Support PROJ >= 6.0.0 #922
Comments
|
I agree that this flag is required now (as PROJ 6 was released on 2019-03-01). Note that PROJ 7 release next year will drop proj_api.h completely. |
|
Do we use proj directly or only via libosmium? |
|
Ah, it seems through libosmium only. |
|
libosmium's associated ticket: osmcode/libosmium#277 |
|
You can close this ticket, but I bet this will be an FAQ. |
|
related PROJ 7.0 discussion happening at osmcode/libosmium#295 |
|
Reopening this. We are going to need osmium-independent proj >= 6 support. |
|
Thanks for reopening. I was going to grumble that I produce some paper maps projected to UTM zone 18 and therefore maintain a small subset of OSM in PostGIS in that projection, so losing support for everything but plate carree and spherical Mercator would be a problem. |
|
#1066 reorders the code so that it becomes fairly easy to implement a generic projection using the proj6 variant and then have cmake detect which one to use. Just add a PRs would be most welcome. |
joto
added a commit
to joto/osm2pgsql
that referenced
this issue
Nov 24, 2020
This commit is adding projection support using the PROJ library from 6.1 onwards. There might be an issue with multithreading, but the old code most likely already had that issue. This needs to be checked. The CMake config will prefer using the old API, simply because I don't know how to check for the new API without it trying to use the new API with version 5-6.0 which don't have everything we need. The code will only work from 6.1 onwards. There is a new CMake cache variable USE_PROJ_LIB which can be set to: * "4": Use version 4 API. Fail CMake if it is not available. * "6": Use version 6 API. Fail CMake if it is not available. * "off": Disable PROJ support (resulting binary will only have WGS84 (4326) and Web Mercator (3857) support) * "auto": Use version 4 API if available, otherwise try version 6 API or fall back to no PROJ support. See osm2pgsql-dev#922
joto
added a commit
to joto/osm2pgsql
that referenced
this issue
Nov 25, 2020
This commit is adding projection support using the PROJ library from 6.1 onwards. There might be an issue with multithreading, but the old code most likely already had that issue. This needs to be checked. The CMake config will prefer using the old API, simply because I don't know how to check for the new API without it trying to use the new API with version 5-6.0 which don't have everything we need. The code will only work from 6.1 onwards. There is a new CMake cache variable USE_PROJ_LIB which can be set to: * "4": Use version 4 API. Fail CMake if it is not available. * "6": Use version 6 API. Fail CMake if it is not available. * "off": Disable PROJ support (resulting binary will only have WGS84 (4326) and Web Mercator (3857) support) * "auto": Use version 4 API if available, otherwise try version 6 API or fall back to no PROJ support. See osm2pgsql-dev#922
joto
added a commit
to joto/osm2pgsql
that referenced
this issue
Nov 25, 2020
This commit is adding projection support using the PROJ library from 6.1 onwards. The CMake config will prefer using the old API, simply because I don't know how to check for the new API without it trying to use the new API with version 5-6.0 which don't have everything we need. The code will only work from 6.1 onwards. There is a new CMake cache variable USE_PROJ_LIB which can be set to: * "4": Use version 4 API. Fail CMake if it is not available. * "6": Use version 6 API. Fail CMake if it is not available. * "off": Disable PROJ support (resulting binary will only have WGS84 (4326) and Web Mercator (3857) support) * "auto": Use version 4 API if available, otherwise try version 6 API or fall back to no PROJ support. Note that we have an issue with multithreading using the PROJ library, because we use it potentially from multiple threads. The information I can find about this seems to indicate that this is only a problem for error reporting, because of the error handling using a global errno-type variable. So we can't detect failed transformations reliably. But we don't care about those anyway in the code we have, so I believe this is probably not a huge problem. This affects the old and the new code, so this isn't something new. The API version 6 allows to do the multithreading correctly using a "context" which this commit does use. But the context is still shared between threads, because we are starting the threads after the reprojections have been initialized. So all of this is technically not correct, but we'll probably get away with it for the time being. We need larger refactorings in the geometry code anyway which we want to tackle soon and hopefully can resolve this correctly then. See osm2pgsql-dev#922
joto
added a commit
to joto/osm2pgsql
that referenced
this issue
Nov 25, 2020
This commit is adding projection support using the PROJ library from 6.1 onwards. The CMake config will prefer using the old API, simply because I don't know how to check for the new API without it trying to use the new API with version 5-6.0 which don't have everything we need. The code will only work from 6.1 onwards. There is a new CMake cache variable USE_PROJ_LIB which can be set to: * "4": Use version 4 API. Fail CMake if it is not available. * "6": Use version 6 API. Fail CMake if it is not available. * "off": Disable PROJ support (resulting binary will only have WGS84 (4326) and Web Mercator (3857) support) * "auto": Use version 4 API if available, otherwise try version 6 API or fall back to no PROJ support. Note that we have an issue with multithreading using the PROJ library, because we use it potentially from multiple threads. The information I can find about this seems to indicate that this is only a problem for error reporting, because of the error handling using a global errno-type variable. So we can't detect failed transformations reliably. But we don't care about those anyway in the code we have, so I believe this is probably not a huge problem. This affects the old and the new code, so this isn't something new. The API version 6 allows to do the multithreading correctly using a "context" which this commit does use. But the context is still shared between threads, because we are starting the threads after the reprojections have been initialized. So all of this is technically not correct, but we'll probably get away with it for the time being. We need larger refactorings in the geometry code anyway which we want to tackle soon and hopefully can resolve this correctly then. See osm2pgsql-dev#922
joto
added a commit
to joto/osm2pgsql
that referenced
this issue
Nov 26, 2020
This commit is adding projection support using the PROJ library from 6.1 onwards. The CMake config will prefer using the old API, simply because I don't know how to check for the new API without it trying to use the new API with version 5-6.0 which don't have everything we need. The code will only work from 6.1 onwards. There is a new CMake cache variable USE_PROJ_LIB which can be set to: * "4": Use version 4 API. Fail CMake if it is not available. * "6": Use version 6 API. Fail CMake if it is not available. * "off": Disable PROJ support (resulting binary will only have WGS84 (4326) and Web Mercator (3857) support) * "auto": Use version 4 API if available, otherwise try version 6 API or fall back to no PROJ support. Note that we have an issue with multithreading using the PROJ library, because we use it potentially from multiple threads. The information I can find about this seems to indicate that this is only a problem for error reporting, because of the error handling using a global errno-type variable. So we can't detect failed transformations reliably. But we don't care about those anyway in the code we have, so I believe this is probably not a huge problem. This affects the old and the new code, so this isn't something new. The API version 6 allows to do the multithreading correctly using a "context" which this commit does use. But the context is still shared between threads, because we are starting the threads after the reprojections have been initialized. So all of this is technically not correct, but we'll probably get away with it for the time being. We need larger refactorings in the geometry code anyway which we want to tackle soon and hopefully can resolve this correctly then. See osm2pgsql-dev#922
bob-beck
pushed a commit
to openbsd/ports
that referenced
this issue
Mar 7, 2021
new website at https://osm2pgsql.org/ osm2pgsql-dev/osm2pgsql#922 was fixed so this now builds against upcoming devel/proj 8 update. use MODPOSTGRESQL_TEST_CMD = ${MODCMAKE_TEST_TARGET} to run the tests using the cmake goop, pleasantly surprised to see that this works !
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Currently the build failed if proj >= 6.0.0 is used. The Flag ACCEPT_USE_OF_DEPRECATED_PROJ_API_H has to be defined.
see: https://github.com/postgis/postgis/pull/354/files#r242068366
The text was updated successfully, but these errors were encountered: