Hi, On Tue, Jan 25, 2022 at 02:42:14AM +0000, huangning290@xxxxxxxxx wrote: > Hi:I created a new variable-length data type, and now I want to create a GIN > index for it. According to the rules of GIN index, I created three functions: > extractValue, extractQuery, and compare. I made sure that the return value of > the first two functions is the address of the array, but when using the index > query, the GIN tuple data obtained by calling PG_GETARG_DATUM in the compare > function is incorrect, and it is misplaced! In memory the size of the data > header becomes something else, and the position of the first byte is not the > header, it becomes the fourth byte. So there is a high probability that the > function called is wrong or my return value is wrong when creating the index > or the error is somewhere else? It's hard to have an opinion without seeing any code extract. Have you checked contrib module for examples of other GIN opclasses, like pg_trgrm, or hstore for varlena datatype with GIN support? Note also that pgsql-hackers is probably a better mailing list for this kind of questions.
