[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: Avoid Scheme-computed Skyline_pair values to get collected before us
|
From: |
dak |
|
Subject: |
Re: Avoid Scheme-computed Skyline_pair values to get collected before use (issue 12708048) |
|
Date: |
Sun, 11 Aug 2013 21:07:03 +0000 |
On 2013/08/11 20:41:48, Keith wrote:
On Sun, 11 Aug 2013 12:46:04 -0700, <mailto:address@hidden> wrote:
>
https://codereview.appspot.com/12708048/diff/1/lily/side-position-interface.cc
> File lily/side-position-interface.cc (left):
>
>
https://codereview.appspot.com/12708048/diff/1/lily/side-position-interface.cc#oldcode216
> lily/side-position-interface.cc:216: Skyline my_dim;
>
> On 2013/08/11 19:30:46, Keith wrote:
>> I am merely curious whether get_*_property() creates a temporary
copy
>> of the Scheme object.
> However, get_property can call callbacks, and then
> the returned value is up for collection as soon as no SCM remains
> referring to the returned value.
>
That's it. The property in question is 'vertical-skylines' and it is
generated
by a callback. From the stacktrace we can determine that
get_*_property() was
called with pure==true to generate the object that is pointed to by
the copy of
'skyp' that is associated with the crash.
So get_*_property() used internal_get_pure_property() which returns
the computed
skylines as a SCM object /without/ storing it in the
'vertical-skylines'
property (because it wants to leave a pointer to callback there for
when we need
the real, unpure, skylines).
Well, I wanted to point out that the continuing presence of the problem
when compiling without optimization makes this problem scenario unlikely
since the SCM variable containing the returned value can't possibly be
optimized out of existence.
But it turns out that the original code never put the returned SCM value
into a variable in the first place but rather unsmobbed it immediately.
So yes, I'd definitely like to hear back whether current master still
exhibits the problem. It's conceivable that this change helped, though
there don't seem to be a lot of opportunities for garbage collection
between return of the SCM value and use of the Skyline_pair.
https://codereview.appspot.com/12708048/