Build vroom as library

[frodrigo]

It will be interesting to build vroom as library containing only the solver and have the main, tsplib and osrm loaders in a command linked with the lib.

[jcoupey]

I can see how useful this could be for integration in other pieces of software. At the moment, a loader could build a problem from a string (or file name) and solving would return a string (or a rapidjson object).

I think a proper library would require to expose more than just this. For example a problem object that could be directly populated, and also an object to represent a solution and use it directly after solving.

[frodrigo]

The idea is to use it as a native library in an Android application (@fijemax)

[jcoupey]

This sounds cool. But how do you plan to re-use the solution output, just parse it again in the client application?

[frodrigo]

The idea with a library if to share objects not exchanging and parsing files.

[jcoupey]

to share objects

Precisely what I mean: we can't do this right now. At the moment, there is no "solution" object that could be shared, only code reaching out all over the place to write the expected json to output.

I think this should be considered in the light of #44 (I just filled this issue to write down things I had in mind before).

[jcoupey]

This is probably getting nearer with the refactor for #44, as involved objects (locations, jobs, vehicles, routes, input problem, solution etc.) now have a proper data representation.

[sashakh]

Hi,

I'm using vroom as library, adding such simple makefile patch:

diff --git a/src/makefile b/src/makefile
index c729fcf..0ff069f 100644
--- a/src/makefile
+++ b/src/makefile
@@ -10,6 +10,7 @@ LDLIBS = -lboost_system -lboost_regex -lboost_log -lboost_log_setup -lpthread -l
 
 # Using all cpp files in current directory.
 MAIN = ../bin/vroom
+LIB = ../lib/libvroom.a
 SRC = $(wildcard *.cpp)\
                        $(wildcard ./algorithms/*.cpp)\
                        $(wildcard ./routing/*.cpp)\
@@ -34,11 +35,14 @@ endif
 OBJ = $(SRC:.cpp=.o)
 
 # Main target.
-all : $(MAIN)
+all : $(MAIN) $(LIB)
 
 $(MAIN) : $(OBJ) main.o
        $(CXX) $(CXXFLAGS) -o $@ $^ $(LDLIBS)
 
+$(LIB) : $(OBJ)
+       $(AR) cr $@ $^
+
 # Building .o files.
 %.o : %.cpp %.h
        $(CXX) $(CXXFLAGS) -c $< -o $@

And my code then looks like:

        input input_data(NULL, false);  // geometry = false

        input_data.add_vehicle(0, optional_coords_t({ p[0].pos->lon,
                                                      p[0].pos->lat}),
                               boost::none);

        unsigned i;
        for (i = 1; i < size; i++) {
                input_data.add_job(i, optional_coords_t({p[i].pos->lon,
                                                         p[i].pos->lat}));
        }

        input_data._max_cost_per_line.assign(size, 0);
        input_data._max_cost_per_column.assign(size, 0);
        input_data._matrix = get_matrix(osrm, input_data._locations,
                                        input_data._max_cost_per_line,
                                        input_data._max_cost_per_column);

        solution sol = input_data.solve(1);       // nb_threads = 1

        if (sol.code != 0) {
                err("sol.code %d: %s\n", sol.code, sol.error.c_str());
                return;
        }

        for (const auto & r:sol.routes) {
                unsigned i = 0; 
                for (const auto & s:r.steps) {
                        //const location_t &l = s.location;
                        //dbg("%lu: (%f,%f)\n", s.job, l.lon.get(), l.lat.get());

                        n[i] = s.job;
                        i++;
                }
                n[0] = 0;
        }

[sashakh]

However #69 will create certain problems with continuous runs.

[jcoupey]

@sashakh thanks for sharing, glad to know this works fine for you. The add_job and add_vehicle functions where added with that exact usage in mind.

Ideally you should not have to worry about internal members like _max_cost_per_[line|column] and calling get_matrix yourself. You can avoid that part by passing your osrm object as first argument in the input ctor (instead of NULL). In the current code-base, this object is constructed here:

vroom/src/utils/input_parser.cpp

Lines 31 to 47 in d75bf82

	std::unique_ptr<routing_io<cost_t>> routing_wrapper;
	if (!cl_args.use_libosrm) {
	// Use osrm-routed.
	routing_wrapper = std::make_unique<routed_wrapper>(cl_args.osrm_address,
	cl_args.osrm_port,
	cl_args.osrm_profile);
	} else {
	#if LIBOSRM
	// Use libosrm.
	if (cl_args.osrm_profile.empty()) {
	throw custom_exception("-l flag requires -m.");
	}
	routing_wrapper = std::make_unique<libosrm_wrapper>(cl_args.osrm_profile);
	#else
	throw custom_exception("libosrm must be installed to use -l.");
	#endif
	}

depending on whether we use libosrm or osrm-routed.

