Thread: pg_dump bug (was Re: [SQL] Slow Inserts Again)
Herouth Maoz wrote: > It's all in the docs. Why don't you try to copy some of the rows into a > temporary table, and dump that table using pg_dump -a -t table_name dbname? > It may give you a clue. this is unrelated to the slow insert bug, but Herouth's suggestion reminded me of something that needs to be looked at before 6.5 is out of beta: pg_dump seems to have problems with mixed case tablenames. There doesn't seem to be a way to send a quoted tablename into pg_dump as the value for a -t option, in 6.4.2 (example below). Can someone try this on 6.5beta? I know some issues with quoting output of pg_dump (i.e. COPY) was addressed, I'm wondering if input parsing got touched. Actually, groveling through the source, it looks like even 6.4.2 should "do the right thing": the query is built with the class (table) name wrapped with fmtId(), which should do exactly this quoting. Anyone else see this? P.S. shouldn't the non existence of the table handed to pg_dump raise a user visible error? P.P.S. How does one go about setting up a second version of PG to test on the same machine, without interference with the production (older) version? I've only got the one machine to test on. test=> create table TestTable (a int, b text); CREATE test=> create table "TestTable" (c int, d text); CREATE test=> \q $ pg_dump -t TestTable test CREATE TABLE "testtable" ( "a" int4, "b" text); COPY "testtable" FROM stdin; \. $ pg_dump -t "TestTable" test CREATE TABLE "testtable" ( "a" int4, "b" text); COPY "testtable" FROM stdin; \. $ pg_dump -t \"TestTable\" test $ pg_dump test CREATE TABLE "testtable" ( "a" int4, "b" text); CREATE TABLE "TestTable" ( "c" int4, "d" text); COPY "testtable" FROM stdin; \. COPY "TestTable" FROM stdin; \. -- Ross J. Reedstrom, Ph.D., <reedstrm@rice.edu> NSBRI Research Scientist/Programmer Computer and Information Technology Institute Rice University, 6100 S. Main St., Houston, TX 77005
"Ross J. Reedstrom" <reedstrm@rice.edu> writes:
> P.S. shouldn't the non existence of the table handed to pg_dump raise a
> user visible error?
Probably.
> P.P.S. How does one go about setting up a second version of PG to test
> on the same machine, without interference with the production (older)
> version? I've only got the one machine to test on.
No problem, I do that all the time. When running "configure", override
the default install location *and* the default port number. For
example, I build test versions using
--with-pgport=5440 --prefix=/users/postgres/testversion
(You might also want --enable-cassert when working with beta code.)
Build, and make install as per normal --- it will go into the directory
you specified as the prefix. Before doing initdb or starting the test
postmaster, make sure you have environment set up to reference the test
location; the critical stuff is PATH, PGLIB, PGDATA, and USER (initdb
uses USER to decide who'll be superuser).
regards, tom lane
> ... pg_dump seems to have problems with mixed case tablenames. There
> doesn't seem to be a way to send a quoted tablename into pg_dump as the
> value for a -t option, in 6.4.2 (example below). Can someone try this on
> 6.5beta? I know some issues with quoting output of pg_dump (i.e. COPY)
> was addressed, I'm wondering if input parsing got touched.
pg_dump explicitly converts all table names to lowercase. I've got a
patch which looks for a table name which starts with a double quote,
and suppresses the case conversion if so:
[postgres@golem pg_dump]$ pg_dump -t '"MixedCase"' postgres
CREATE TABLE "MixedCase" (
"i" int4);
COPY "MixedCase" FROM stdin;
1
2
\.
Patch enclosed for you to try. Bruce, any reason not to apply this to
the tree?
- Tom
--
Thomas Lockhart lockhart@alumni.caltech.edu
South Pasadena, California*** pg_dump.c.orig Thu Apr 15 05:08:53 1999
--- pg_dump.c Tue May 4 13:47:01 1999
***************
*** 606,615 ****
int i;
tablename = strdup(optarg);
! for (i = 0; tablename[i]; i++)
! if (isascii((unsigned char) tablename[i]) &&
! isupper(tablename[i]))
! tablename[i] = tolower(tablename[i]);
}
break;
case 'v': /* verbose */
--- 606,626 ----
int i;
tablename = strdup(optarg);
! /* quoted string? Then strip quotes and preserve case... */
! if (tablename[0] == '"')
! {
! strcpy(tablename, &tablename[1]);
! if (*(tablename+strlen(tablename)-1) == '"')
! *(tablename+strlen(tablename)-1) = '\0';
! }
! /* otherwise, convert table name to lowercase... */
! else
! {
! for (i = 0; tablename[i]; i++)
! if (isascii((unsigned char) tablename[i]) &&
! isupper(tablename[i]))
! tablename[i] = tolower(tablename[i]);
! }
}
break;
case 'v': /* verbose */
Apply, please. It is a bug fix.
> > ... pg_dump seems to have problems with mixed case tablenames. There
> > doesn't seem to be a way to send a quoted tablename into pg_dump as the
> > value for a -t option, in 6.4.2 (example below). Can someone try this on
> > 6.5beta? I know some issues with quoting output of pg_dump (i.e. COPY)
> > was addressed, I'm wondering if input parsing got touched.
>
> pg_dump explicitly converts all table names to lowercase. I've got a
> patch which looks for a table name which starts with a double quote,
> and suppresses the case conversion if so:
>
> [postgres@golem pg_dump]$ pg_dump -t '"MixedCase"' postgres
> CREATE TABLE "MixedCase" (
> "i" int4);
> COPY "MixedCase" FROM stdin;
> 1
> 2
> \.
>
> Patch enclosed for you to try. Bruce, any reason not to apply this to
> the tree?
>
> - Tom
>
> --
> Thomas Lockhart lockhart@alumni.caltech.edu
> South Pasadena, California
> *** pg_dump.c.orig Thu Apr 15 05:08:53 1999
> --- pg_dump.c Tue May 4 13:47:01 1999
> ***************
> *** 606,615 ****
> int i;
>
> tablename = strdup(optarg);
> ! for (i = 0; tablename[i]; i++)
> ! if (isascii((unsigned char) tablename[i]) &&
> ! isupper(tablename[i]))
> ! tablename[i] = tolower(tablename[i]);
> }
> break;
> case 'v': /* verbose */
> --- 606,626 ----
> int i;
>
> tablename = strdup(optarg);
> ! /* quoted string? Then strip quotes and preserve case... */
> ! if (tablename[0] == '"')
> ! {
> ! strcpy(tablename, &tablename[1]);
> ! if (*(tablename+strlen(tablename)-1) == '"')
> ! *(tablename+strlen(tablename)-1) = '\0';
> ! }
> ! /* otherwise, convert table name to lowercase... */
> ! else
> ! {
> ! for (i = 0; tablename[i]; i++)
> ! if (isascii((unsigned char) tablename[i]) &&
> ! isupper(tablename[i]))
> ! tablename[i] = tolower(tablename[i]);
> ! }
> }
> break;
> case 'v': /* verbose */
-- Bruce Momjian | http://www.op.net/~candle maillist@candle.pha.pa.us | (610)
853-3000+ If your life is a hard drive, | 830 Blythe Avenue + Christ can be your backup. | Drexel Hill,
Pennsylvania19026
> Apply, please. It is a bug fix.
> > > ... pg_dump seems to have problems with mixed case tablenames.
> > pg_dump explicitly converts all table names to lowercase. I've got a
> > patch which looks for a table name which starts with a double quote,
> > and suppresses the case conversion if so:
Already done :)
- Tom
--
Thomas Lockhart lockhart@alumni.caltech.edu
South Pasadena, California