Add a helper for sorting search words from largest to smallest. This
should help search performance. For example, let's say our search index
is like so:
"the pokemon is cool"
the
the
the
...
* 1000
Our root word search would have to start 1000 new recursive queries. By
sorting by the largest word:
pokemon
pokemon
pokemon
...
* 10
We only have to do 10 recursive searches, assuming larger words are less
common, which will likely be the case most of the time
Signed-off-by: William Casarin <jb55@jb55.com>
This fixes an allocation issue with ndb_filter_init_with for small
page sizes. instead of allocating the buffer around pages, we allocate
based on total buffer size.
Fixes: f7aac3215575 ("filter: introduce ndb_filter_init_with")
Signed-off-by: William Casarin <jb55@jb55.com>
We forgot to move one DEBUG instance to NDB_LOG
Fixes: b4c2ff3d270a ("Only log to stdout if NDB_LOG is defined")
Signed-off-by: William Casarin <jb55@jb55.com>
Just a static function for now for creating smaller filter sizes. We
will use this for filters that we know are small so that we don't have
to allocate so many pages at once. It's likely the OS will only allocate
a single page anyways, but its nice to be explicit.
Changelog-Added: Add ndb_filter_init_with
Signed-off-by: William Casarin <jb55@jb55.com>
Make fulltext indices and note blocks optional. This will be useful for
quickly building databases when testing, since more stuff in the write
queue when writing can slow things down.
Signed-off-by: William Casarin <jb55@jb55.com>
We should be specifying that we've matched the id here, not authors. Not
that this would have any effect.. but still.
Signed-off-by: William Casarin <jb55@jb55.com>
This function can be used to check if a db is an index or not. We
will use this in future functions that rebuild indices.
Signed-off-by: William Casarin <jb55@jb55.com>
technically more accurate. we are about to introduce a new type called:
ndb_ts_u64_id
which would be confusing if we didn't do this
Signed-off-by: William Casarin <jb55@jb55.com>
This was the only thing that wasn't threadsafe. Add a simple mutex
instead of a queue so that polling is quick.
This also means we can't really return the internal subscriptions
anymore, so we remove that for now until we have a safer
interface.
Fixes: https://github.com/damus-io/nostrdb/issues/55
Signed-off-by: William Casarin <jb55@jb55.com>
subset testing for filters. Can be used to see if one subset is
redundant in the presence of a another in the local relay model
Changelog-Added: Add ndb_filter_is_subset_of
Signed-off-by: William Casarin <jb55@jb55.com>
filter equality testing. this works because field elements are sorted
Changelog-Added: Add ndb_filter_eq for filter equality testing
Signed-off-by: William Casarin <jb55@jb55.com>
Expose a way to get the set of filters for a subscription. On the rust
side, we should likely ndb_filter_clone each filter asap, because the
result of this function will only be valid up until the subscription
ends.
Signed-off-by: William Casarin <jb55@jb55.com>
This will find strings which match the beginning of other strings,
which seems wrong.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: William Casarin <jb55@jb55.com>
Sure, this format would be nice, but it's not what the code does.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: William Casarin <jb55@jb55.com>
If we make unknown_field simply discard, we can remove decoders and
have them discard those fields.
Now we can cut down struct bolt11 to only the fields needed by
invoice.c, and also speed up parsing a little.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: William Casarin <jb55@jb55.com>
Copy the latest, which has parsing fixes. We make a new explicit
"bolt11_decode_minimal" which doesn't check sigs, rather than neutering
the bolt11_decode logic.
As a bonus, this now correctly parses "LIGHTNING:BECH32..." format
(upper case, with prefix).
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: William Casarin <jb55@jb55.com>