Then calling solve on an input object will automatically take care of setting up the matrix via set_matrix.

[sashakh]

Hi Julien, On Nov 1, 2017 11:18, "Julien Coupey" <notifications@github.com> wrote: @sashakh <https://github.com/sashakh> thanks for sharing, glad to know this works fine for you. The add_job and add_vehicle functions where added with that exact usage in mind. Thanks for the great piece of code. ) Ideally you should not have to worry about internal members like _max_cost_per_[line|column] and calling get_matrix yourself. You can avoid that part by passing your osrm object as first argument in the input ctor (instead of NULL). In the current code-base, this object is constructed here: https://github.com/VROOM-Project/vroom/blob/d75bf82a27b999f025d1fa7492a2fc d126f0fbbd/src/utils/input_parser.cpp#L31-L47 depending on whether we use libosrm or osrm-routed. Absolutely correct. It was the way how I started to use vroom. In my project I'm already using OSRM for other purposes and in this example get_matrix() is my custom function. In generic examle those lines are not needed, but instead osrm_wrapper can be initialized and passed to input() constructor.

[jcoupey]

@sashakh could you submit your makefile patch as a pull request when you find time ? I'd be happy to merge it in. The support for library use is still quite experimental but it does seem relevant for several use-cases.

[sashakh]

Will do. Do you wish to keep the library in lib/ directory (then we will need to update build instuction or add additional makefile rule to create bin/ and lib/ directories, the last one is even simpler)? Sasha On Nov 1, 2017 15:22, "Julien Coupey" <notifications@github.com> wrote: @sashakh <https://github.com/sashakh> could you submit your makefile patch as a pull request when you find time ? I'd be happy to merge it in. The support for library use is still quite experimental but it does seem relevant for several use-cases. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#42 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABCIjDwQiyg5C70-ukjaKmpg8IxR6iiSks5syHChgaJpZM4J5sAu> .

[jcoupey]

@sashakh creating the bin and lib directories from makefile rules is probably the more convenient option. Thanks!

[sashakh]

Ok.

…

On Nov 2, 2017 09:59, "Julien Coupey" ***@***.***> wrote: @sashakh <https://github.com/sashakh> creating the bin and lib directories from makefile rules is probably the more convenient option. Thanks! — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#42 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABCIjHEc5xeO98Z6bja1kt66LAP9Sxe8ks5syXZqgaJpZM4J5sAu> .

[sashakh]

Hi Julien. I've created pull request: #71 Thanks, Sasha

…

On Thu, Nov 2, 2017 at 10:00 AM, Sasha Khapyorsky ***@***.***> wrote: Ok. On Nov 2, 2017 09:59, "Julien Coupey" ***@***.***> wrote: > @sashakh <https://github.com/sashakh> creating the bin and lib > directories from makefile rules is probably the more convenient option. > Thanks! > > — > You are receiving this because you were mentioned. > Reply to this email directly, view it on GitHub > <#42 (comment)>, > or mute the thread > <https://github.com/notifications/unsubscribe-auth/ABCIjHEc5xeO98Z6bja1kt66LAP9Sxe8ks5syXZqgaJpZM4J5sAu> > . >

[jcoupey]

FYI: next step toward this landed with #84, introducing a input::set_matrix(matrix<cost_t>&&) function that is used throughout the codebase (_matrix member in input is now private).

See updated libvroom.cpp example for usage from a C++ context.

[jcoupey]

I flagged this as an experimental feature in the v1.2.0 release. It should be fully working. Yet I think we should keep that ticket open for reference as the C++ API might still evolve. Also there is no proper documentation besides the examples in libvroom_examples/libvroom.cpp.

[dbhoot]

I second this request. I think it'd be good to have javascript and or python wrappers so that the vroom solver can be invoked from other high level languages

[jcoupey]

@dbhoot thanks for your interest. This ticket was primarily about compiling a C++ library, which is now effective, and I've had a couple reports from people using this successfully in their own C++ code. The reason I did not close here is that I still consider it as experimental (read "I don't want to be bound to the C++ API, and any internal change will not be considered as a breaking change"). I try to keep the example file up to date though.

I don't have a need myself for bindings to other languages, but I definitely see how that might be useful to others. If you're interested in setting something up, then feel free to open a dedicated ticket here to discuss the best way to proceed.

For the sake of completeness: you might want to check out vroom-express, an expressjs-based wrapper to use VROOM with HTTP POST requests.

