Memory leak in Base - combining Tasks with eval

[schlichtanders]

Hi there

First time I am trying to use Julia for infinite jobs (streaming). It is horrible I have to admit unfortunately. Lost 1 full week to track down memory leaks. Here another one:

for _ in 1:50
    @eval fetch(schedule(Task(() -> ())))
    GC.gc(true); GC.gc(false)
    Core.println(Base.gc_live_bytes() / 2^20)
end

running this will output something like

1.36197
1.36372
1.36546
1.36721
1.36896
1.3707
1.37245
1.3742
1.37595
1.37769
1.37944
1.38119
1.38293
1.38468
1.38643
1.38818
1.38992
1.39167

[KristofferC]

Aren't you generating infinite amount of code here with generating new closures?

For example, something like

const clos = () -> ()
for _ in 1:50
    @eval fetch(schedule(Task(clos)))
    GC.gc(true); GC.gc(false)
    Core.println(Base.gc_live_bytes() / 2^20)
end

seems to be stable.

Probably, dup of #14495

[schlichtanders]

indeed, I may create new closures repeatedly (this is running as part of Pluto.jl).

if a closure is not cleaned up automatically when nothing references it, how can it be deleted manually?

[schlichtanders]

I tried naming the inner function but this also does not work

for _ in 1:50
    @eval function myfunc() end
    GC.gc(true); GC.gc(false)
    Core.println(Base.gc_live_bytes() / 2^20)
end

...
4.07169
4.07333
4.07497
4.07661
4.07825
4.07989
4.08153
4.08317

[schlichtanders]

Realized this is a complete duplicate 😄

The original issue already creates a named function and still experiences these problems.
The Task wrapper is not really needed, but yeah, it is a common context where this occurs.

I guess the greatest value I can add is to mention that this memory bug also happens when using Pluto.