add valhalla's time options #1

nilsnolde · 2023-08-06T18:47:43Z

Sorry, I got carried away a little bit. Seems I gave bit of a lousy review when adding OTP, there's been a few inconsistencies with other routers, which I patched here as well. Ideally I'd have separated that entirely in a different PR, sorry for a bit of a mess.. Maybe I should've also waited until we merged your PR, then PR valhalla separately. Anyways, let me know if you strongly prefer another way and you can just cherry-pick what you want from here for your PR and I'll deal with valhalla after we merged nilsnolde#116 for OTP & Google only.

I mainly wanted to make the review back and forth a bit easier and have smth tangible to show what I mean.Let me know smth looks weird or is not clear, I'm not 100% I didn't mess anything up.

nilsnolde · 2023-08-06T18:49:14Z

pyproject.toml

@@ -25,6 +25,7 @@ contextily = {version = "^1.1.0", optional = true}
 geopandas = {version = "^0.8.2", optional = true}
 descartes = {version = "^1.0.0", optional = true}
 pytz = "^2023.3"
+timezonefinder = "^6.2.0"


for valhalla, which (unfortunately) doesn't return a tz aware date_time string

routingpy/convert.py

nilsnolde · 2023-08-06T18:53:13Z

routingpy/routers/google.py

@@ -189,8 +187,8 @@ def directions(  # noqa: C901
        language: Optional[str] = None,
        region: Optional[str] = None,
        units: Optional[str] = None,
-        arrival_time: Optional[int] = None,
-        departure_time: Optional[int] = None,
+        date_time: Optional[datetime.datetime] = datetime.datetime.now(datetime.timezone.utc),


I made them all more consistent in terms of time arguments:

date_time is everywhere supposed to be a tz-aware date_time object, defaulting to UTC

date_time_type is either depart_at or arrive_by

I put those arguments wherever time is needed, but sometimes they're not doing anything, e.g. OTP isochrones only support depart_at scenarios

We could've done a enum or so, but I thought with a docstring it's ok, and it's one less thing to import for others

Great, I'd planned to tackle that later too! However, I would have preferred a "departure_datetime" and an "arrival_datetime". I find it more concise and more explicit for the user.

I'm open to that. As we're breaking stuff right now, we're flexible. And I kinda agree.

I'll change it here before touching the tests.

routingpy/routers/google.py

routingpy/routers/opentripplanner_v2.py

nilsnolde · 2023-08-06T18:57:57Z

routingpy/routers/opentripplanner_v2.py

@@ -135,13 +105,13 @@ def directions(
        query = f"""
            {{
                plan(
-                    date: "{ date.strftime("%Y-%m-%d") }"
-                    time: "{ time.strftime("%H:%M:%S") }"
+                    date: "{ date_time.date().strftime("%Y-%m-%d") }"


this should do the same as before, just omitting the timezone

Yes, we must validate this in the unit tests.

nilsnolde · 2023-08-06T19:01:35Z

routingpy/routers/opentripplanner_v2.py

-            arrival_datetime = self._timestamp_to_utc_datetime(itinerary["endTime"])
+            distance, geometry = OpenTripPlannerV2._parse_legs(itinerary["legs"])
+            departure_datetime = convert.timestamp_to_tz_datetime(itinerary["startTime"], "UTC")
+            arrival_datetime = convert.timestamp_to_tz_datetime(itinerary["endTime"], "UTC")


this I don't really understand @khamaileon : does that mean that in the OTP result there's no info which timezone the destination (or departure) is referring to? Is there no way to get the timezone info back from OTP? If not, I'd suggest to find the timezone similar to valhalla with timezonefinder, otherwise anyone using this has the burden to look up the timezone to get the local time.

I think we can consider the input timezone as the output timezone. That's why I insisted on a datetime aware input.

But how does OTP expect it? It expects a UTC date & time in the request right? Then these outputs here by OTP is UTC as well? If we can't find out the timezone, we shouldn't assume the same as the input, it might've changed from origin to destination. I couldn't find the docs for OTP request parameters though..

It expects a local date & time in the request. But we ask the user for an aware datetime so we can use the same timezone to transform the output. Maybe I'm wrong, but I find it hard to imagine a user requesting an itinerary on one timezone with a result on another timezone.

BTW, there is a way to retrieve the timezone for transit legs via agency/timezone.

