On 2022-12-19 Mo 17:48, David Rowley wrote: > On Tue, 20 Dec 2022 at 11:42, Tom Lane <t...@sss.pgh.pa.us> wrote: >> I think Peter is entirely right to question whether *this* type's >> output function is performance-critical. Who's got large tables with >> jsonpath columns? It seems to me the type would mostly only exist >> as constants within queries. > The patch touches code in the path of jsonb's output function too. I > don't think you could claim the same for that. >
I agree that some of the uses in the jsonpath code could reasonably just be converted to use appendStringInfoString() There are 5 uses in the jsonb code where the length param is a compile time constant: andrew@ub22:adt $ grep appendBinary.*[0-9] jsonb* jsonb.c: appendBinaryStringInfo(out, "null", 4); jsonb.c: appendBinaryStringInfo(out, "true", 4); jsonb.c: appendBinaryStringInfo(out, "false", 5); jsonb.c: appendBinaryStringInfo(out, ": ", 2); jsonb.c: appendBinaryStringInfo(out, " ", 4); None of these really bother me much, TBH. In fact the last one is arguably nicer because it tells you without counting how many spaces there are. Changing the type of the second argument to appendBinaryStringInfo to void* seems reasonable. cheers andrew -- Andrew Dunstan EDB: https://www.enterprisedb.com
