[jhawthorn]

Previously to_a was Enumerable#to_a which requires a call to step and done for each row. It's faster to just accumulate the full result.

require "sqlite3"
require "ips"

db = SQLite3::Database.new(":memory:")
hdb = SQLite3::Database.new(":memory:", results_as_hash: true)
[db, hdb].each do |d|
  d.execute("create table t (id integer primary key, i integer, s text)")
  d.transaction { 100.times { |k| d.execute("insert into t values (?, ?, ?)", [k, k * 7, "name-#{k}"]) } }
end
stmt = db.prepare("select * from t")

IPS.ips do |x|
  x.report("Statement#to_a")      { stmt.reset!; stmt.to_a }
  x.report("execute")             { db.execute("select * from t") }
  x.report("execute (as hash)")   { hdb.execute("select * from t") }
end

Before:

$ ruby -I/work/sqlite3-base/lib to_a.rb
      Statement#to_a:    40.241k i/s (± 8.7%, GC  8.9%)
             execute:    39.403k i/s (± 8.1%, GC  9.1%)
   execute (as hash):    16.734k i/s (± 6.8%, GC 17.1%)

Summary
  Statement#to_a ran
    1.02 ± 0.12 times faster than execute
    2.40 ± 0.26 times faster than execute (as hash)

After:

$ ruby -I/work/sqlite3-ruby/lib to_a.rb
      Statement#to_a:    52.023k i/s (± 6.4%, GC 11.3%)
             execute:    46.407k i/s (± 7.8%, GC 10.4%)
   execute (as hash):    33.971k i/s (± 6.9%, GC 10.5%)

Summary
  Statement#to_a ran
    1.12 ± 0.11 times faster than execute
    1.53 ± 0.14 times faster than execute (as hash)