database/postgres
1
0
Fork 0
mirror of https://github.com/postgres/postgres.git synced 2025-11-19 13:42:17 +03:00
Code Activity

Revert refactoring of hex code to src/common/

Browse Source 
This is a combined revert of the following commits:
- c3826f8, a refactoring piece that moved the hex decoding code to
src/common/.  This code was cleaned up by aef8948, as it originally
included no overflow checks in the same way as the base64 routines in
src/common/ used by SCRAM, making it unsafe for its purpose.
- aef8948, a more advanced refactoring of the hex encoding/decoding code
to src/common/ that added sanity checks on the result buffer for hex
decoding and encoding.  As reported by Hans Buschmann, those overflow
checks are expensive, and it is possible to see a performance drop in
the decoding/encoding of bytea or LOs the longer they are.  Simple SQLs
working on large bytea values show a clear difference in perf profile.
- ccf4e27, a cleanup made possible by aef8948.

The reverts of all those commits bring back the performance of hex
decoding and encoding back to what it was in ~13.  Fow now and
post-beta3, this is the simplest option.

Reported-by: Hans Buschmann
Discussion: https://postgr.es/m/1629039545467.80333@nidsa.net
Backpatch-through: 14
This commit is contained in:
Michael Paquier
2021-08-19 09:20:13 +09:00
parent 2313dda9d4
commit 2576dcfb76
9 changed files with 127 additions and 304 deletions
Download Patch File Download Diff File Expand all files Collapse all files

15
src/backend/utils/adt/varlena.c
View File

@@ -22,7 +22,6 @@
#include "catalog/pg_collation.h"
#include "catalog/pg_type.h"
#include "common/hashfn.h"
#include "common/hex.h"
#include "common/int.h"
#include "common/unicode_norm.h"
#include "lib/hyperloglog.h"
@@ -305,12 +304,10 @@ byteain(PG_FUNCTION_ARGS)
if (inputText[0] == '\\' && inputText[1] == 'x')
{
size_t len = strlen(inputText);
uint64 dstlen = pg_hex_dec_len(len - 2);
bc = dstlen + VARHDRSZ; /* maximum possible length */
bc = (len - 2) / 2 + VARHDRSZ; /* maximum possible length */
result = palloc(bc);
bc = pg_hex_decode(inputText + 2, len - 2, VARDATA(result), dstlen);
bc = hex_decode(inputText + 2, len - 2, VARDATA(result));
SET_VARSIZE(result, bc + VARHDRSZ); /* actual length */
PG_RETURN_BYTEA_P(result);
@@ -399,15 +396,11 @@ byteaout(PG_FUNCTION_ARGS)
if (bytea_output == BYTEA_OUTPUT_HEX)
{
uint64 dstlen = pg_hex_enc_len(VARSIZE_ANY_EXHDR(vlena));
/* Print hex format */
rp = result = palloc(dstlen + 2 + 1);
rp = result = palloc(VARSIZE_ANY_EXHDR(vlena) * 2 + 2 + 1);
*rp++ = '\\';
*rp++ = 'x';
rp += pg_hex_encode(VARDATA_ANY(vlena), VARSIZE_ANY_EXHDR(vlena), rp,
dstlen);
rp += hex_encode(VARDATA_ANY(vlena), VARSIZE_ANY_EXHDR(vlena), rp);
}
else if (bytea_output == BYTEA_OUTPUT_ESCAPE)
{