch_all(conn, stmt) end},
  inputs: %{
    "10 rows" => 10,
    "100 rows" => 100,
    "1000 rows" => 1000,
    "10000 rows" => 10000
  },
  before_scenario: fn rows ->
    {:ok, conn} = Sqlite3.open(":memory:", [:readonly, :nomutex])
    {:ok, stmt} = Sqlite3.prepare(conn, sql)
    Sqlite3.bind(conn, stmt, [rows])
    %{conn: conn, stmt: stmt}
  end,
  after_scenario: fn %{conn: conn, stmt: stmt} ->
    Sqlite3.release(conn, stmt)
    Sqlite3.close(conn)
  end
)

Results:

##### With input 10 rows #####
Name                ips        average  deviation         median         99th %
fetch_all       87.22 K       11.47 μs    ±71.28%        9.83 μs       47.25 μs

##### With input 100 rows #####
Name                ips        average  deviation         median         99th %
fetch_all       11.47 K       87.22 μs    ±14.59%       86.13 μs      128.13 μs

##### With input 1000 rows #####
Name                ips        average  deviation         median         99th %
fetch_all        1.24 K      809.13 μs     ±1.90%      808.29 μs      850.19 μs

##### With input 10000 rows #####
Name                ips        average  deviation         median         99th %
fetch_all        139.13        7.19 ms     ±9.00%        7.45 ms        8.45 ms

Master

Mix.install [:exqlite, :benchee], force: true

sql = """
with recursive cte(i) as (
  values(0)
  union all
  select i + 1 from cte where i < ?
)
select i, i / 1.0, null, x'000000', 'hello' || i from cte
"""

alias Exqlite.Sqlite3

Benchee.run(
  %{"fetch_all" => fn %{conn: conn, stmt: stmt} -> Sqlite3.fetch_all(conn, stmt) end},
  inputs: %{
    "10 rows" => 10,
    "100 rows" => 100,
    "1000 rows" => 1000,
    "10000 rows" => 10000
  },
  before_scenario: fn rows ->
    {:ok, conn} = Sqlite3.open(":memory:", [:readonly, :nomutex])
    {:ok, stmt} = Sqlite3.prepare(conn, sql)
    Sqlite3.bind(conn, stmt, [rows])
    %{conn: conn, stmt: stmt}
  end,
  after_scenario: fn %{conn: conn, stmt: stmt} ->
    Sqlite3.release(conn, stmt)
    Sqlite3.close(conn)
  end
)

Results:

##### With input 10 rows #####
Name                ips        average  deviation         median         99th %
fetch_all       50.65 K       19.74 μs    ±39.01%       18.42 μs       59.86 μs

##### With input 100 rows #####
Name                ips        average  deviation         median         99th %
fetch_all       10.91 K       91.70 μs    ±13.32%       90.00 μs      129.14 μs

##### With input 1000 rows #####
Name                ips        average  deviation         median         99th %
fetch_all        1.19 K      840.99 μs     ±1.63%      840.01 μs      880.36 μs

##### With input 10000 rows #####
Name                ips        average  deviation         median         99th %
fetch_all        130.85        7.64 ms     ±8.54%        7.88 ms        8.82 ms

[warmwaffles]

I prefer to have brackets around all if statements. I've been bitten too many times by not having them.

Unsure on what to do when sqlite3_column_count returns 0. The behavior I originally implemented just returned an error, but I don't know if that is appropriate. 🤔

[warmwaffles]

This can return 0

https://www.sqlite.org/c3ref/column_count.html

We should handle this. I don't think an error is appropriate here. I'm not 100% certain.

[ruslandoga]

Suggested change

	if (!cells) {
	if (!cells && column_count > 0) {

Something like this?

[warmwaffles]

Yea I think that would work. It would be safe because the list of results returned would be an empty array.

[ruslandoga]

Suggested change

	if (!cells) {
	if (!cells && column_count > 0) {

[warmwaffles]

This needs to be rebased.