Boost requirement

[RagnarDa]

Nice work! How much would be needed to remove the boost dependency?

[daniele77]

Thank you.
The CLI library uses boost asio to provide the remote session feature, and to get the async engine.
IMHO is almost impossible to remove the dependency.

[nightlark]

I recently replaced Boost.Asio (to get rid of the dependency on the compiled Boost.System library it uses) in a project. It was really straightforward, since Asio is pretty much a drop-in replacement aside from changing the namespace for functions and using things like std::error instead of the boost::system error.

[daniele77]

Great. I knew already Asio, maybe I'm gonna do some experiment in a branch.
Thank you very much!

[berge2007]

you can drop the parser too and build interactivity upon this: https://github.com/CLIUtils/CLI11

[qsc-jhndnn]

It was pretty straightforward to remove any boost requirement from cli.h - just replacement of boost::lexical_cast with a simple std::istringstream version. I was then able to hook it up to my own socket server. I've attached a diff - note that this has only been tested with VS2015 and then only slightly - YMMV. But it might be a good starting point for officially removing the boost requirement - or at least making it optional.
deboost.txt

[daniele77]

@jhndnnqsc yes, as one can see from your patch, removing boost from cli.h is easy enough (I already knew: I used boost lexical_cast just to cut some LOC because I already had the boost dependency used in the telnet server).

I guess the hard work would be removing boost from the telnet server. I won't put an implementation of mine because really don't want to face all the issues with multiplatform socket programming.
Instead, a good idea could be to use Asio as a replacement of Boost.Asio (as suggested by @nightlark ), but to get some benefit I should be able to embed Asio in my library (providing Asio license allows me to do so: I see Asio comes with Boost License. Does that mean I can embed the Asio sources directly in my library?)

[nightlark]

The Boost Software License has less requirements to comply with the license than BSD (at least 2 and 3 clause variants), and doesn't have the vagueness regarding binaries that the MIT license has.

It let's you copy/use/modify the software for anything you want (commercial or non-commercial) without a fee, so it can be copied/redistributed (git submodule or git subtree might make managing a copy of Asio in your repository easier).

To follow the copyright notice requirements:

• a copy of the license must be included with copies/redistributions of the source code
• executables/binary copies ("machine-executable object code generated by a source language processor") do not require the license be included

The copyright notices in the Software and this entire statement, including
the above license grant, this restriction and the following disclaimer,
must be included in all copies of the Software, in whole or in part, and
all derivative works of the Software, unless such copies or derivative
works are solely in the form of machine-executable object code generated by
a source language processor.

[qsc-jhndnn]

I think removing the boost requirement from cli.h is useful even if you do have to supply your own telnet or ssh server. At least in my case - I already have a libevent based telnet server in my software and don't wish to add a boost requirement just to hook up to my server.

[daniele77]

Ok, I'll see what I can do here. The ideal solution for me would be: make the cli using a generic communication server (not necessarily telnet, BTW) that could be:

• the boost version currently provided by the library OR
• implemented by the library user.

The boost server will be implemented in a separated header file so the boost dependency will disappear if it's not used.

What do you think? Could be a good solution for you?

[qsc-jhndnn]

I think that would work. If what was passed into Session was a wrapper for a socket that could either contain a boost socket or some other socket that implemented the same functionality I think that would work for me.

As a data point, here's an idea of what my socket interface looks like that I would like to swap the boost implementation for

class tcp_socket_t
{
public:
enum event_e { ev_connected, ev_reconnect, ev_data, ev_closed, ev_error, ev_timeout, ev_login_failed };
enum eol_e { eol_any, eol_crlf, eol_crlf_strict, eol_lf, eol_null, eol_custom };
public:
tcp_socket_t(...);
virtual ~tcp_socket_t();
bool is_connected() const;
void disconnect();
void write(const std::string& data);
void disconnect();
sigc::signal<void, tcp_socket_t*, event_e> signal;
virtual std::string read(int count); // nonblocking
std::string peek(int count); // nonblocking
virtual bool read_line(std::string& line, eol_e eol, const char* custom, size_t len); // nonblocking
};

[daniele77]

@jhndnnqsc well, actually I was oriented more towards an interface of higher level than sockets, e.g., transform Server and Session classes into interfaces.
In this way, one implementation could be the current BoostServer / BoostSession and the user can provide its own not based on boost.

[qsc-jhndnn]

That would work as well. As long as code like what's in remotecli.h can be shared since you've already figured out the tricky bits of dealing with a telnet connection and it'd be great to reuse those.

[connorjak]

It looks like using Asio standalone instead of boost/asio has been mentioned but not conversed about; that would be a drop-in replacement for all the networking features from what I was able to see in the code. Asio has a TINY compile time compared to full boost.

Another option for vcpkg users is to segment the remote cli features into a module, so you could vcpkg install cli or cli[with_remote].