I like your roadmaps, it looks nice.
first pair of functions : I think less dependencies could improve compatibility and reduce the difficulty of future developments. But maybe use Boost::Geometry instead of CGAL will change a lot, which means a lot of work to do. In my opinion, it is not urgent for now.
pgr_trsp rewrite : I don’t know this functionality much. If many users use it and it has many problems for now, I think maybe it is an important job for us developers.
···
2018-04-11 7:51 GMT+08:00 Cayetano Benavent <cayetano.benavent@geographica.gs>:
Thank you Vicky, and sorry for my late response. I’m sure this roadmap draws a good and solid future for pgRouting. I would like to add some thoughts from my own experience:
- CGAL dependent functions: I’m agree with you that pgr_alphaShape and pgr_pointsAsPolygon should be inside PostGIS as they are pure geometric functions and CGAL is a too big dependence. However both functions works well and they are so usefull to build geometries for catchment areas.
- Personal thoughts about TRSP and Withpoints functions:
- TRSP: the use of turn restrictions is critical to build real world vehicle routing applications, so I think this is a very important function. Putting the focus on this family of functions is a success due to the high level of interest it has (and will have much more). But I think there is an important weakness with TRSP: if you want to use with arbitrary points (edge fractions in this case, but without road side) you need to use old implementation that it is not very solid (performance is bad and there are some bugs). In the future, when TRSP function will be more estable, it could be interesting too to have a TRSP cost matrix function to compute big cost matrices.
- Withpoints: When you develop a vehicle routing application you almost always get input data (source and target) as points with road side. For this reason this family of functions is so important too. It works well, but it has an importante weakness: there is no way to take control of turn restrictions and this is a big problem working with vehicles (I’m very worried about u-turns). The perfect shortest path function when you are working with vehicles would be the current Withpoint function capable of using a restrictions table.
Related to osm2pgrouting it would be interesting to continue with the experiments with Libosmium to build restrictions tables form OSM data. With real data we could test and debug in a better way TRSP functions.
Finally I would like to take this opportunity to announce you that it has been released osm2pgrouting version 2.3.4 last week and the next Milestone (2.3.5) starts his way:
https://github.com/pgRouting/osm2pgrouting/releases
Thanks,
Cayetano
pgrouting-dev mailing list
pgrouting-dev@lists.osgeo.org
https://lists.osgeo.org/mailman/listinfo/pgrouting-dev
Cayetano Benavent
Head of Data
+34 954 095 876
www.geographica.gs
2018-04-10 19:31 GMT+02:00 Vicky Vergara <vicky@georepublic.de>:
Hello all,
Thanks all for your feedback.
Its difficult to know how many users are using the proposed or experimental functions.
Moving up one level the functions implies bug fixing and documentation enhancement. (we could even move down functions?)
“Priorities” is a subjective term: what is a priority to me might not be a priority to you, but in general:
bug fixing (no matter where in the classification, (official, proposed, experimental) the function is, and move them up a level as we see fit.
TRSP rewrite
What works don’t fix (aka, pgr_alphaShape + pgr_pointsAsPolygons)
All your comments lead me to think that road map #2 with a twist:
Next version to be 3.0, but the alpha version on march 2019 with feature freeze, and make the release on August/September 2019
Reasons:
-
Documentation can be improved between march and August/September
-
Give some time for osm2pgRouting & pgRoutingLayers to work for 3.0
-
Give time for the alpha to be tested and do bug fixes
-
Because pgRouting is FOSS, “normally” we work on it on our free time, so making the alpha on march will give us more time to do better work on making the 3.0 happen
I would make the develop branch to be 3.0.0dev version this weekend, (unless another suggestion is made)
Vicky
pgrouting-dev mailing list
pgrouting-dev@lists.osgeo.org
https://lists.osgeo.org/mailman/listinfo/pgrouting-dev
On Tue, Apr 10, 2018 at 1:49 AM, Rohith Reddy <rohithreddy2219@gmail.com> wrote:
Hello ,
The road map looks great !!! I really like the idea of moving some experimental functions to proposed functions in both the road maps. I believe that this will continue to be in all the future road maps, as the experimental section keep accumulating with functions every year by the addition of new functionality by the GSoC students. So the question now is how can we efficiently convert these into proposed functions. I would suggest that moving the experimental functions to the proposed functions can be carried out based on a priority which can be the following
-
Number of non-pgRouting users using the experimental function
-
Number of issues related to the experimental function
-
Dependency of a stable function on the experimental function for optimization.
Based on the priorities, we could start testing the experimental functions, find bugs and solve them. I feel that this would be an efficient way of enriching the pgRouting functionality and usage in the upcoming releases.
I tried to use the contraction function (developed by me) an experimental function, in one of my research projects and then realised that there are bugs in contraction function. I had opened issues [1], [2], [3], [4] for the same. I am interested in working on moving the contraction function to a stable function. I am currently working on solving the issues and soon will have the functionality well tested.
I am also excited to see the rewrite of pgr_trsp with a dijkstra implementation. As Anthony mentioned his interest in using osm2pgrouting to extract turn data, I remember having a discussion on a rewrite of osm2pgrouting using osmosis. The functionality of extracting turn data can be considered in the rewrite of osm2pgrouting, so that it can easily be used with pgr_trsp rewrite.
Regards,
Rohith Reddy.
