Attempting to free stack allocated channel_s object about facil.io HOT 3 CLOSED

Hi @adamansky ,

Thank you very much for opening this issue.

I can't replicate the issue at the moment, I'll need to upgrade my gcc and VM for that, which I'll try to do tonight.

In the meantime, I wonder - could this be a compiler error?

The stack memory is assigned a high reference count, preventing the fio_channel_free function from freeing stack allocated objects.

The result is that stack objects are either ignored or that they are "copied" to the heap (with a new reference count), but never freed.

In this case, the if (fio_atomic_sub(&ch->ref, 1)) statement is always true (always 7, which is luckier than always -1, but produces the same result).

Did I miss something?

Kindly,
Bo.

from facil.io.

I installed GCC 7... which installed version 7.4.0 (gcc version 7.4.0 (Ubuntu 7.4.0-1ubuntu1~16.04~ppa1))...

...the warning doesn't seem to pop up. I can't reproduce.

Perhaps it's a GCC issue that was fixed?

from facil.io.

Hi Boaz! I think compiler is quite objective here, for example

In function ‘fio_channel_free.part.26’,
    inlined from ‘fio_ch_set__insert_or_overwrite_.constprop’,
    inlined from ‘fio_filter_dup_lock_internal’ at lib/facil/fio.h:6127:10,
    inlined from ‘fio_channel_match_dup_lock.isra.67’ at lib/facil/fio.c:5416:14,
    inlined from ‘fio_subscribe’ at lib/facil/fio.c:5457:8:

Where ch is placed on stack in fio_channel_match_dup_lock() and then used in fio_channel_free() It seems the compiler is not smart enough to prove that fio_atomic_sub(&ch->ref, 1) will be always true in runtime. May be 7.4.0 is able to derive it :)

Yes I see your comment: /* avoid freeing stack memory */ in https://github.com/boazsegev/facil.io/blob/master/lib/facil/fio.c#L5395

So seems to be no problems here. Thank you!

from facil.io.