Lua - rewriting coroutine recursion to tail-call recursion

Question

I have to write an iterator that can traverse nested tables. I wrote one using coroutine .

It created an array of (path, value) pairs, eg {{key1, key2, key3}, value} meaning that to get value you got to do nested_table[key1][key2][key3] .

Upon that I wrote find() , findall() , in() with ease, life was bright.

function table.extend(tbl, new_value)
  local tbl = {table.unpack(tbl)}
  table.insert(tbl, new_value)
  return tbl
end

function iterate(tbl, parent)
  local parent = parent or {}
  if (type(tbl)=="table") then
    for key, value in pairs(tbl) do
      iterate(value, table.extend(parent, key))
    end
  end
  coroutine.yield(parent, tbl)
end

function traverse(root)
   return coroutine.wrap(iterate), root
end

Then I realized that the Lua environment I have to work with has coroutine blacklisted. We can not use that. So I try to get the same functionality without coroutine .

-- testdata

local pool = {}
test = {
  ['a'] = 1,
  ['b'] = {
    ['c'] = {2, 3},
    ['d'] = 'e'
  }
}

-- tree traversal

function table.extend(tbl, element)
  local copy = {table.unpack(tbl)}
  table.insert(copy, element)
  return copy
end

local function flatten(value, path)
  path = path or {'root'}
  pool[path] = value -- this is the 'yield'
  if type(value) == 'table' then
    for k,v in pairs(value) do
      flatten(v, table.extend(path, k))
    end
  end
end

-- testing the traversal function

flatten(test)

for k, v in pairs(pool) do
  if type(v) == 'table' then v = '[table]' end
  print(table.concat(k, ' / ')..' -> '..v)
end

This code returns what I need:

root -> [table]
root / b / c / 1 -> 2
root / b -> [table]
root / a -> 1
root / b / d -> e
root / b / c / 2 -> 3
root / b / c -> [table]

But I still have a problem: I can not use a global variable, pool , this code is called paralleled. And I can not do proper tail-call recursion ( return flatten(...) ) from a for cycle as it would return only once.

So my question: how do I package this function into something that can be called in parallel? Or in other words: can I achieve what the 'yield' part does with a return value, instead of piping the results into a global variable?

I tried to make it an object, following the patterns here , but I could not get it work.

Answer 1

You can make pool variable local:

test = {
   ['a'] = 1,
   ['b'] = {
      ['c'] = {2, 3},
      ['d'] = 'e'
   }
}

-- tree traversal

function table.extend(tbl, element)
   local copy = {table.unpack(tbl)}
   table.insert(copy, element)
   return copy
end

local function flatten(value, path, pool)    -- third argument is the pool
   path = path or {'root'}
   pool = pool or {}                                    -- initialize pool
   pool[path] = value
   if type(value) == 'table' then
      for k,v in pairs(value) do
         flatten(v, table.extend(path, k), pool)  -- use pool in recursion
      end
   end
   return pool                           -- return pool as function result
end

-- testing the traversal function

local pool = flatten(test)

for k, v in pairs(pool) do
   if type(v) == 'table' then v = '[table]' end
   print(table.concat(k, ' / ')..' -> '..v)
end