xflatten fails to update two fields in a nested record

[Totktonada]

How to reproduce:

#!/usr/bin/tarantool

local avro = require('avro_schema')
local log = require('log')
local json = require('json')

local schema={
    name='foo',
    type='record',
    fields={
        {name='bar', type={
            name='bar_t',
            type='record',
            fields={
                {name='value_1', type='string'},
                {name='value_2', type='string'},
            }}  
        },  
    }   
}

local _, schema_handler = avro.create(schema)
local _, model = avro.compile(schema_handler)

-- uncomment to produce case 2
-- model.unflatten({"ccc", "ddd"})

local obj_partial = {bar={value_1="aaa",value_2="bbb"}}
local ok, exps = model.xflatten(obj_partial)
log.info('ok: ' .. json.encode(ok))
log.info('exps: ' .. json.encode(exps))

Example of output:

./avro_case_3.lua: ok: false
./avro_case_3.lua: exps: "Internal error: unknown code"

Or with uncommented unflatten:

./avro_case_3.lua: ok: true
./avro_case_3.lua: exps: [["=",2,"bbb"],"value_2",["=",1,"aaa"]]

gdb output after break in unparse_msgpack() ('unknown code' case, w/ commented unflatten):

(gdb) p nitems
$1 = 10
(gdb) p state->ot
$4 = (uint8_t *) 0x9427d0 "\v\v\022\004\b"
(gdb) p state->ot+5
$9 = (uint8_t *) 0x9427d5 ""
(gdb) p state->ot+6
$10 = (uint8_t *) 0x9427d6 "\v\022\004\b"
(gdb) p *typeid
$14 = 0 '\000'

I will follow up with temporary workarounds in the comments.

[Totktonada]

Fast-written & dirty workaround for both cases described above.

local _slow_xflatten_helper
_slow_xflatten_helper = function(schema_names, top_path, obj_partial)
    local exps = {}
    for k, v in pairs(obj_partial) do
        local cur_path = (top_path == '') and k or (top_path .. '.' .. k)
        if type(v) == 'table' then
            local ok, new_exps = _slow_xflatten_helper(
                schema_names, cur_path, v)
            if not ok then
                return false, new_exps
            end
            for _, exp in ipairs(new_exps) do
                exps[#exps + 1] = exp
            end
        else
            local column_no = -1
            for i, name in ipairs(schema_names) do
                if name == cur_path then
                    column_no = i
                    break
                end
            end
            if column_no == -1 then
                local reason = 'cannot find fields name: ' .. cur_path
                return false, reason
            end
            exps[#exps + 1] = {'=', column_no, v}
        end
    end
    return true, exps
end

-- Note: no array, no enum support
local slow_xflatten = function(schema_handle, obj_partial)
    local schema_names = avro.get_names(schema_handle)
    return _slow_xflatten_helper(schema_names, '', obj_partial)
end

local xflatten_wrapper = function(model, schema_handle, obj_partial)
    local valid_exps = function(exps)
        for _, exp in ipairs(exps) do
            local valid_exp = type(exp) == 'table' and #exp == 3
            if valid_exp then
                -- XXX: more checks?
                -- local op, field_no, value = unpack(exp)
            else
                return false
            end
        end
        return true
    end
    
    local ok, exps = model.xflatten(obj_partial)
    local unkn_msg = 'Internal error: unknown code'
    local is_case_1 = not ok and type(exps) == 'string' and exps == unkn_msg
    local is_case_2 = ok and type(exps) == 'table' and not valid_exps(exps)
    if not (is_case_1 or is_case_2) then
        return ok, exps
    end
    return slow_xflatten(schema_handle, obj_partial)
end

Then invoke xflatten_wrapper(model, schema_handle, obj_partial) instead of model.xflatten(obj_partial), where schema_handle is what avro.create() returns and model is what avro.compile() returns (assuming local avro = require('avro_schema')).

[Totktonada]

Works correctly in both cases with debug (w/o optimization):

local _, model = avro.compile({schema_handler, debug=true})

The input of unparse_msgpack w/o optimization:

unparse_msgpack; *typeid: 0x0B (11) -- PUTARRAYC; value: 2   !!!
unparse_msgpack; *typeid: 0x0B (11) -- PUTARRAYC; value: 3
unparse_msgpack; *typeid: 0x12 (18) -- PUTSTRC; value: 1
unparse_msgpack; *typeid: 0x04 (4) -- PUTINT / PUTLONG; value: 2
unparse_msgpack; *typeid: 0x08 (8) -- PUTSTR; value: 3
unparse_msgpack; *typeid: 0x0B (11) -- PUTARRAYC; value: 3
unparse_msgpack; *typeid: 0x12 (18) -- PUTSTRC; value: 1
unparse_msgpack; *typeid: 0x04 (4) -- PUTINT / PUTLONG; value: 1
unparse_msgpack; *typeid: 0x08 (8) -- PUTSTR; value: 3

The input w/ optimization (broken):

unparse_msgpack; *typeid: 0x0B (11) -- PUTARRAYC; value: 3   !!!
unparse_msgpack; *typeid: 0x0B (11) -- PUTARRAYC; value: 3
unparse_msgpack; *typeid: 0x12 (18) -- PUTSTRC; value: 1
unparse_msgpack; *typeid: 0x04 (4) -- PUTINT / PUTLONG; value: 2
unparse_msgpack; *typeid: 0x08 (8) -- PUTSTR; value: 3
unparse_msgpack; *typeid: 0x00 (0) -- (zero); value: 0       !!!
unparse_msgpack; *typeid: 0x0B (11) -- PUTARRAYC; value: 3
unparse_msgpack; *typeid: 0x12 (18) -- PUTSTRC; value: 1
unparse_msgpack; *typeid: 0x04 (4) -- PUTINT / PUTLONG; value: 1
unparse_msgpack; *typeid: 0x08 (8) -- PUTSTR; value: 3