metatable tab-completions don't seem to work as expected

[szym]

I like most trepl's features but the tab-completion of methods doesn't seem to work properly

require 'optim'
l = optim.Logger('out.log')
l:<TAB>

luajit alone gives:

l:add(       l:epsfile    l:idx.       l:names.     l:plot(      l:style(     l:symbols.
l:empty      l:file:      l:name       l:new(       l:setNames(  l:styles.

th (luajit with trepl) gives:

_G              break           false           io              module          pairs           repeat          sys             utils
_LAST           collectgarbage  for             ipairs          monitor_G       paths           repl_linenoise  table           while
_RESULTS        coroutine       function        jit             newproxy        pcall           repl_readline   then            who
_VERSION        debug           gcinfo          l               next            print           require         tonumber        xerror
...

Tab-completions of l.<TAB> are likewise wrong.

[soumith]

thanks, fixed via 817d362

[szym]

The fix is good but not perfect.

require 'nn'
m = nn.Sequential()
m:<TAB>

trepl:

__add                    __init                   __pow                    add                      share
__call                   __le                     __sub                    evaluate                 size
__concat                 __len                    __tostring               get                      training
__div                    __lt                     __tostring__             insert                   updateGradInput
__eq                     __mod                    __unm                    new                      updateOutput
__factory                __mul                    accGradParameters        parameters               updateParameters
__index                  __newindex               accUpdateGradParameters  reset                    zeroGradParameters

luajit:

model:accGradParameters(              model:float(                          model:sharedAccUpdateGradParameters(
model:accUpdateGradParameters(        model:forward(                        model:size(
model:add(                            model:get(                            model:training(
model:backward(                       model:getParameters(                  model:type(
model:backwardUpdate(                 model:insert(                         model:updateGradInput(
model:clone(                          model:modules.                        model:updateOutput(
model:cuda(                           model:new(                            model:updateParameters(
model:double(                         model:parameters(                     model:zeroGradParameters(
model:evaluate(                       model:reset(
model:findModules(                    model:share(

TAB-completions for nn.Module work as expected, so it seems that trepl completions just need to follow the inheritance chain. I can't say I understand how inheritance works in torch, but I think torch.getmetatable(torch.typename(t.__metatable)) will allow recursing into the super-metatable.