ANNOUNCEMENT: Live Wireshark University & Allegro Packets online APAC Wireshark Training Session
April 17th, 2024 | 14:30-16:00 SGT (UTC+8) | Online

Wireshark-dev: Re: [Wireshark-dev] Visual Studio 2013/2015/2017 compatibility and libraries

From: Graham Bloice <graham.bloice@xxxxxxxxxxxxx>
Date: Mon, 24 Apr 2017 18:01:45 +0100


On 24 April 2017 at 16:59, Peter Wu <peter@xxxxxxxxxxxxx> wrote:
On Mon, Apr 24, 2017 at 04:41:27PM +0100, Graham Bloice wrote:
> On 24 April 2017 at 15:28, Pascal Quantin <pascal.quantin@xxxxxxxxx> wrote:
>
> >
> >
> > 2017-04-24 16:25 GMT+02:00 Graham Bloice <graham.bloice@xxxxxxxxxxxxx>:
> >
> >>
> >>
> >> On 24 April 2017 at 14:56, Pascal Quantin <pascal.quantin@xxxxxxxxx>
> >> wrote:
> >>
> >>> Hi Peter
> >>>
> >>> 2017-04-24 15:43 GMT+02:00 Peter Wu <peter@xxxxxxxxxxxxx>:
> >>>
> >>>> Hi,
> >>>>
> >>>> Are there possible issues to be aware of when using the libraries (built
> >>>> with mingw/msvc2013) with the Wireshark binaries built with VS2017?
> >>>> When trying it with a friend, it seems to build and run with no issues.
> >>>>
> >>>> I thought that there can be problems with combining different MSVC
> >>>> runtime versions in one binary? Looking through the libraries, it seems
> >>>> that there is a combination of at least MingW (GeoIP?) and MSVC (Lua).
> >>>> (And apparently VS2015 (14.0) and VS2017 (14.1) are binary compatible.)
> >>>>
> >>>
> >>> Just a side note: most of our 3rd party libraries are compiled with
> >>> MinGW(32|64) except zlib (that we compile from scratch if I remember
> >>> properly, cannot check right now), Lua for 2.0.X and 2.2.X and Qt.
> >>> Historically we were using a MinGW Lua build but that was triggering a
> >>> specific MinGW bug and we switched to MSVC based Lua library to work around
> >>> this. But it required us to package Lua for every supported MSVC variants
> >>> (as you need to have the corresponding C runtime installed on your PC while
> >>> Wireshark installer packages the C runtime of the version you used to
> >>> compile), which was painful. That's also why it is highly suggested to
> >>> install the Qt package matching the MSVC version you intend to use for
> >>> compilation.
> >>> Since the MinGW bug was fixed, so master branch is again using the MinGW
> >>> based Lua library. The only pre-compiled 3rd party package built with MSVC
> >>> I can think to right now is Qt. As of today they offer MSVC2013 and
> >>> MSVC2015 flavors (there is also a MinGW build but for 32 bits only). It
> >>> looks like Qt 5.9 will provide a MSVC2017 version according to
> >>> http://lists.qt-project.org/pipermail/development/2016-Decem
> >>> ber/028159.html
> >>>
> >>> I think Graham proposed to discuss which MSVC to use (for 2.5 I guess)
> >>> during Sharkfest US.
> >>>
> >>>
> >>>
> >> The major issue with using DLL's that are compiled with a different C
> >> runtime (CRT) from the main application is the issue of different heaps for
> >> each CRT and mallocing from one and freeing from a piece of code using
> >> another leads to bad things happening.  If the malloc\free is all done from
> >> within code that uses the same CRT then you should be OK.
> >>
> >
> > This is exactly the issue we faced with our GeoIP library (bug 13578) and
> > why I added a GeoIP_free symbol.
> >
> >
> >> The reason why the CRT changed with each version of Visual Studio was
> >> that this was the only way for MS to service the CRT with bug fixes as it
> >> couldn't be upgraded with Windows Updates as that might break various
> >> applications (that used the ill-advised static linking to the CRT).
> >>
> >> The Universal CRT was introduced with Windows 10\VS 2015 to allow Windows
> >> Updates to service a portion of the runtime [1].  The part of the CRT that
> >> deals with heaps is now serviced by the OS, so theoretically apps built
> >> with VS2015 or later (with the Windows 10 SDK) can now use DLL's compiled
> >> with VS2015 or later.
> >>
> >
> >>
> >> [1]: https://blogs.msdn.microsoft.com/vcblog/2015/03/03/intr
> >> oducing-the-universal-crt/
> >>
> >
> > Nice. But do you know if practice is working as good as theory ?
> >
>
> I don't.  This [1] earlier MS blog entry goes into a little more detail
> about the split and says that the VS part contains process start-up and
> exception handling stuff.
>
> I do note however that vcpkg doesn't appear to build VS specific versions
> of libraries and is available for VS 2015 update 3 or later.

I doubt that they will make it available for pre-2015 versions, but
according to a vcpkg developer, there might be version suffixes in the
future:
https://www.reddit.com/r/cpp/comments/5ud9sr/if_youre_doing_windows_dev_and_not_using_vcpkg/

    | | vcpkg install rapidjson:x64-windows
    | And what compiler/version is this for?

    We use a system of triplets that each define an entire, consistent
    toolchain and build target. We only support targets of VS2015/2017
    today (they're binary-interchangeable).

    So, in the case of :x64-windows, this is built using the v140/v141
    ABI against Windows Desktop amd64. In the future, we look forward to
    adding :x64-win-vXYZ and beyond to ensure you never mix ABIs!


Who knows what will be in the next Visual Studio.  I haven't seen any announcements, but as VS 2017 was only released just over a month ago I don't expect any public announcements yet.

It's possible that future C++ language changes may force them to change the ABI.

--
Graham Bloice