Add callbacks for delete actions#2344
Conversation
This special handling has been already disabled for flex output when we introduced the untagged callbacks.
src/output-pgsql.cpp
Outdated
| m_enable_way_area = read_style_file(options.style, &exlist); | ||
|
|
||
| m_ignore_untagged_objects = !options.extra_attributes; | ||
|
|
There was a problem hiding this comment.
Should be initialized with the other members above.
src/output-flex.cpp
Outdated
| { | ||
| if (m_delete_relation) { | ||
| m_relation_cache.init(rel); | ||
| select_relation_members(); |
There was a problem hiding this comment.
Deleted relations have no members, so there is no way of selecting any of them for two-stage processing. So this line can be removed.
| void output_flex_t::node_delete(osmium::Node const &node) | ||
| { | ||
| if (m_delete_node) { | ||
| m_context_node = &node; |
There was a problem hiding this comment.
You are reusing an existing context. This is fine, but the comments in src/output-flex.hpp for the calling_context enum referring to the process_* callbacks should be changed, too. (We probably already should have changed them when we added the untagged callbacks.)
| } | ||
| lua_rawset(lua_state, -3); | ||
|
|
||
| // Set the metatable of this object |
There was a problem hiding this comment.
It is weird that the process callbacks get a "real Lua object", and the delete callbacks only get a "Lua table" as parameter. This might have all sorts of consequences later, no sure. It definitely prevents us from ever calling methods on deleted OSM objects.
There was a problem hiding this comment.
This is just an implementation detail. If we need functions later, we can change it back to being a real Lua object (possibly of its own type). The user shouldn't rely on the type and assume duck typing is in place.
| Given the input file 'liechtenstein-2013-08-03.osm.pbf' | ||
| And the lua style | ||
| """ | ||
| change = osm2pgsql.define_table{ |
| { column = 'extra', sql_type = 'int' } | ||
| }} | ||
|
|
||
| function process(object) |
There was a problem hiding this comment.
Confusing that this function is called process like the process callbacks. Also should be local.
| name = 'change', | ||
| ids = {type = 'any', id_column = 'osm_id', type_column = 'osm_type'}, | ||
| columns = { | ||
| { column = 'extra', sql_type = 'int' } |
There was a problem hiding this comment.
type = 'int' should work fine.
|
|
||
| Then SELECT osm_type, count(*), sum(extra) FROM change GROUP BY osm_type | ||
| | osm_type | count | sum | | ||
| | N | 16773 | 16779| |
There was a problem hiding this comment.
mode superpicky on Spaces at the end missing in this and following lines.
| name = 'change', | ||
| ids = {type = 'any', id_column = 'osm_id', type_column = 'osm_type'}, | ||
| columns = { | ||
| { column = 'extra', sql_type = 'int' } |
2 participants
This adds callbacks
delete_node\delete_way\delete_relation. They are called for every object that is explicitly marked as deleted in a change file. The callbacks are not called for objects that are implicitly deleted to make room for a new version of the object.The callbacks receive an OSM object as a single parameter -- just like the other callbacks --, only the tags, nodes and member attributes as well as all geometry creation functions are disabled. This kind of information is usually not available for deleted objects. You can still access meta information like version, user etc. and use the object to insert something into a table.
This change required some cleanup to the code that handles untagged nodes. The pgsql output has some special handling in that it ignores untagged objects unless attribute handling is requested. We've long since removed the special handling for flex but the code for filtering was still in osmdata. This PR moves the handling into the pgsql output, which is a bit cleaner. This changes some of the outcome in the tests. See first commit for details.