Subject: Re: [boost] "peer reviewed" - Rights and responsibilities of maintainers
From: Robert Ramey (ramey_at_[hidden])
Date: 2018-10-18 15:17:02
On 10/18/18 4:12 AM, John Maddock via Boost wrote:
> I was about to agree with you, but there is one very important use case
> - that of the application pluggin.Â In that situation one would expect
> that as long as everything is built with -fvisibility-hidden so that
> each shared library is entirely self contained, then everything should
> really be fine.
Hmm - I'm not sure about this. Consider the C++ runtime library on
windows with msvc. The app specifies dynamic linking of some boost
library. I believe that will of necessity use the dynamically linked
version of the C++ runtime library. Now we call a plugin which includes
another boost library - statically linked this time. The plugin will
use the statically linked version of the C++ runtime library. Now as
the program executes, it might be calling different versions of the
C++standard library and who knows what else. Sooooo - I would be pretty
reluctant to start down this path. Perhaps this is why microsoft, in
it's wisdom, has separate "visibility" attributes for import / export.
In any case, this is one aspect that should really be solved at the
committee level - but they're occupied with more pressing issues such as
the spaceship operator.
Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk