| commit | 11854f7ea4e1b4c8fc6f0004e5126d4ce2ee4717 | |
| author | Olly Betts <olly@survex.com> | |
| Tue, 2 Jan 2018 00:16:46 +0000 (2 13:16 +1300) | ||
| committer | Olly Betts <olly@survex.com> | |
| Tue, 2 Jan 2018 00:16:46 +0000 (2 13:16 +1300) | ||
| tree | 2b4e674ec0d9217f611cc146877375760abb2f70 | treesnapshot (tar.gz zip) |
| parent | 8f1d78307ecd71c375c476a16fe5ba7d0f797009 | commitdiff |
Reduce overheads of PostList positional data support
Probably the majority of PostList objects aren't involved in phrases
or OP_NEAR during the match, so avoid the overhead of a position
list object when we know we won't need it. For glass, we lazily
create it (so only have the overhead of a pointer, while for honey
we use a special subclass to eliminate the size overhead completely).
We also now avoid copying interpolative coded positional data in
order to iterate it in most cases.
Probably the majority of PostList objects aren't involved in phrases
or OP_NEAR during the match, so avoid the overhead of a position
list object when we know we won't need it. For glass, we lazily
create it (so only have the overhead of a pointer, while for honey
we use a special subclass to eliminate the size overhead completely).
We also now avoid copying interpolative coded positional data in
order to iterate it in most cases.
30 files changed: