Re: [Quickfix-developers] quickfix memory leak

[K. F. <kfr...@gm...>]

Hi Oren!

Thanks for your reply.

On Fri, Sep 12, 2014 at 11:53 AM,  <or...@qu...> wrote:
> We have been pretty conservative up until this point, but that was
> largely due to how poor and divergent compilers were when the project
> first started.
> ...
> We are also starting to get more aggressive dropping out of date
> toolchains.  We supported VS6 for a long long time, but we now only
> supports VS10 and up.
>
> The philosophy on the whole has been to get quickfix up and running for
> people with as few dependencies as possible.

I like that philosophy.  (But why then xml?  That was the one thing that
caused me the most nuisance "porting" QuickFIX to mingw_w64.)

> That left libraries like boost out.

I wholeheartedly agree.  Boost has some good stuff in it, but it's hugely
bloated.  I think there may be some conceit on boost's part.  (Our code
don't stink ...)  Why should I have to download a template-metaprogramming
state-machine framework just to use a smart pointer?  Boost really ought
to make a (small) base library of low-level utilities that are clearly broadly
useful (e.g. smart pointers).  Then they could implement all of their
whiz-bang experiments in separate add-on libraries.

> ...
> With the
> latest 1.14.1 coming out, we were able to drop the dependencies on
> libxml2 and msxml.  This made our PyPI installation a lot easier too.

As noted above, xml was something of an issue for me.  How did you
get rid of the libxml2 / msxml dependency?

> I feel that if anyone is running a legacy system that old, they must
> really love legacy software and can continue to use previous stable
> releases of QuickFIX.

Personally, I agree with this.  But I understand and respect that
many projects take a more legacy-focused approach.  But I'm with
you guys -- keep moving forward!

> I'd really like to use more of C++11.  In fact I was looking to convert
> all our std for loops to range based loops, but alas VS 2010 doesn't
> support them.  They make the code so much cleaner.

(Then drop VS 2010.  Hee-hee ...)

Another, more substantive change you could make would be to switch
your threading to c++11's std::thread (and friends).  Let the tool vendors
(i.e., VS and gcc) worry about the platform-specific code.  std::thread is
broadly supported now, among others, by VS, gcc on linux, and gcc on
windows (mingw-w64).


Best regards.


K. Frank


>> -------- Original Message --------
>> Subject: Re: [Quickfix-developers] quickfix memory leak
>> From: "K. Frank" <kfr...@gm...>
>>
>> Hi Oren!
>>
>> On Fri, Sep 12, 2014 at 12:49 AM,  <or...@qu...> wrote:
>> >
>> > So it turns out there was a leak here.  The returned copies were not
>> > being stored in that collection and were never disposed of.  I solved
>> > this right now by putting them in std::shared_ptr objects.
>> > ...
>> > I think it would be nice to be able to use some modern smart pointers.
>> > When the project was started back in 2001, there was really nothing
>> > standardized that was widely distributed other than the dependency
>> > monster boost.  Now that there is I think it would help us avoid these
>> > types of memory pitfalls.
>>
>> I would personally advocate for the use of std::shared_ptr (and other
>> standard smart pointers, as appropriate), but, of course, this would
>> make QuickFIX (partially) dependent on c++11.
>>
>> For me, that's fine (in fact preferable), but I understand that many
>> users may have older toolchains that don't support the standard
>> smart pointers (or other c++11 / c++14 features).  What's been the
>> QuickFIX philosophy about adopting new language features when
>> they're beneficial?
>>
>> Thanks.
>>
>> K. Frank
>> ...