On Thu, Aug 4, 2016 at 9:39 AM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> Andreas Seltenreich <seltenreich@gmx.de> writes:
>> the following statement triggers an assertion in tsearch:
>
>> select ts_delete(array_to_tsvector('{smith,smith,smith}'::text[]), '{smith,smith}'::text[]);
>> -- TRAP: FailedAssertion("!(k == indices_count)", File: "tsvector_op.c", Line: 511)
>
> Confirmed here. I notice that the output of array_to_tsvector() is
> already fishy in this example:
>
> regression=# select array_to_tsvector('{smith,smith,smith}'::text[]);
> array_to_tsvector
> -------------------------
> 'smith' 'smith' 'smith'
> (1 row)
>
> Shouldn't those have been merged together? You certainly don't get
> results like that from other tsvector-producing operations:
>
> regression=# select to_tsvector('smith smith smith');
> to_tsvector
> ---------------
> 'smith':1,2,3
> (1 row)
> regression=# select 'smith smith smith'::tsvector;
> tsvector
> ----------
> 'smith'
> (1 row)
>
> However, that does not seem to be the proximate cause of the crash
> in ts_delete, because this non-duplicated case still crashes:
>
> select ts_delete(array_to_tsvector('{smith,smithx,smithy}'::text[]), '{smith,smith}'::text[]);
>
> It kinda looks like you need more than one deletion request for
> the first entry in the sorted tsvector, because for example
> {smith,foo,toolbox} works but not {smith,too,toolbox}.
>
> I'm thinking there are two distinct bugs here.
The assertion in tsvector_delete_by_indices fails because its counting
algorithm doesn't expect indices_to_delete to contain multiple
references to the same index. Maybe that could be fixed by
uniquifying in tsvector_delete_arr before calling it, but since
tsvector_delete_by_indices already qsorts its input, it should be able
to handle duplicates cheaply. I was thinking something like this:
for (i = j = k = 0; i < tsv->size; i++) {
+ bool drop_lexeme = false;
+ /* * Here we should check whether current i is present in *
indices_to_deleteor not. Since indices_to_delete
is already sorted
- * we can advance it index only when we have match.
+ * we can advance it index only when we have match. We do this
+ * repeatedly, in case indices_to_delete contains
duplicate references
+ * to the same index. */
- if (k < indices_count && i == indices_to_delete[k])
+ while (k < indices_count && i == indices_to_delete[k]) {
+ drop_lexeme = true; k++;
- continue; }
+ if (drop_lexeme)
+ continue;
But that doesn't seem to be enough, there is something else wrong here
resulting in garbage output, maybe related to the failure to merge the
tsvector...
--
Thomas Munro
http://www.enterprisedb.com
