Strange assert_equals behaviour for tables returned from or passed to server:exec

[sergepetrenko]

TL; DR:

1. t.assert_equals(expect_tbl, return_tbl) on line 19 fails, but
   t.assert_equals(return_tbl, expect_tbl) on line 20 passes!
2. t.assert_equals(tbl, expect) on line 24 fails, but
   t.assert_equals(expect, tbl) on line 23 passes!

The problems I see:
a) server:exec() encodes table arguments / return values with extra box.NULL when passing them to / from other server
b) assert_equals works when the argument passed over the net is first, but doesn't work when that argument is second.

The details:
Consider the following reproducer:

repro_test.lua

local t = require('luatest')
local server = require('luatest.server')
local g = t.group('repro')

g.test_repro = function(cg)
    local server = server:new{}
    server:start()
    local expect_tbl = server:exec(function()
        return {1, 2, 3, 4, 5, 6, 7, 8, 9}
    end)
    local return_tbl = server:exec(function(expect)
        local tbl = {1, 2, 3, 4, 5, 6, 7, 8, 9}
        t.assert_equals(expect, tbl)
        t.assert_equals(tbl, expect)
        tbl[16] = 'something'
        return tbl
    end, {expect_tbl})
    expect_tbl[16] = 'something'
    t.assert_equals(expect_tbl, return_tbl)
    t.assert_equals(return_tbl, expect_tbl)
    server:exec(function(expect)
        local tbl = {1, 2, 3, 4, 5, 6, 7, 8, 9, [16] = 'something'}
        t.assert_equals(expect, tbl)
        t.assert_equals(tbl, expect)
    end, {expect_tbl})
    server:drop()
end

When I run it with luatest, luatest repro_test.lua -v, it fails like this:

fail 1

Failed tests:
-------------

1) repro.test_repro
/Users/s.petrenko/Source/sptnt/test/box-luatest/r_test.lua:19: expected: 
{
    1,
    2,
    3,
    4,
    5,
    6,
    7,
    8,
    9,
    cdata<void *>: NULL,
    cdata<void *>: NULL,
    cdata<void *>: NULL,
    cdata<void *>: NULL,
    cdata<void *>: NULL,
    cdata<void *>: NULL,
    "something",
}
actual: 
{1, 2, 3, 4, 5, 6, 7, 8, 9, [16] = "something"}
List difference analysis:
* lists A (actual) and B (expected) have the same size
* lists A and B start differing at index 17
* lists A and B are equal again from index 1
* Common parts:
  = A[1], B[1]: 1
  = A[2], B[2]: 2
  = A[3], B[3]: 3
  = A[4], B[4]: 4
  = A[5], B[5]: 5
  = A[6], B[6]: 6
  = A[7], B[7]: 7
  = A[8], B[8]: 8
  = A[9], B[9]: 9
  = A[10], B[10]: nil
  = A[11], B[11]: nil
  = A[12], B[12]: nil
  = A[13], B[13]: nil
  = A[14], B[14]: nil
  = A[15], B[15]: nil
  = A[16], B[16]: "something"
* Common parts at the end of the lists
  = A[1], B[1]: 1
  = A[2], B[2]: 2
  = A[3], B[3]: 3
  = A[4], B[4]: 4
  = A[5], B[5]: 5
  = A[6], B[6]: 6
  = A[7], B[7]: 7
  = A[8], B[8]: 8
  = A[9], B[9]: 9
  = A[10], B[10]: nil
  = A[11], B[11]: nil
  = A[12], B[12]: nil
  = A[13], B[13]: nil
  = A[14], B[14]: nil
  = A[15], B[15]: nil
  = A[16], B[16]: "something"
stack traceback:
	/Users/s.petrenko/Source/sptnt/test/box-luatest/r_test.lua:19: in function 'repro.test_repro'
	...
	[C]: in function 'xpcall'

Now, if I comment the line 19 (t.assert_equals(expect_tbl, return_tbl)), it starts failing in another place.
(Note, that line 20 passes successfully (t.assert_equals(return_tbl, expect_tbl)) and it has the same arguments as line 19 in reverse order.

fail2

Tests with errors:
------------------

1) repro.test_repro
{"status":"fail","class":"LuatestError","message":"\/Users\/s.petrenko\/Source\/sptnt\/test\/box-luatest\/r_test.lua:24: expected: \n{\n    1,\n    2,\n    3,\n    4,\n    5,\n    6,\n    7,\n    8,\n    9,\n    cdata<void *>: NULL,\n    cdata<void *>: NULL,\n    cdata<void *>: NULL,\n    cdata<void *>: NULL,\n    cdata<void *>: NULL,\n    cdata<void *>: NULL,\n    \"something\",\n}\nactual: \n{1, 2, 3, 4, 5, 6, 7, 8, 9, [16] = \"something\"}\nList difference analysis:\n* lists A (actual) and B (expected) 
stack traceback:
	/Users/s.petrenko/Source/sptnt/test/box-luatest/r_test.lua:21: in function 'repro.test_repro'
	...
	[C]: in function 'xpcall'

Ok, now I comment out line 24 (t.assert_equals(_G.ballot, expect)) and the test passes! So:

1. t.assert_equals(expect_tbl, return_tbl) on line 19 fails, but
   t.assert_equals(return_tbl, expect_tbl) on line 20 passes!
2. t.assert_equals(tbl, expect) on line 24 fails, but
   t.assert_equals(expect, tbl) on line 23 passes!

The problems I see:
a) server:exec() encodes table arguments / return values with extra box.NULL when passing them to / from other server
b) assert_equals works when the argument passed over the net is first, but doesn't work when that argument is second.

[Totktonada]

I use the following helper for the strict validation (with nil/box.NULL differentiation).

-- Compare types deeply.
--
-- The main motivation is to differentiate nil and box.NULL.
--
-- The following calls do not raise an error:
--
-- * t.assert_equals(nil, box.NULL)
-- * t.assert_equals(box.NULL, nil) too.
-- * t.assert_equals({foo = box.NULL}, {foo = nil})
--
-- However, the following does:
--
-- * t.assert_equals({foo = nil}, {foo = box.NULL})
--
-- See also https://github.com/tarantool/luatest/issues/310
local function assert_type_equals(a, b)
    t.assert_equals(type(a), type(b))
    if type(a) ~= 'table' then 
        return
    end  
    assert(type(b) == 'table')

    for k, v in pairs(a) do
        assert_type_equals(v, b[k])
    end  
    for k, v in pairs(b) do
        assert_type_equals(v, a[k])
    end  
end

My usage:

t.assert_equals(res, exp)
assert_type_equals(res, exp)

I doubt that we can strengthen the assert_equals() function, but it should be OK to add assert_equals_strict() (or whatever we'll name it).

OTOH, we added the strict flag for tap's is_deeply() method four years ago (see tarantool/tarantool@8e33785 plus tarantool/tarantool@676582c) and it was never used, at least in tarantool tests.