mirror of
https://github.com/postgres/postgres.git
synced 2025-07-26 01:22:12 +03:00
Fix ts_headline() to handle ORs and phrase queries more honestly.
This patch largely reverts what I did in commitsc9b0c678d
and78e73e875
. The maximum cover length limit that I added in78e73e875
(to band-aid over c9b0c678d's performance issues) creates too many user-visible behavior discrepancies, as complained of for example in bug #17691. The real problem with hlCover() is not what I thought at the time, but more that it seems to have been designed with only AND tsquery semantics in mind. It doesn't work quite right for OR, and even less so for NOT or phrase queries. However, we can improve that situation by building a variant of TS_execute() that returns a list of match locations. We already get an ExecPhraseData struct representing match locations for the primitive case of a simple match, as well as one for a phrase match; we just need to add some logic to combine these for AND and OR operators. The result is a list of ExecPhraseDatas, which hlCover can regard as having simple AND semantics, so that its old algorithm works correctly. There's still a lot not to like about ts_headline's behavior, but I think the remaining issues have to do with the heuristics used in mark_hl_words and mark_hl_fragments (which, likewise, were not revisited when phrase search was added). Improving those is a task for another day. Patch by me; thanks to Alvaro Herrera for review. Discussion: https://postgr.es/m/840.1669405935@sss.pgh.pa.us
This commit is contained in:
@ -206,6 +206,9 @@ extern bool TS_execute(QueryItem *curitem, void *arg, uint32 flags,
|
||||
extern TSTernaryValue TS_execute_ternary(QueryItem *curitem, void *arg,
|
||||
uint32 flags,
|
||||
TSExecuteCallback chkcond);
|
||||
extern List *TS_execute_locations(QueryItem *curitem, void *arg,
|
||||
uint32 flags,
|
||||
TSExecuteCallback chkcond);
|
||||
extern bool tsquery_requires_match(QueryItem *curitem);
|
||||
|
||||
/*
|
||||
|
Reference in New Issue
Block a user