ipcluster touching ~/.ipython even though --ipython-dir is specified

[mforbes]

When trying to run ipcluster for testing in a temporary directory, ~/.ipython still gets created.

rm -rf ~/.ipython
mkdir /tmp/tmp5aa3_zqu
ipcluster start --profile=testing1 --n=2 --ipython-dir=/tmp/tmp5aa3_zqu

$ tree ~/.ipython/
/home/mforbes/.ipython/
`-- profile_default
    |-- db
    |-- log
    |-- pid
    |-- security
    `-- startup
        `-- README

Shouldn't --ipython-dir prevent this? (I don't want my tests creating files in someones home directory).

Note, this is very minor: /tmp/tmp5aa3_zqu is used for all the important stuff.

[mforbes]

Related probably is that if one locks ~/.ipython then starting a cluster populates many tempdirs instead, which do not get cleaned up.

mkdir ~/.ipython /tmp/tmp5aa3_zqu
chmod a-rwx ~/.ipython
ipcluster start --profile=testing1 --n=2 --ipython-dir=/tmp/tmp5aa3_zqu

$ tree /tmp
/tmp
...
|-- tmp5aa3_zqu                      # This is expected
|   |-- extensions
|   |-- nbextensions
|   `-- profile_testing1
|       |-- log
|       |-- pid
|       |   |-- ipcluster.pid
|       |   `-- ipcontroller.pid
|       |-- security
|       `-- startup
|           `-- README
|-- tmp60l8kd6o                      # This should not be created
|   `-- profile_default
|       |-- db
|       |-- log
|       |-- pid
|       |-- security
|       `-- startup
|           `-- README
|-- tmpup6dmrce                      # Neither should this...
|   `-- profile_default
|       |-- db
|       |-- log
|       |-- pid
|       |-- security
|       `-- startup
|           `-- README
...

[minrk]

Will have to investigate exactly what's the cause, but my guess sis that something is following IPython's default initialization machinery before the ipython-dir cli option is handled. Does it work if you set the IPYTHONDIR env instead of the command-line arg?

The separate temp directories are presumably the same thing - something is not getting passed the ipython-dir option and/or getting it after some initialization step has already done what it was going to do.

The temp directory is presumably this logic in get_ipython_dir(), which would fire if get_ipython_dir() is ever called. My guess is it's in one or more places for defaults, and maybe called even if not ultimately used. I think if we used the ipython-dir cli arg to set IPYTHONDIR env as soon as possible, it might be fixed.