proper check for filename before calling subprocess

[prakhar2b]

Based on discussion here, we should properly check for filename before process calls in utils.check_output(), and raise more intuitive exception and error message.

cc @jayantj

[tmylk]

@AverageS A tutorial to get started is in https://github.com/RaRe-Technologies/gensim/blob/develop/docs/notebooks/FastText_Tutorial.ipynb

[tmylk]

The cpython check_output code https://github.com/python/cpython/blob/master/Lib/subprocess.py

[AverageS]

@tmylk @prakhar2b
#1501
It should fix the problem

[pratikmjoshi]

@menshikh-iv Can I go ahead with this?

[menshikh-iv]

@pratikmjoshi of course

[pratikmjoshi]

@menshikh-iv Just to clarify, fix #1501 did not solve this issue, right? Am I supposed to add an exception and error message for an incorrect filepath to the FastText directory, and should it indicate that the path be changed to the executable? What should the error message look like? I assume that I'll have to take care of the windows case as well.

[menshikh-iv]

@pratikmjoshi #1501 solved this issue, but partially

• the solution should work on main OS system (Linux, MacOSX, Windows)
• the idea is simple: before running anything, check that this path exists & executable

[pratikmjoshi]

@menshikh-iv Sorry, but I've been getting a bit confused with the code. In utils.py, is a function like this for path checking ok? The code is basically from #1501 , but I've added a check for amending the path depending on the platform. What will be the Windows equivalent of /usr/bin/ (I've just put C:\Users\ ,pretty sure that its incorrect)? Am I missing anything else?

And at check_output....

[pratikmjoshi]

@menshikh-iv After making the necessary changes, the respective error messages are raised as below:

However, after entering the right path, I'm still getting this error on the notebook. Any advice?

[menshikh-iv]

@pratikmjoshi about path (/usr/bin, etc): they can be many varieties only on Linux + user can have some custom path, and we have not started with other OS. IMO this way doesn't look like the right way.

Also, sometimes, wappers want to path to "folder" (and have hard-coded executable), please check, how it works for all wrappers (it's not so simple).