[sashakh]

Yet another report. I'm using libvroom constantly and successfully. ) Sometimes (rear) need to upgrade API - not a big deal. )

…

On Tue, Aug 20, 2019, 14:20 Julien Coupey ***@***.***> wrote: @dbhoot <https://github.com/dbhoot> thanks for your interest. This ticket was primarily about compiling a C++ library, which is now effective, and I've had a couple reports from people using this successfully in their own C++ code. The reason I did not close here is that I still consider it as experimental (read "I don't want to be bound to the C++ API, and any internal change will not be considered as a breaking change"). I try to keep the example file <https://github.com/VROOM-Project/vroom/blob/master/libvroom_examples/libvroom.cpp> up to date though. I don't have a need myself for bindings to other languages, but I definitely see how that might be useful to others. If you're interested in setting something up, then feel free to open a dedicated ticket here to discuss the best way to proceed. For the sake of completeness: you might want to check out vroom-express <https://github.com/VROOM-Project/vroom-express>, an expressjs-based wrapper to use VROOM with HTTP POST requests. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#42?email_source=notifications&email_token=AAIIRDDTSUIUTWMIIKUJFE3QFPHPXA5CNFSM4CPGYAXKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOD4V57GI#issuecomment-522968985>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAIIRDAP7CWEMQKBHFDCE73QFPHPXANCNFSM4CPGYAXA> .

[dbhoot]

For the sake of completeness: you might want to check out vroom-express, an expressjs-based wrapper to use VROOM with HTTP POST requests.

I saw the express project. If I remember correctly, it basically shells out to invoke the binary.

The PR sashakh made is great for c++ projs. I didn't try creating v8 bindings using swig or something based on his work but I'm sure it's possible with some work.

My point in commenting above is that I generally think it makes the project more useful.

[dndll]

I struggled a little bit trying to get libvroom to run with Rust, gave up and did a shell command.

Did the same thing in Kotlin too as it was difficult/not enough time to build JNI bindings for it

That said, it would be mega useful if it was easier to utilise rather than the above methods as command is expensive in the JVM

[jcoupey]

@awesomeibex not sure exactly what you're trying to achieve here as the purpose of libvroom is to call the native functions from C++.

If you're interested in using it from your C++ code, then you'll find an example on how to link the library and how to setup and solve a problem.

If you're interested in bindings for other languages, then please check my previous comment above that still applies to date.

[yhilem]

Hi,
I'm trying to use the vroom lib from Java. I thought about using JavaCPP (https://github.com/bytedeco/javacpp) and if possible create a module in JavaCPP Presets (https://github.com/bytedeco/javacpp-presets). I am not a C ++ expert. I would need some help creating the cppbuild.sh file (https://github.com/bytedeco/javacpp-presets/wiki/Create-New-Presets#the-cppbuildsh-file).
Thanks in advance.

[jcoupey]

I'm not familiar at all with JavaCPP (and not that much with Java either), so I can't really comment on the whole process. You'll probably get better generic advise from the JavaCPP maintainers.

On the specific question of the cppbuild.sh script: the example you're pointing to seems to get down to building using make. So if you don't need all the fancy multi-platform stuff, you should be able to simply paste the usual build instructions for vroom in your script, including pulling the repo and installing dependencies.

Interested to have any feedback if you end up using the project from Java.

[yhilem]

Thank you for your reply. We are already using vroom from java with the ProcessBuilder class (https://docs.oracle.com/javase/9/docs/api/java/lang/ProcessBuilder.html). Same as what vroom-express does.
It is not scalable for our new use cases: between 20,000 to 40,000 requests, in real time, for optimizing routes in the morning between 8 a.m. and 10 a.m.
We want first reduce process startup overhead.
In addition, we want to deploy vroom on our agents' android smartphones with a cost matrix that corresponds to the geographic coverage area of our sites.
I will share the results of our work as soon as it is done.

[kkarbowiak]

I flagged this as an experimental feature in the v1.2.0 release. It should be fully working. Yet I think we should keep that ticket open for reference as the C++ API might still evolve. Also there is no proper documentation besides the examples in libvroom_examples/libvroom.cpp.

I think the library part of this project is a fact and already a number of people depend on it (myself included). I would therefore propose to close this issue and handle any outstanding topics as new issues.

Any thoughts?

[jcoupey]

I think the library part of this project is a fact

Yes you're right of course. Again my only concern here is that I don't want to be tied by the C++ API and have to use work-arounds to make it evolve in a non-breaking way.

Maybe we could advertise the C++ API but add some warnings stating that the non-breaking semantic versioning approach only applies to the json API?

[kkarbowiak]

Maybe we could advertise the C++ API but add some warnings stating that the non-breaking semantic versioning approach only applies to the json API?

Works for me.

[dbhoot]

👍