I find it hard to imagine a user requesting an itinerary on one timezone with a result on another timezone.

That's actually quite common for car or long-distance transit (e.g. Amtrak crossing the US).

BTW, there is a way to retrieve the timezone for transit legs via agency/timezone.

I think that's not granular enough. We'd need the stops.txt/stop_timezone (if OTP exposes that) and I wouldn't be sure that's consistently handled by all operators who indeed have stops in different timezones.

FWIW, I asked a OTP core dev offline (Cc @leonardehrenfried), maybe he has some more insight/recommendations.

Haha he sent me your matrix questions and his answers. So it's the same situation for OTP as we have in Valhalla. It's a rare situation that one operator has timezone-crossing stops, but still it's common to cross timezone in one itinerary. I'd use the same approach for now as in Valhalla and look it up dynamically by its location with timezonefinder. Apparently they want to change it in OTP similar to what I'd like in Valhalla, so this would hopefully only be a temp solution. Ok for you?

That's actually quite common for car or long-distance transit (e.g. Amtrak crossing the US).

Yeah, I see. I think I was too focused on public transport where you only have one timezone per network.

I think that's not granular enough. We'd need the stops.txt/stop_timezone (if OTP exposes that) and I wouldn't be sure that's consistently handled by all operators who indeed have stops in different timezones.

OTP exposes timezone a the stop level, but maybe that's too much information to bring up.

FWIW, I asked a OTP core dev offline (Cc @leonardehrenfried), maybe he has some more insight/recommendations.

Lol I had asked the question on the OTP chat and it was him who answered me!

Yes I'm ok with it! Don't forget timezonefinder has a cost (near 1s / 10000 query)

Come to think of it, now that we won't be using the input timezone for output, do we still need an input datetime aware? And what happens if we enter a departure timezone that isn't the one at the point of origin? Sorry, I can make the changes if we rollback again.

yeah right, it’s a bit of cost, but I guess it’s drowning well in the OTP/Valhalla response time:)

God I hate datetime stuff! It always the head quite a bit.. I’m not sure about OTP‘s isochrones, but I’m now fairly sure that expects a local time as well (timezone would be meaningless). If that’s the case, all routers only need the local time and we can ditch tz aware input? The output is probably the only thing that should have timezone awareness. Google does it right, OTP & Valhalla hopefully soon too.

routingpy/routers/opentripplanner_v2.py

nilsnolde · 2023-08-06T19:03:27Z

routingpy/routers/opentripplanner_v2.py

+        profile: str,
+        intervals: List[int],
+        interval_type: Optional[str] = None,


same here: I changed those to be consistent with other routers

Yes, I was planning to do that in the 2nd phase. I wanted the 1st version to reflect the router's API as closely as possible.

Ah ok, that'd have been fine too, as long as we don't release, which we have already. It's not that bad though, the next release will break the date_time stuff anyways, so we'll need to increment major version.

nilsnolde · 2023-08-06T19:35:30Z

routingpy/routers/valhalla.py

+        # TODO: maybe we can get rid of that in the future, in case we'll have timezone aware strings in the response:
+        #   https://github.com/valhalla/valhalla/issues/4241
+        if origin.get("date_time"):
+            departure_time = timestamp_to_tz_datetime(
+                origin["date_time"], lonlat_to_timezone(origin["lon"], origin["lat"])
+            )


valhalla doesn't return the timezone either. it could do that in the date_time string and I think we should. it won't change any public facing API if we change it in the future, so we can start with this

Same as above. I'd take the input timezone.

Same comment as above:) destination can have a different timezone

nilsnolde · 2023-08-06T19:43:16Z

And last question: do you want to be invited as a maintainer to routingpy? No pressure, I don't mind either way. It just makes collaboration a bit easier, at the least the Github goodies like assigning reviewers (not possible across forks).

nilsnolde · 2023-08-06T19:58:38Z

And I should mention this is missing unit tests for valhalla (I didn't do changes to OTP/Google tests either yet).

khamaileon · 2023-08-07T08:12:37Z

routingpy/routers/google.py

@@ -189,8 +187,8 @@ def directions(  # noqa: C901
        language: Optional[str] = None,
        region: Optional[str] = None,
        units: Optional[str] = None,
-        arrival_time: Optional[int] = None,
-        departure_time: Optional[int] = None,
+        date_time: Optional[datetime.datetime] = datetime.datetime.now(datetime.timezone.utc),


