Interesting GC.start

Ruby
1

irb:
name='1’10241024*100;nil # 100m
name=rand
GC.start

$ ps -ef|grep irb
ax 14081 13958 0 14:57 pts/2 00:00:00 irb

$ top 14081

14081 ax 20 0 114m 112m 1960 S 0 11.3 0:00.66 irb
14081 ax 20 0 13824 11m 1960 S 0 1.2 0:00.54 irb

when “name=rand;GC.start”, 112m->11m,why?

but “name=nil;GC.start”,dont work

2

2009/10/30 Haoqi H. [email protected]:

14081 ax 20 0 114m 112m 1960 S 0 11.3 0:00.66 irb
14081 ax 20 0 13824 11m 1960 S 0 1.2 0:00.54 irb

when “name=rand;GC.start”, 112m->11m,why?

Because the large string is eligible for collection.

but “name=nil;GC.start”,dont work

What do you mean? If “name=nil” does not work you should probably
recompile your Ruby interpreter from the sources.

If you mean that you then do not observe a memory drop the reason is
that although the large string is eligible for collection there is no
guarantee when it will be collected. That’s how automatic GC works -
and is supposed to work.

Cheers

robert

3

but “name=nil;GC.start”,dont work

If you mean that you then do not observe a memory drop the reason is
that although the large string is eligible for collection there is no
guarantee when it will be collected. That’s how automatic GC works -
and is supposed to work.

Cheers

robert
“name=nil;GC.start” dont work:
I mean that then do not observe a memory drop.
And “name=‘1’*1;GC.start” it also works(observer a memory drop)

4

Haoqi H. wrote:

irb:
name='1’10241024*100;nil # 100m
name=rand
GC.start

…

when “name=rand;GC.start”, 112m->11m,why?

but “name=nil;GC.start”,dont work

GC is tricky, because depending on what method calls, “ghost” references
to the original name might still exist on the stack.
You could try it with jruby/REE/1.9 to see if one of them works better,
but in this instance it appears that running “some methods” clears the
stack while others leave it dirty.
-r