Add workaround for Ruby garbage collection bug

[stanhu]

The Ruby interpreter has a bug in String#concat where the appended array may be garbage collected prematurely because the compiler optimized out a Ruby stack variable. We now call to_ary on the Protobuf object to ensure the array lands on the Ruby stack so the garbage collector sees it.

The real fix in the interpreter is described in https://bugs.ruby-lang.org/issues/18140#note-2, but most current Ruby
interpreters won't have this fix for some time.

[stanhu]

I considered adding this test:

diff --git a/spec/lib/parse_spec.rb b/spec/lib/parse_spec.rb
index 3705e28..7b4470b 100644
--- a/spec/lib/parse_spec.rb
+++ b/spec/lib/parse_spec.rb
@@ -1609,4 +1609,11 @@ $BODY$
       expect(query.select_tables).to eq(['bar', 'baz'])
     end
   end
+
+  it "parses a query with garbage collection" do
+    GC.stress = true
+    tables = described_class.parse(%(SELECT 1 FROM test WHERE 1 AND 2 AND 3)).tables
+    expect(tables).to eq(['test'])
+    GC.stress = false
+  end
 end

It slows the test down by a few seconds, and it doesn't always fail, depending on the environment and Ruby interpreter you have.

[stanhu]

@lfittl Do you think we should include the test above?

[lfittl]

@stanhu Thanks for the contribution & for investing the time to get to the root cause of this! (and for your patience - I was out on leave and missed this PR)

I think we can merge this without a test, since as you note the test slows things down, and isn't fully reliable. I'll add a comment at the top of the method linking the Ruby bug, so its clear why we're doing the .to_ary calls.

[lfittl]

Merged in 945cb8f

[stanhu]

@lfittl Thanks! Would you mind releasing a new version? It seems our team worked around the problem once by disabling garbage collection before running PgQuery#parse, but they ran into another case where they forgot to do this.

[lfittl]

@lfittl Thanks! Would you mind releasing a new version? It seems our team worked around the problem once by disabling garbage collection before running PgQuery#parse, but they ran into another case where they forgot to do this.

Yep - Released!