Great, I'd planned to tackle that later too! However, I would have preferred a "departure_datetime" and an "arrival_datetime". I find it more concise and more explicit for the user.

khamaileon · 2023-08-07T08:16:03Z

routingpy/routers/google.py

@@ -378,7 +349,7 @@ def _parse_direction_json(self, response, alternatives):

        directions = []
        for route in response["routes"]:
-            duration, distance, geometry, departure_datetime, arrival_datetime = self._parse_legs(
+            duration, distance, geometry, departure_datetime, arrival_datetime = Google.parse_legs(


Maybe it's just me, but it seems a bit odd. But I can live with it and we stay consistent with the rest of the code, which is the most important thing.

khamaileon · 2023-08-07T08:16:56Z

routingpy/routers/google.py

-        arrival_time: Optional[int] = None,
-        departure_time: Optional[int] = None,
+        date_time: Optional[datetime.datetime] = datetime.datetime.now(datetime.timezone.utc),
+        date_time_type: Optional[str] = None,


Same as above.

khamaileon · 2023-08-07T08:19:10Z

routingpy/routers/opentripplanner_v2.py

-        time: Optional[datetime.time] = datetime.datetime.now().time(),
-        arrive_by: Optional[bool] = False,
+        profile: Optional[str],
+        date_time: Optional[datetime.datetime] = datetime.datetime.now(datetime.timezone.utc),


Same as above.

khamaileon · 2023-08-07T08:21:16Z

routingpy/routers/opentripplanner_v2.py

@@ -135,13 +105,13 @@ def directions(
        query = f"""
            {{
                plan(
-                    date: "{ date.strftime("%Y-%m-%d") }"
-                    time: "{ time.strftime("%H:%M:%S") }"
+                    date: "{ date_time.date().strftime("%Y-%m-%d") }"


Yes, we must validate this in the unit tests.

khamaileon · 2023-08-07T08:24:33Z

routingpy/routers/opentripplanner_v2.py

-            arrival_datetime = self._timestamp_to_utc_datetime(itinerary["endTime"])
+            distance, geometry = OpenTripPlannerV2._parse_legs(itinerary["legs"])
+            departure_datetime = convert.timestamp_to_tz_datetime(itinerary["startTime"], "UTC")
+            arrival_datetime = convert.timestamp_to_tz_datetime(itinerary["endTime"], "UTC")


I think we can consider the input timezone as the output timezone. That's why I insisted on a datetime aware input.

khamaileon · 2023-08-07T08:27:10Z

routingpy/routers/opentripplanner_v2.py

+        profile: str,
+        intervals: List[int],
+        interval_type: Optional[str] = None,


Yes, I was planning to do that in the 2nd phase. I wanted the 1st version to reflect the router's API as closely as possible.

routingpy/routers/opentripplanner_v2.py

khamaileon · 2023-08-07T08:35:45Z

routingpy/routers/valhalla.py

+        # TODO: maybe we can get rid of that in the future, in case we'll have timezone aware strings in the response:
+        #   https://github.com/valhalla/valhalla/issues/4241
+        if origin.get("date_time"):
+            departure_time = timestamp_to_tz_datetime(
+                origin["date_time"], lonlat_to_timezone(origin["lon"], origin["lat"])
+            )


Same as above. I'd take the input timezone.

khamaileon · 2023-08-07T08:43:48Z

I'm fine with doing it this way. Apart from the comments I've already made, the tests on Valhalla are missing and the results of the dates also need to be tested. Tell me how you want us to proceed, I should have 1 or 2 hours available today. And I'd be delighted to become a contributor. I had already done a code of this type privately a few years ago. It's unmaintainable on its own. Long live FOSS!

nilsnolde · 2023-08-07T10:12:58Z

Ok cool I'll add you to the repo then, thanks!

I'd do the changes to this PR:

add tests for valhalla
adapt tests for google/otp
move to arrival_datetime & destination_datetime, both being mutually exclusive

Just needs clarification on the OTP /route endpoint and how it handles timezones. Is there a sane doc for OTP API parameters? I struggled to find one.

Sounds ok for you? Next collab will happen with a bit more thought on my side:)

khamaileon · 2023-08-07T10:25:42Z

Ok cool I'll add you to the repo then, thanks!

