joto created an issue (osm2pgsql-dev/osm2pgsql#2509)
The code in `src/db-copy-mgr.hpp` and `src/db-copy.[ch]pp` could do with some
refactoring. Here are the reasons:
* The `db_copy_mgr_t` class mixes two things: It handles the formatting of the
data as used in the COPY command.
And it handles the actual copy, the commit/rollback, and the interfacing with
the processor. You can see the
split when looking at where the m_processor member ist used. This mixture
makes it hard to unit test the
COPY format code.
This formatting part is extended in the new binary COPY format code in #2503
making this even more awkward.
* Making all of this dependend on the DELETER type doesn't look right to me. It
seems overkill to use a
template parameter for the DELETER code. The `db_deleter_by_id_t` and
`db_deleter_by_type_and_id_t` classes
really differ only in the `delete_rows()` function. Storing the type for each
entry in `m_deletables`
(in `item_t`) isn't so expensive that we couldn't jst do it in both cases. So
maybe all of this should
just be a flag somewhere to choose the right code in `delete_rows()`.
* In #2426 it was proposed to run multiple COPYs in parallel. If we split the
code doing the actual COPY from
the code that prepares the COPY data, this would be easier to implement.
I am not sure how all of this can look like, it needs some experimenting, but
here are some ideas:
* Refactoring the DELETER code would make `db_copy_mgr_t::m_current` not depend
on a template parameter any more
* This opens up the possiblity to move the COPY formatting code out of
`db_copy_mgr_t` into its own class that
is then only used in `db_copy_mgr_t`.
* Then `db_copy_mgr_t` can be redesigned to work with parallel COPYs.
These could hopefully be done in several small steps.
--
Reply to this email directly or view it on GitHub:
https://github.com/osm2pgsql-dev/osm2pgsql/issues/2509
You are receiving this because you are subscribed to this thread.
Message ID: <osm2pgsql-dev/osm2pgsql/issues/[email protected]>
_______________________________________________
Tile-serving mailing list
[email protected]
https://lists.openstreetmap.org/listinfo/tile-serving
