ffi.lua:56: expected align(#) on line 579

[LukaOo]

I have torch with Lua 5.2 installed on ubuntu 16.04. When I use require 'hdf5' I have got error

/root/torch/install/share/lua/5.2/hdf5/ffi.lua:56: expected align(#) on line 579
stack traceback:
/root/torch/install/share/lua/5.2/itorch/main.lua:166: in function </root/torch/install/share/lua/5.2/itorch/main.lua:159>
[C]: in function 'cdef'
/root/torch/install/share/lua/5.2/hdf5/ffi.lua:56: in function 'loadHDF5Header'
/root/torch/install/share/lua/5.2/hdf5/ffi.lua:60: in main chunk
[C]: in function 'dofile'
/root/torch/install/share/lua/5.2/torch/init.lua:54: in function 'include'
/root/torch/install/share/lua/5.2/hdf5/init.lua:29: in main chunk
[C]: in function 'require'
[string "hdf5 = require 'hdf5'..."]:1: in main chunk
[C]: in function 'xpcall'
/root/torch/install/share/lua/5.2/itorch/main.lua:209: in function </root/torch/install/share/lua/5.2/itorch/main.lua:173>
(...tail calls...)
/root/torch/install/share/lua/5.2/lzmq/poller.lua:75: in function 'poll'
/root/torch/install/share/lua/5.2/lzmq/impl/loop.lua:307: in function 'poll'
/root/torch/install/share/lua/5.2/lzmq/impl/loop.lua:325: in function 'sleep_ex'
/root/torch/install/share/lua/5.2/lzmq/impl/loop.lua:370: in function 'start'
/root/torch/install/share/lua/5.2/itorch/main.lua:381: in main chunk
[C]: in function 'require'
(command line):1: in main chunk
[C]: in ?

It seems the problem is in lua 5.2. I solved this issue when installed luajit.

[wpf5511]

I also get this problem , so I want to know does that mean I should reinstall torch use luajit?

[wpf5511]

@LukaOo

[LukaOo]

Yes, under luajit hdf5 works fine. I reinstalled torch with luajit

[FabbyD]

Same as #73, is anyone aware of this problem? What if we can't use LuaJIT?

[himanshuUsc]

I am having the same issue . I am working on Ubuntu 16.04 . I have lua version 5.2.3.

[ambarpal]

@FabbyD Did you find a fix? I also have the same problem, working on Ubuntu 15.10 with Lua 5.2.3. I cannot install LuaJIT because of memory issues.

[FabbyD]

Hmm, sadly no. Not by using Lua 5.1 atleast. I ended up using LuaJIT. I dont know what kind of memory issue you have, but it seems like most memory allocation in the torch library (mostly Tensors) are done in C which bypasses the LuaJIT's garbage collector. From what I have read, the memory limit in LuaJIT comes from its current GC. I know they are working on a new version that should get rid of this limitation (fingers crossed). Correct me if I'm wrong.
Anyways. Using LuaJIT on Ubuntu 16.04, I am able to load 20GB in memory from an hdf5 file. Have you given it a try? Since you are using Lua 5.2, you might have some refactoring to do since LuaJIT is not quite 5.2-compatible yet.

[ambarpal]

Ah, I see. I had this memory problem long ago with LuaJIT and then used 5.2 since. I will give it a try

[FabbyD]

If you're loading straight Lua arrays from the HDF5 file then you most probably will still have the same problem (if you load more than 2GB).

[ambarpal]

LuaJIT made hdf5 work, but now I'm struggling to load models saved in Lua 5.2 into LuaJIT. Getting incompatible byteCodes all over the place. I guess I will get the array problem also once this fixes :(

[kprimice]

Any update on that issue? I would also prefer to use Lua 5.2

[ambarpal]

@kprimice I still don't know what fixed it. But going back to Ubuntu 14.04 with a previous version of GCC(4.8.4) and Lua 5.1 worked. Lua 5.1 does not have the memory issue, so I'm working with this for now

[lymrty]

lxy@lxy-Lenovo:~/neuraltalk2$ th eval.lua -model /home/lxy/model_id1-501-1448236541.t7 -image_folder /home/lxy/Flickr8k_Dataset/Flicker8k_Dataset/ -num_images 10
/home/lxy/torch/install/bin/lua: /home/lxy/torch/install/share/lua/5.2/trepl/init.lua:389: /home/lxy/torch/install/share/lua/5.2/trepl/init.lua:389: /home/lxy/torch/install/share/lua/5.2/hdf5/ffi.lua:56: expected align(#) on line 579
stack traceback:
[C]: in function 'error'
/home/lxy/torch/install/share/lua/5.2/trepl/init.lua:389: in function 'require'
eval.lua:8: in main chunk
[C]: in function 'dofile'
.../lxy/torch/install/lib/luarocks/rocks/trepl/scm-1/bin/th:145: in main chunk
[C]: in ?

well, i am also meet this expected align(#) on line 579 issue when i am trying to running Andrej Karpathy's program NeuralTalk2, so thanks everyone who gives an solution about my issue.

[gdh1995]

Lua 5.1 on Ubuntu 16 also has such a problem.

I can't use LuaJIT because the memory cost of my project is larger than 4GB...

[obbrc]

I have the same issue on Ubuntu 16.04 using gcc-5. Following @ambarpal's suggestion, I have installed gcc/g++ 4.8 and used Lua 5.1, the above error is not happening anymore.

However I still have the same memory error (both Lua 5.1 and LuaJIT give "not enough memory" with 6GB while reading 9GB H5 file (system has more than 20GB memory).

[McLawrence]

I get a similar error despite using LuaJIT:

th -e "require 'hdf5'" ...<UserName>/torch/install/share/lua/5.1/trepl/init.lua:389: ...s/<UserName>/torch/install/share/lua/5.1/hdf5/ffi.lua:56: ')' expected near '_close' at line 1436

[jeamland]

And another one on FreeBSD:

> uname -a
FreeBSD bjrbsd 12.0-CURRENT FreeBSD 12.0-CURRENT #0 r320364: Mon Jun 26 12:35:03 PDT 2017     benno@bjrbsd:/src/obj/src/freebsd/sys/GENERIC-NODEBUG  amd64
> th -e "require 'hdf5'"
/src/torch/torch/install/share/lua/5.1/trepl/init.lua:389: /src/torch/torch/install/share/lua/5.1/hdf5/ffi.lua:56: unexpected symbol near '_Alignof' at line 133

[t-li]

I happen to need to load a bunch of large json files and I am suffering from the same issue on Lua5.1/5.2. For those who need to bypass LuaJIT heap limitation, might want to try torch-tds. And call collectgarbage("collect") properly as a temporary solution.

[spx1997]

torch-hdf5 with lua 5.2 and gcc 4.8 can work. I think it has a relationship with version of gcc. (ps: My version of g++ is also 4.8)