I'd do the changes to this PR:
* add tests for valhalla

* adapt tests for google/otp

* move to `arrival_datetime` & `destination_datetime`, both being mutually exclusive

Great, let me know if you need help.

Just needs clarification on the OTP /route endpoint and how it handles timezones. Is there a sane doc for OTP API parameters? I struggled to find one.

You can look at the GraphQL doc on a local instance: http://localhost:8081/graphiql

Sounds ok for you? Next collab will happen with a bit more thought on my side:)

No worries, it's nice to share it with you already :)

Co-authored-by: Nils <[email protected]>

nilsnolde · 2023-12-15T21:22:27Z

sorry for the long wait. I finally found the time to take care of this. let me know if it's fine for you.

khamaileon · 2024-01-08T15:37:01Z

Hi @nilsnolde, happy new year :)
Sorry end of the year was intensive for me too. I'll have a look ASAP!

leonardehrenfried · 2024-02-21T16:10:05Z

BTW, times which include the offset (~= time zone) will be coming to OTP: opentripplanner/OpenTripPlanner#5660

nilsnolde · 2024-02-21T17:04:50Z

Great:) forgot to update that we also added it a few weeks ago: valhalla/valhalla#4491. returning both tz string and offset

nilsnolde commented Aug 6, 2023

View reviewed changes

routingpy/convert.py Show resolved Hide resolved

nilsnolde commented Aug 6, 2023

View reviewed changes

routingpy/routers/google.py Outdated Show resolved Hide resolved

nilsnolde commented Aug 6, 2023

View reviewed changes

routingpy/routers/opentripplanner_v2.py Show resolved Hide resolved

nilsnolde commented Aug 6, 2023

View reviewed changes

routingpy/routers/opentripplanner_v2.py Outdated Show resolved Hide resolved

nilsnolde commented Aug 6, 2023

View reviewed changes

nilsnolde mentioned this pull request Aug 6, 2023

Add departure and arrival datetime to Direction nilsnolde/routingpy#116

Open

khamaileon reviewed Aug 7, 2023

View reviewed changes

sadikkuzu and others added 13 commits September 18, 2023 01:27

Update pre-commit hooks (#118)

e5eb103

Co-authored-by: Nils <[email protected]>

fix: preference parameter for isochrones and matrix with Valhalla (#121)

2f0594b

handle ZERO_RESULTS in google matrix (#122)

230da12

Update CHANGELOG.md

da93ea1

Add departure and arrival datetime to Direction

8f5a213

Naive version

3659d53

Improve code

7520276

Rollback to timezone aware version

b3ccdbd

revert making google matrix parser non-static and 'protected'

6efc3e6

break out the time functions into utils.py

9fe6434

add valhalla plus a ton of other edits.. sorry for that..

d0df5cc

few corrections

5d5818b

remove a whole function

ed5f64a

nilsnolde added 6 commits December 15, 2023 20:17

otp isochrones still were using arrive_by arg

08e6c9c

todo for future

ce92629

dep/arr time for valhalla

1cd4699

dep/arr time for google

56df7f2

dep/arr time for otp

7900898

fix tests and other stuff

78c2549

nilsnolde force-pushed the nn-add-valhalla-time branch from b04027f to 78c2549 Compare December 15, 2023 21:13

add valhalla's time options #1

Are you sure you want to change the base?

add valhalla's time options #1

Conversation

nilsnolde commented Aug 6, 2023

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

khamaileon Aug 7, 2023 • edited Loading

Choose a reason for hiding this comment

nilsnolde Aug 7, 2023 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

khamaileon Aug 7, 2023 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

nilsnolde commented Aug 6, 2023 • edited Loading

nilsnolde commented Aug 6, 2023

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

khamaileon commented Aug 7, 2023

nilsnolde commented Aug 7, 2023

khamaileon commented Aug 7, 2023

nilsnolde commented Dec 15, 2023

khamaileon commented Jan 8, 2024

leonardehrenfried commented Feb 21, 2024

nilsnolde commented Feb 21, 2024

khamaileon Aug 7, 2023 •

edited

Loading

nilsnolde Aug 7, 2023 •

edited

Loading

khamaileon Aug 7, 2023 •

edited

Loading

nilsnolde commented Aug 6, 2023 •

edited

Loading