Import data through Nominatim

.clang-format

@@ -0,0 +1,14 @@
+---
+# We'll use defaults from the LLVM style, but with 4 columns indentation.
+BasedOnStyle: GNU
+ColumnLimit: 100
+---
+Language: Cpp
+AllowShortFunctionsOnASingleLine: Inline
+AlwaysBreakAfterDefinitionReturnType: None
+AlwaysBreakAfterReturnType: None
+SpaceBeforeParens: ControlStatements
+
+AlignConsecutiveAssignments: Consecutive
+AlignConsecutiveDeclarations: Consecutive
+AlignConsecutiveDeclarations: Consecutive

.gitignore

@@ -15,3 +15,7 @@
 *~
 
 /*.pro.user
+build/
+.vscode
+*.code-workspace
+CMakeLists.txt.user

CMakeLists.txt

@@ -5,19 +5,22 @@ project(geocoder-nlp
 	DESCRIPTION "Geocoder NLP")
 
 set(CMAKE_INCLUDE_CURRENT_DIR ON)
-set(CMAKE_CXX_STANDARD 11)
+set(CMAKE_CXX_STANDARD 17)
 set(CMAKE_CXX_STANDARD_REQUIRED True)
 
 include(FindPkgConfig)
 include(FeatureSummary)
 include(GNUInstallDirs)
 
 find_package(PkgConfig REQUIRED)
+find_package(nlohmann_json 3.2.0 REQUIRED)
+find_package(Boost 1.30 COMPONENTS program_options REQUIRED)
 
 pkg_check_modules(MARISA marisa IMPORTED_TARGET)
 pkg_check_modules(KYOTOCABINET kyotocabinet IMPORTED_TARGET)
-pkg_check_modules(POSTAL postal IMPORTED_TARGET)
+pkg_check_modules(POSTAL libpostal IMPORTED_TARGET)
 pkg_check_modules(SQLITE3 sqlite3 IMPORTED_TARGET)
+pkg_check_modules(LIBPQXX libpqxx IMPORTED_TARGET)
 
 set(SRC
   src/geocoder.cpp
@@ -31,31 +34,70 @@ set(HEAD
 include_directories(thirdparty/sqlite3pp/headeronly_src)
 include_directories(src)
 
+# boost
+include_directories(${Boost_INCLUDE_DIR})
+
+# importer
+set(IMPSRC
+  importer/src/config.h
+  importer/src/main.cpp
+  importer/src/hierarchy.cpp
+  importer/src/hierarchy.h
+  importer/src/hierarchyitem.cpp
+  importer/src/hierarchyitem.h
+  importer/src/normalization.cpp
+  importer/src/normalization.h
+  importer/src/utils.cpp
+  importer/src/utils.h
+)
+add_executable(geocoder-importer ${SRC} ${HEAD} ${IMPSRC})
+target_link_libraries(geocoder-importer
+  PkgConfig::MARISA 
+  PkgConfig::KYOTOCABINET
+  PkgConfig::POSTAL
+  PkgConfig::SQLITE3
+  PkgConfig::LIBPQXX
+  nlohmann_json::nlohmann_json
+  ${Boost_LIBRARIES})
+
 # demo codes
 add_executable(geocoder-nlp
   demo/geocoder-nlp.cpp
   ${SRC}
   ${HEAD})
 
 target_link_libraries(geocoder-nlp
-  -lmarisa -lkyotocabinet -lpostal -lsqlite3)
+  PkgConfig::MARISA 
+  PkgConfig::KYOTOCABINET
+  PkgConfig::POSTAL
+  PkgConfig::SQLITE3)
 
 add_executable(nearby-line
   demo/nearby-line.cpp
   ${SRC}
   ${HEAD})
 
 target_link_libraries(nearby-line
-  -lmarisa -lkyotocabinet -lpostal -lsqlite3)
+  PkgConfig::MARISA 
+  PkgConfig::KYOTOCABINET
+  PkgConfig::POSTAL
+  PkgConfig::SQLITE3)
 
 add_executable(nearby-point
   demo/nearby-point.cpp
   ${SRC}
   ${HEAD})
 
 target_link_libraries(nearby-point
-  -lmarisa -lkyotocabinet -lpostal -lsqlite3)
+  PkgConfig::MARISA 
+  PkgConfig::KYOTOCABINET
+  PkgConfig::POSTAL
+  PkgConfig::SQLITE3)
 
