Thread: Which parts of src/backend/nodes/print.c are used?

src/backend/nodes/print.c contains a number of functions that print node 
types, mostly to stdout.  Most of these are not actually used anywhere 
in the code.  Are they meant to be inserted into the code ad hoc for 
debugging?  Is anyone using these?

This file has clearly not been updated substantially in a long time, and 
functions like print_expr() are clearly outdated.

elog_node_display() and its callees are used, but I suppose these could 
be kept locally in postgres.c.

Other than that, is this file still needed?

Peter Eisentraut <peter@eisentraut.org> writes:
> src/backend/nodes/print.c contains a number of functions that print node 
> types, mostly to stdout.  Most of these are not actually used anywhere 
> in the code.  Are they meant to be inserted into the code ad hoc for 
> debugging?  Is anyone using these?

Personally, I use pprint() a lot.  (I invoke it manually from gdb and
then look into the postmaster log for results.)  Its cousins such as
format_node_dump look like they were added by people with slightly
different tastes in output format, so they probably have a
constituency somewhere.

I tend to agree that print_rt() and the other tree-printing routines
below it (down to, but not including, print_slot) are not as useful
as invoking the outfuncs.c code; but others might think differently.
Sometimes you don't want all the gory detail.

            regards, tom lane

On 12/4/23 05:50, Tom Lane wrote:
> Peter Eisentraut <peter@eisentraut.org> writes:
>> src/backend/nodes/print.c contains a number of functions that print node
>> types, mostly to stdout.  Most of these are not actually used anywhere
>> in the code.  Are they meant to be inserted into the code ad hoc for
>> debugging?  Is anyone using these?
> 
> Personally, I use pprint() a lot.  (I invoke it manually from gdb and
> then look into the postmaster log for results.)  Its cousins such as
> format_node_dump look like they were added by people with slightly
> different tastes in output format, so they probably have a
> constituency somewhere.
> 
> I tend to agree that print_rt() and the other tree-printing routines
> below it (down to, but not including, print_slot) are not as useful
> as invoking the outfuncs.c code; but others might think differently.
> Sometimes you don't want all the gory detail.

I've wondered about these functions for years. I use pprint a lot, and I've wanted to use 
print_slot/print_rt/print_tl (especially print_slot), but they never seemed to do anything. For 
instance debugging `SELECT 1;`:

(gdb) b ExecResult
Breakpoint 1 at 0x5fcc25f1ffcb: file nodeResult.c, line 68.
(gdb) c
Continuing.

Breakpoint 1, ExecResult (pstate=0x5fcc285272f8) at nodeResult.c:68
68      {
(gdb) call print_rt(((PlannedStmt *)pstate->plan)->rtable)
(gdb) call print_slot(pstate->ps_ResultTupleSlot)
(gdb)

Even with log_min_messages and client_min_messages set to DEBUG5, nothing appears in psql or the log 
or gdb. How are you supposed to use these functions?

Or if you want a real table, I still see no output after `ExecScanFetch` with:

create table s(i) as select generate_series(1,10);
select i from s;

I even tried dup'ing the backend's stdout to a file, but still got nothing:

(gdb) call creat("/tmp/pgout", 0600)
$1 = 103
(gdb) call dup2(103, 1)
'dup2' has unknown return type; cast the call to its declared return type
(gdb) call (int)dup2(103, 1)
$2 = 1
(gdb) b ExecScanFetch
Breakpoint 1 at 0x5fcc25ef026e: file execScan.c, line 37.
(gdb) c
Continuing.

Breakpoint 1, ExecScanFetch (node=node@entry=0x5fcc2852d348, 
accessMtd=accessMtd@entry=0x5fcc25f20b74 <SeqNext>, recheckMtd=recheckMtd@entry=0x5fcc25f20b28 
<SeqRecheck>) at execScan.c:37
37      {
(gdb) fin
Run till exit from #0  ExecScanFetch (node=node@entry=0x5fcc2852d348, 
accessMtd=accessMtd@entry=0x5fcc25f20b74 <SeqNext>, recheckMtd=recheckMtd@entry=0x5fcc25f20b28 
<SeqRecheck>) at execScan.c:37
0x00005fcc25ef044a in ExecScan (node=0x5fcc2852d348, accessMtd=accessMtd@entry=0x5fcc25f20b74 
<SeqNext>, recheckMtd=recheckMtd@entry=0x5fcc25f20b28 <SeqRecheck>) at execScan.c:180
180                     return ExecScanFetch(node, accessMtd, recheckMtd);
Value returned is $3 = (TupleTableSlot *) 0x5fcc2852d538
(gdb) call print_slot($3)
(gdb)

Yours,

-- 
Paul              ~{:-)
pj@illuminatedcomputing.com