On Sun, Sep 29, 2013 at 6:38 PM, Jakub Zawadzki
<darkjames-ws@xxxxxxxxxxxx> wrote:
> On Sun, Sep 29, 2013 at 05:35:59PM -0400, Evan Huus wrote:
>> On Sun, Sep 29, 2013 at 3:56 PM, Jakub Zawadzki
>> <darkjames-ws@xxxxxxxxxxxx> wrote:
>> > But back to topic (cause you'll probably see this problem few more times).
>> > I don't quite get a point why we need to change everything to wmem.
>> > (To be honest I still don't quite get why we need wmem_ at all, but let's skip that).
>>
>> Wmem is mostly a replacement for emem, because emem was becoming
>> impossible to maintain and reason about (recall, for example, bug
>> #5284). If you could still deal with emem, congratulations, but the
>> rest of us mortals needed something more comprehensible.
>
> Evan, I *really* don't see what wmem_ is fixing.
Then I must be doing an absolutely awful job of explaining. I
apologize, I'm just not sure how else to approach it.
> I mostly see that we have few
> extra assertions to dissallow use it when not dissecting packets.
> But if we want to remove totally ep_ we'd need to fix this shit anyway.
>
> I'm not fan of this, sorry. Right now wmem_ is quite long in our eco system,
> so this talk is pointless. Please, just let's skip this discussion.
> I'll make some comments from time to time (/me just being a little conservative),
> but I think you can life with it? :-)
I wish we were all on the same page, I think that if we could come to
a common understanding then things would go more smoothly.
> Eventually I might try to lobby replacing wmem_packet_scope() with some pinfo_current()->pool [or pinfo->pool if we have pinfo pointer]
This is part of the plan I have for wmem, definitely. It's just not as
high priority for now.
> - that's all.
>
>> > But if we really want to do that (change everything to wmem_), we NEED some ep-like temporary pool (which will work both for UI and dissection),
>> > or some function which will return packet-pool or gui-pool if there's no dissection. Otherwise we need to remove some functionality.
>>
>> No. This is one of the things I dislike most about emem. Using ep_
>> memory outside of packet dissection provides no guarantees at all as
>> to when that memory is going to be freed, and makes it difficult if
>> not impossible to reason about scope. The majority of such uses I
>> expect to be converted to manually-managed memory. If there is some
>> obvious common UI-scope then we can easily create a new pool for that
>> memory, but the cases I have reviewed have not had any obvious common
>> scope, they are simply using ep_ and assuming that it goes away at
>> some point (and not while they still need it!)
>
> Ok, this can be fixed with g_free() after val_to_str_[dup|format].
> I'm fine with this. +1 from me.
>
>> P.S. I have a pleasant day-dream where libwireshark gets rewritten in
>> a garbage-collected language like Go, but I somehow suspect that isn't
>> going to happen...
>
> Why Go? if we talk about better languages I'd rather use D.
> If we talk about glib-language-environment than Vala has reference couting.
Because I already know Go :) It was just an idle thought, not a
serious proposal.
> Still there's popular GC for C: http://www.hpl.hp.com/personal/Hans_Boehm/gc/
> and there's alredy interested in using that: http://www.wireshark.org/lists/wireshark-dev/201210/msg00229.html
I have not looked closely, but I think it would be easy to cause
memory leaks by embedding memory addresses in packet bytes. Maybe
worth experimenting with though.
