undefined symbol: ...

[emanjavacas]

Hi there,

I have a clean ucto installation from sudo apt install ucto.
When I compile the python extension, however, I can't import it since it fails with:

ImportError: /home/manjavacas/.pyenv/versions/anaconda3-4.4.0/lib/python3.6/site-packages/ucto.cpython-36m-x86_64-linux-gnu.so: undefined symbol: _ZN9Tokenizer14TokenizerClass4initERKSs

Not sure what might be going bad, since ucto works perfectly fine and the extension manages to compile without errors.

Any ideas?

[proycon]

Hmm, hard to say exactly what happened here. Might there have been two different ucto version installed or ucto was upgraded after the Python extension was installed? Or compiled against a different Python version than your anaconda one? If you want to dive into this, you can use ldd to check what the library was linked to exactly.

A solution to get things working quickly may be to simply use LaMachine, which installs both ucto and python-ucto for you (in this case you'd have to uninstall the ucto package and dependencies first though)

[emanjavacas]

You are right, there must've been some dependency issue. I've removed now everything related to ucto (ticcutils, libfolia, ucto). But installing ucto from the apt packages and trying to compile the extensions now results in a compilation failure not finding libfolia/folia.h:

running build
running build_ext
cythoning ucto_wrapper.pyx to ucto_wrapper.cpp
building 'ucto' extension
gcc -pthread -Wsign-compare -DNDEBUG -g -fwrapv -O3 -Wall -Wstrict-prototypes -fPIC -I/usr/include/ -I/usr/include/libxml2 -I/usr/local/include/ -I/home/manjavacas/.pyenv/versions/anaconda3-4.4.0/include/python3.6m -c ucto_wrapper.cpp -o build/temp.linux-x86_64-3.6/ucto_wrapper.o --std=c++0x -D U_USING_ICU_NAMESPACE=1
cc1plus: warning: command line option ‘-Wstrict-prototypes’ is valid for C/ObjC but not for C++
ucto_wrapper.cpp:455:28: fatal error: libfolia/folia.h: No such file or directory
compilation terminated.
error: command 'gcc' failed with exit status 1

However, libfolia did get installed as a dep of ucto:

[11:06:15][manjavacas@station ~/code/vendor/python-ucto](master %)$ ldconfig -p | grep folia
	libfolia.so.2 (libc6,x86-64) => /usr/lib/libfolia.so.2

[proycon]

You'll need libfolia-dev as well which probably didn't get pulled in automatically.

(Also be aware that these debian packages are a bit older than the latest releases and there is currently an open issue (proycon/python-frog#6) regarding a missing libfrog-dev preventing compilation of python-frog)

[emanjavacas]

Alright, thanks for the info:

I've installed libfolia-dev, libticcutils2-dev and libucto-dev.
After this, compilation fails on ucto with:

In file included from ucto_wrapper.cpp:457:0:
/usr/include/ucto/tokenize.h:33:20: fatal error: config.h: No such file or directory

I guess from your comment on the debian packages being older I gather that this issue is probably related to python-ucto relying on a newer version of ucto..

[proycon]

That does seem likely yes, you could roll-back to an older python-ucto but that's not really recommended. I've been trying to get the debian packages updated for a while already but it takes quite some time and our primary effort goes into LaMachine.

[emanjavacas]

I will try to compile the deps from scratch. I remember trying it not so long ago and having an issue with libtool version mismatches. I think it was ticcutils, so perhaps it makes sense to open an issue over there.

[proycon]

Though it's a bit of a hassle, that should always work yeah (it's what LaMachine does too under the hood). Just make sure all debian packages are purged otherwise there may be duplicates lingering which will surely mess things up.

[emanjavacas]

Alright, I managed to compile the extension after compiling deps from scratch.
For the record, the issue I was having with libtool was because anaconda was using its own installed version of it. Solution was to not just anaconda.