You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Resolve the problem of using external components #20
This is especially important to resolve before submitting my XMPP client.
XMPP requires an XML parser, and various other components for security and authentication. It definitely falls outside the scope of our little project to rewrite the world and, in the case of some algorithms for cryptography, may violate the Boost license. So, it's necessary to define a policy for including such code in cpp-netlib or allowing the library to be extended to use such code.
First, it seems evident in our case that this can be partly done by deferring these to traits, so in the case of XML parsers we have:
namespace boost { namespace network { namespace traits {
template <
class Tag
>
struct xml_parser; // etc.
}}}
But where to put the parser code? So far, I've put things in the top level directory:
gnutls is released under LGPL so I think we can still use it. Am I right in thinking that we can't distribute anything that uses GPL, but that doesn't stop users of cpp-netlib from writing such code?
One problem I'm trying to solve here is that a lot of this 3rd party code is going to be reused across different protocols, so we'd need a subdirectory in the top level directory, such as boost/network/ext/ or boost/network/external/
I'd include a big compile-time hit under the "not scaling well" header. This could be a reason against adoption.
Just to update this discussion, now that cpp-netlib is not strictly header-only anymore, I think we have some leeway with getting some of the implementation parts moved out into linked-in modules. I'm no longer dogmatic on the header-only idea, but will still strive to get as much of the library implemented in a header-only fashion and where it's practical to do so.
Right now I'm thinking about allowing the HTTP implementation details that can be updated without causing a recompile of the client code. I think this will increase the likelihood that cpp-netlib would actually be useful in situations where being able to upgrade without recompilation is concerned.
I think this is important. Many of the issues reported on the ML describe problems upgrading different versions of cpp-netlib and watching it fail in different ways with different versions of boost. Consequently, I think we should test this with at least the latest stable version of boost, but probably also previous stable versions.
Now that I think about it a little more, I think we ought to have a long-running effort to move implementations into non-header only versions and just have a flag to determine whether to use the library in header-only mode. This will help a lot with the compile times and with deployment.
I'll deal with the Boost acceptability issues later on if/when they come up.
Activity
glynos commentedon Oct 19, 2010
deanberris commentedon Aug 4, 2011
Just to update this discussion, now that cpp-netlib is not strictly header-only anymore, I think we have some leeway with getting some of the implementation parts moved out into linked-in modules. I'm no longer dogmatic on the header-only idea, but will still strive to get as much of the library implemented in a header-only fashion and where it's practical to do so.
Right now I'm thinking about allowing the HTTP implementation details that can be updated without causing a recompile of the client code. I think this will increase the likelihood that cpp-netlib would actually be useful in situations where being able to upgrade without recompilation is concerned.
Thoughts?
glynos commentedon Aug 4, 2011
I think this is important. Many of the issues reported on the ML describe problems upgrading different versions of cpp-netlib and watching it fail in different ways with different versions of boost. Consequently, I think we should test this with at least the latest stable version of boost, but probably also previous stable versions.
deanberris commentedon Aug 5, 2011
Now that I think about it a little more, I think we ought to have a long-running effort to move implementations into non-header only versions and just have a flag to determine whether to use the library in header-only mode. This will help a lot with the compile times and with deployment.
I'll deal with the Boost acceptability issues later on if/when they come up.
glynos commentedon Jun 19, 2013
No longer relevant.