+# install
+install(TARGETS geocoder-importer
+    DESTINATION ${CMAKE_INSTALL_BINDIR})
 
+# summary
 feature_summary(WHAT ALL FATAL_ON_MISSING_REQUIRED_PACKAGES)
 

Database.md

@@ -0,0 +1,49 @@
+# Geocoder NLP database format
+
+The geocoder database consists of several files which are expected to be in the
+same directory. All locations are described using singe coordinate to keep the
+files as small as possible.
+
+The files composing a database are:
+
+1. geonlp-primary.sqlite: SQLite database with location description and coordinate
+2. geonlp-normalized.trie: MARISA database with normalized strings
+3. geonlp-normalized-id.kch: Kyoto Cabinet database for linking MARISA and primary IDs
+
+## geonlp-primary.sqlite
+
+SQLite database contains location description, their organization into hierarchy
+of objects. 
+
+Table `object_primary` keeps location description. In this table, objects are
+stored sequentially (in terms of their `id`) according to the positioning in the
+object hierarchy with the children stored after parents. Table `hierarchy` has a
+record for each item (`id` from `object_primary`) with the children consisting
+of parent ID (`prim_id`) and the ID of the last child (`last_subobject`). 
+
+Object types are stored separately in `type` table with the type ID used in
+`object_primary`.
+
+Spatial queries are indexed using R-Tree with `box_id` used as a reference in
+`object_primary`. Namely, as all objects are stored as points, for storage
+efficiency, objects next to each other are set to have the same `box_id` and are
+found through `-rtree` tables.
+
+Table `meta` keeps database format version and is used to check version
+compatibility.
+
+## geonlp-normalized.trie
+
+All normalized strings are stored in MARISA database
+(https://github.com/s-yata/marisa-trie). Normalized strings are formed from
+`name` and other similar fields of `object_primary` table in
+`geonlp-primary.sqlite`. All strings are pushed into MARISA database that
+assigns its internal ID for each of the strings.
+
+## geonlp-normalized-id.kch
+
+Kyoto Cabinet (https://dbmx.net/kyotocabinet/) database for linking MARISA and
+primary IDs. Hash database variant is used where `key` is an ID provided by
+MARISA for a search string and value is an array of bytes consisting of
+`object_primary` IDs stored as `uint32_t` one after another. The array is stored
+using `std::string`.

README.md

@@ -1,8 +1,11 @@
-# geocoder-nlp
+# Geocoder NLP
 
-This is a geocoder C++ library that uses libpostal to parse the user
+This is a geocoder C++ library that targets offline use by mobile 
+applications. It is able to perform forward and reverse geocoding. 
+For forward geocoding, it uses libpostal to parse the user
 request, normalize the parsed result, and search for the match in
-geocoder database. 
+geocoder database. In addition to traditional reverse geocoding, it is 
+able to find points of interest close to the reference point or line.
 
 The library includes demo program showing how to use it. Its also used
 as one of the geocoders in OSM Scout Server
@@ -29,7 +32,7 @@ libraries mentioned above.
 ## Databases
 
 At present, the datasets required for the geocoder to function are distributed 
-as a part of OSM Scout Server datasets . 
+as a part of OSM Scout Server datasets.
 
 If you use the geocoder with the full libpostal installation, you don't need to 
 get the libpostal datasets from that location, but can use the datasets 
@@ -43,8 +46,10 @@ To use country-specific datasets, you would have to get:
 In addition, the prepared geocoder databases are available at 
 geocoder/SELECT THE NEEDED ONES.
 
+Database format is described in [separate document](Database.md).
+
 ## Acknowledgments
 
-libpostal: https://github.com/openvenues/libpostal
+libpostal: Used for input parsing; https://github.com/openvenues/libpostal
 
-libosmscout: http://libosmscout.sourceforge.net
+Nominatim: Used for data import; https://nominatim.org/