summaryrefslogtreecommitdiffstats
path: root/searchlib
Commit message (Collapse)AuthorAgeFilesLines
* use same name for abstract node typesArne H Juul2021-10-197-24/+22
|
* remove leftoversArne H Juul2021-10-191-2/+0
|
* extend unit testArne H Juul2021-10-191-0/+2
|
* extend unit testArne H Juul2021-10-191-2/+18
|
* handle new itemsArne H Juul2021-10-191-0/+4
|
* use custom types for true/false (NodeTypes::True, ProtonTrue, etc)Arne H Juul2021-10-198-9/+44
|
* use search::endDocIdArne H Juul2021-10-191-1/+1
|
* add ITEM_TRUE and ITEM_FALSE query nodesArne H Juul2021-10-1823-3/+138
|
* report more issues from searchlibHåvard Pettersen2021-10-1515-43/+57
|
* Merge pull request #19555 from vespa-engine/arnej/docsumreply-classArne H Juul2021-10-157-40/+94
|\ | | | | make DocsumReply a proper class
| * make destructor defaultArne H Juul2021-10-151-1/+1
| |
| * update after reviewArne H Juul2021-10-154-23/+28
| |
| * more useful DocsumReplyArne H Juul2021-10-146-16/+47
| |
| * make DocsumReply a proper classArne H Juul2021-10-146-39/+57
| |
* | Use the enum idx that has already been looked up and cached.Henning Baldersheim2021-10-145-10/+22
|/
* Merge pull request #19534 from vespa-engine/arnej/only-slime-docsumsArne H Juul2021-10-147-42/+18
|\ | | | | Arnej/only slime docsums
| * track API change - "root slime" is always true nowArne H Juul2021-10-133-7/+3
| |
| * only permit docsums as slimeArne H Juul2021-10-132-21/+10
| | | | | | | | | | * remove alternative with vector of binary blobs, we only use the "root slime" option now.
| * remove useRootSlime() optionArne H Juul2021-10-132-14/+5
| | | | | | | | | | * production path always sets this flag, so we should always use the code path for "useRootSlime() == true"
* | report onnx model evaluation as an issueHåvard Pettersen2021-10-132-1/+17
| |
* | report more issuesHåvard Pettersen2021-10-131-3/+4
|/ | | | also drop the prefix of issues that are logged by the default handler
* Consider the execute instructions comming in the rankprofiles config too.Henning Baldersheim2021-10-123-3/+41
|
* Merge pull request #19513 from ↵Henning Baldersheim2021-10-122-5/+13
|\ | | | | | | | | vespa-engine/havardpe/avoid-core-report-issue-on-unsupported-operation-during-attribute-create-blueprint avoid core and report issue
| * avoid core and report issueHåvard Pettersen2021-10-122-5/+13
| |
* | revert to logging for issue detected during setupArne H Juul2021-10-111-4/+2
| |
* | report issues instead of logging warnings in some featuresArne H Juul2021-10-1116-96/+98
|/
* extend element size to at least contain all matchesHåvard Pettersen2021-10-081-1/+2
| | | | | | | Increase element size rather than reduce the number of matches. This is to avoid getting scores over 1 for cases where we have matches outside the element. We do not know if or when this might happen, but reversing the guard makes it more robust.
* wire reported issues into search/docsum repliesHåvard Pettersen2021-10-078-3/+94
|
* Do not install fat jar for searchlib (unused)gjoranv2021-10-071-1/+0
|
* Update Verizon Media copyright notices.gjoranv2021-10-07129-129/+129
|
* Update 2019 Yahoo Holdings copyright notices.gjoranv2021-10-072-2/+2
|
* Update 2018 copyright notices.gjoranv2021-10-0767-68/+68
|
* Update 2017 copyright notices.gjoranv2021-10-071862-1874/+1874
|
* Merge pull request #19431 from vespa-engine/havardpe/add-errors-to-replyHåvard Pettersen2021-10-061-0/+6
|\ | | | | add errors to reply messages
| * add errors to reply messagesHåvard Pettersen2021-10-061-0/+6
| |
* | Remove temporary test caseLester Solbakken2021-10-061-8/+0
| |
* | Add non-primitive tensor expand functionLester Solbakken2021-10-063-2/+31
|/
* Merge pull request #19398 from ↵Henning Baldersheim2021-10-014-10/+48
|\ | | | | | | | | vespa-engine/havardpe/estimate-or-as-saturated-sum use saturated sum as hit estimate for OR
| * use saturated sum as hit estimate for ORHåvard Pettersen2021-10-014-10/+48
| | | | | | | | | | | | | | | | | | | | The upper bound for the hit estimate is the docid limit. The lower bound for the docid limit is the maximum child estimate. In the future we probably want a solution where everyone knows the docid limit from the start, which might also require us to have a separate way of estimating the document frequency if we want to avoid ranking score regression.
* | Pass destinationJon Bratseth2021-09-301-1/+1
|/ | | | | This allows embedders to switch on it to enable bucket testing and similar.
* file should reflect name of functionArne H Juul2021-09-291-1/+1
|
* use common binary_hamming_distance functionArne H Juul2021-09-281-17/+2
|
* Specify stack size for compiler when resolving blueprint.Tor Egge2021-09-271-3/+9
|
* Merge pull request #19260 from vespa-engine/arnej/java-add-hammingArne H Juul2021-09-236-3/+12
|\ | | | | Arnej/java add hamming
| * add unit testArne H Juul2021-09-232-0/+3
| |
| * add "hamming" functionArne H Juul2021-09-234-3/+9
| |
* | allow generated PB files outside source treeArne H Juul2021-09-221-1/+1
|/
* Remove explicit tracking of enum store and multi-value address space usage.Geir Storli2021-09-212-11/+0
| | | | | This has been replaced by generic tracking of address space usage among components in attributes vectors in all sub databases.
* Change loading of nearest neighbor index to use direct I/O instead of mmapping.Geir Storli2021-09-1512-82/+102
| | | | This should reduce memory spike during loading.
* ExpressionTransformer: only setChildren if any child nodes is transformedLester Solbakken2021-09-151-4/+14
|