Garbage collection issues using IJulia?

[jnoynaert]

The Julia kernel appears to refuse to garbage collect allocations on Windows?

Repeatedly calling:

x = fill(NaN, (5000000,8))

increases the size of the Julia process in memory by 305 mb chunks (with no limit, until reaching out of memory).
-Reassigning x = 0 does not reduce the process size in memory as reported by task manager.
-Calling GC.gc() does not release the additional memory.

When using the standard Julia REPL, memory growth halts at +710 mb, with the additional allocation released (to +305mb, expected size of the array in memory) after a manual garbage collection call.

This behavior holds true for any large array operations so far.

IJulia 1.14.1 (using default miniconda setup, conda v1.1.1)
Julia 1.0.2
Windows 7 SP1

[stevengj]

The Out array records all of the outputs of the previous cells, similar to IPython. So if x is the output of your cells, it gets saved. You can do empty!(Out) to force IJulia to forget the old outputs.

[stevengj]

See e.g. here and here for similar discussions about IPython.

[jnoynaert]

Makes sense. Knowing that saves the usability of jupyter/hydrogen for me for a lot of work--thanks!