[1] https://github.com/pgRouting/pgrouting/issues/1003
[2] https://github.com/pgRouting/pgrouting/issues/1004
[3] https://github.com/pgRouting/pgrouting/issues/1005
[4] https://github.com/pgRouting/pgrouting/issues/1006
pgrouting-dev mailing list
pgrouting-dev@lists.osgeo.org
https://lists.osgeo.org/mailman/listinfo/pgrouting-dev
–
Georepublic UG (haftungsbeschränkt)
Salzmannstraße 44,
81739 München, Germany
Vicky Vergara
Operations Research
eMail: vicky@[georepublic.de](http://georepublic.de)
Web: [https://georepublic.info](https://georepublic.info)
Tel: +49 (089) 4161 7698-1
Fax: +49 (089) 4161 7698-9
Commercial register: Amtsgericht München, HRB 181428
CEO: Daniel Kastl
On Tue, Apr 10, 2018 at 3:47 AM, Aditya Pratap Singh <adityapratap.singh28@gmail.com> wrote:
Hello,
pgrouting-dev mailing list
pgrouting-dev@lists.osgeo.org
https://lists.osgeo.org/mailman/listinfo/pgrouting-dev
Thanks
Aditya Pratap Singh
On Tue, Apr 3, 2018 at 9:06 PM, Vicky Vergara <vicky@georepublic.de> wrote:
Hello fellow developers,
As you know we are rewriting pgRouting, since v2.1, to remove all “bugs”,
like the non standard way of constructing a graph, and using more C++
features.
The rewrite is “almost done”:
First pair of functions
- pgr_alphaShape - Alpha shape computation
- pgr_pointsAsPolygon - Polygon around a set of points
Both functions are tightly dependant, and I believe that they should be a
postGIS function as is a geometry function, and the closest one that
resembles this pair of functions is ST_ConcaveHull [1] Some time ago I
compared both functions on [2]
I went to see how to add that function to postGIS, and for the moment that
is not possible, it sould be done first in GEOS, and GEOS is so big, right
now I dont feel is the time to do that, but rewriting to use boost:Geometry
would eliminate the need of CGAL, so one less pre-requisite to build
pgRouting would be needed. or they can stay as they are now.
The other function is pgr_trsp
Which has so many problems at postgres level [4], Last year on GSoC,
Vidham tried to do a rewrite, confirming that the problem is much harder
than expected for the 3 month program. But as far as I know many people use
them because it has the “points” version.
The C++ code is not well designed [5], just a look at that, my_dijkstra
calls my_dijkstra and that calls my_dijkstra, it still uses pointers, so
great chance of memory leaks, etc.
Basically is Use at your own risk
Beside those functions mentioned above, now we are in the moment where
there are more proposed functions [7] than official functions. and a lot
of deprecated functions that have to be maintained and tested because of
backwards compatibility.
** So, here is my road map #1 **
Version 2.7 to be released on September 2018 where fix bug of
pgr_withPoints [8] is a must have, and the possibilities of what other
things it could have:
- New functionality done by GSoC students on “experimental section”
- Write a substitution (with a different name maybe pgr_dijkstraTR) for
pgr_trsp (one vertex to one vertex) based on dijkstra
- Rewrite pgr_alphaShape to use boost:graph instead of CGAL
- move some experimental functions up to proposed
Version 3.0 I would like it to be on September 2019
Where a complete cleanup of the deprecated functions would be done,
Move proposed functions to official pgRouting functions.
** So, here is my road map #2 **
Version 3.0 to be released on March 2019 where fix bug of pgr_withPoints
[8] and Complete cleanup of the deprecated functions are a must have, and
the possibilities of what other things it could have:
- New functionality done by GSoC students on “experimental section”
- Write a substitution (with a different name maybe pgr_dijkstraTR) for
pgr_trsp (one vertex to one vertex) based on dijkstra
- Rewrite pgr_alphaShape to use boost:graph instead of CGAL
- move some experimental functions up to proposed
- Move proposed functions to official pgRouting functions.
Probably this second road map will make the second version of the
pgRouting book a little easier to write, (please Robe, comment on this)
We would like comments from the community about both road maps and if
possible to test proposed functions and experimental functions to help
decide which proposed functions can go up one level. (don’t forget to open
issues)
Regards
pgRouting team
[1] https://postgis.net/docs/ST_ConcaveHull.html
[2] https://github.com/cvvergara/pgrouting/issues/57
[3] https://github.com/pgRouting/pgrouting/blob/master/sql/
alpha_shape/alpha_shape.sql#L62
[4] https://github.com/pgRouting/pgrouting/tree/master/doc/trsp
[5] https://github.com/pgRouting/pgrouting/blob/master/include/trsp/GraphDefinition.h#L94
[6] https://github.com/pgRouting/pgrouting/tree/master/sql/pickDeliver
[7] http://docs.pgrouting.org/latest/en/proposed.html
[8] https://github.com/pgRouting/pgrouting/issues/760
–
Georepublic UG (haftungsbeschränkt)
Salzmannstraße 44,
81739 München, Germany
Vicky Vergara
Operations Research
eMail: vicky@georepublic.de
Web: https://georepublic.info
Tel: +49 (089) 4161 7698-1
Fax: +49 (089) 4161 7698-9
Commercial register: Amtsgericht München, HRB 181428
CEO: Daniel Kastl
When I was reading documentation I found a bug in pgr_withPointsDD and I created an issue ( https://github.com/pgRouting/pgrouting/issues/979 ) which was finally resolved but as you pointed this function still have some other bugs, So it should be our first prior to fix bugs of pgr_withPoints function.
I think it is right time to take some proposed function to official function as there are very less official functions and some experiment function to proposed function.
Seriously telling I don’t know the functionality of pgr_alphaShape but as Daniel said that there are very few non-pgRouting users, that find the AlphaShape function useful. So according to me it should be our least prior but if possible and time permit we can implement using boost in #2 road map.
Another important function is pgr_trsp. I know the functionality of this function and as you have pointed this function is not well designed. So we will fix bugs of this function ASAP as many users are using this functions. So, I’ll also go with road map #2 as it looks more stable to me. I am going to learn lot of things in release of version 3.0 of pgRouting.