Sometime readline do not have set_completer_delims

[tebeka]

No description provided.

[minrk]

@tebeka can you give an example of when this is unavailable?

[tebeka]

@minrk Every time I run Anaconda jupyter console on my Archlinux machine

[Carreau]

set_completer_delims is supposed to be available on readline module, if it's not it's:

• either a bug of the readline implementation on arch/anaconda-arch (in which case it should be fixed/ reported upstream)
• a missing statement in the docs (in which case it also need to be fixed)
• a bug in our code that fail to find set_completer_delims (we have a shim around readline)

can you import readline, and give us the result of readline.__file__, so that we can try to track that down ?

In [1]: import readline
In [2]: readline.__file__
Out[2]: '/Users/bussonniermatthias/anaconda3/lib/python3.5/lib-dynload/readline.cpython-35m-darwin.so'

[tebeka]

IMO this is a version mismatch between the installed readline on arch vs the one anaconda expects.
If I run the arch system python you'll see

Python 3.5.1 (default, Dec  7 2015, 12:58:09) 
[GCC 5.2.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import readline
>>> readline.__file__
'/usr/lib/python3.5/lib-dynload/readline.cpython-35m-x86_64-linux-gnu.so'

However if you run the Anaconda Python:

Python 3.5.1 |Anaconda 2.5.0 (64-bit)| (default, Dec  7 2015, 11:16:01) 
[GCC 4.4.7 20120313 (Red Hat 4.4.7-1)] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import readline
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
ImportError: libncursesw.so.5: cannot open shared object file: No such file or directory
>>> 

There is /opt/anaconda/lib/python3.5/lib-dynload/readline.cpython-35m-x86_64-linux-gnu.so however:

ldd /opt/anaconda/lib/python3.5/lib-dynload/readline.cpython-35m-x86_64-linux-gnu.so
    linux-vdso.so.1 (0x00007ffdbd1fd000)
    libreadline.so.6 => /opt/anaconda/lib/python3.5/lib-dynload/../../libreadline.so.6 (0x00007f5d7ba11000)
    libncursesw.so.5 => not found
    libpython3.5m.so.1.0 => /opt/anaconda/lib/python3.5/lib-dynload/../../libpython3.5m.so.1.0 (0x00007f5d7b4e8000)
    libpthread.so.0 => /usr/lib/libpthread.so.0 (0x00007f5d7b2cb000)
    libc.so.6 => /usr/lib/libc.so.6 (0x00007f5d7af2a000)
    libdl.so.2 => /usr/lib/libdl.so.2 (0x00007f5d7ad25000)
    libutil.so.1 => /usr/lib/libutil.so.1 (0x00007f5d7ab22000)
    librt.so.1 => /usr/lib/librt.so.1 (0x00007f5d7a91a000)
    libm.so.6 => /usr/lib/libm.so.6 (0x00007f5d7a614000)
    /usr/lib64/ld-linux-x86-64.so.2 (0x0000560a7aa0a000)

You'll see it depends on libncursesw.so.5 which is not in the system. Where's the system one uses libncursesw.so.6

ldd /usr/lib/python3.5/lib-dynload/readline.cpython-35m-x86_64-linux-gnu.so
    linux-vdso.so.1 (0x00007ffd9fbf9000)
    libreadline.so.6 => /usr/lib/libreadline.so.6 (0x00007f8646b8d000)
    libpython3.5m.so.1.0 => /usr/lib/libpython3.5m.so.1.0 (0x00007f86466bf000)
    libpthread.so.0 => /usr/lib/libpthread.so.0 (0x00007f86464a2000)
    libc.so.6 => /usr/lib/libc.so.6 (0x00007f8646100000)
    libncursesw.so.6 => /usr/lib/libncursesw.so.6 (0x00007f8645e93000)
    libdl.so.2 => /usr/lib/libdl.so.2 (0x00007f8645c8f000)
    libutil.so.1 => /usr/lib/libutil.so.1 (0x00007f8645a8b000)
    libm.so.6 => /usr/lib/libm.so.6 (0x00007f8645786000)
    /usr/lib64/ld-linux-x86-64.so.2 (0x0000558c0f766000)

[minrk]

That seems like a significant problem to bring up with Continuum, since it suggests that the readline isn't built correctly.

[Carreau]

ping @damianavila, @bollwyvl do you know who we should ping on the continuum team about that ?

[bollwyvl]

@ilanschnell is the man with the plan on this one. Certainly worth raising on https://github.com/ContinuumIO/anaconda-issues

[Carreau]

Thanks !

Might be related to ContinuumIO/anaconda-issues#455

[tebeka]

I agree that Continuum should address the problem their end. But IMO they can't cover all the issue with readline. This small patch guards against such problems in the future.

[Carreau]

This small patch guards against such problems in the future.

You can see that through the opposite view which is this patch hides an obvious bug in an underlying implementation or build of a specific package. It also confuses other dev into thinking set_completer_delims is not mandatory.

If there is not fixes in conda that get released we can weight the cost of adding that in, which could make sens, but I prefer (for now) to wait on wether ContinuumIO/anaconda-issues#455 will be fixed.

[takluyver]

(N.B. We're also ditching readline for 5.0, so this won't be a longstanding problem)

[Carreau]

Thanks, merging, if we do a 4.2 release this should be backported, not sure there will be a 4.2 though.

Sorry for the delay,

[bollwyvl]

Thanks, merging, if we do a 4.2 release this should be backported, not sure there will be a 4.2 though.

News to me: what's the development that suggests otherwise? Love to help out if I can!

[takluyver]

The reason not to do 4.2 is that we might just go straight to 5.0. IIRC there are only two minor things tagged for 4.2.