Обсуждение: pg_proc.dat "proargmodes is not a 1-D char array"

On 2020-Sep-30, Craig Ringer wrote:

> Hi all
> 
> Random tip for future searchers. If you've modified pg_proc.dat and  initdb
> fails with "proargmodes is not a 1-D char array" - it could well actually
> be that the length of proargmodes does not match the length of
> proallargtypes given the test
> 
>                         ARR_DIMS(arr)[0] != numargs ||
> 
> in funcapi.c.

Perhaps we can improve these error messages like below.  (Or maybe just
keep it one message "proargmodes is not a 1-D char array of %d
elements"?)  There are about 5 places to change I think.

diff --git a/src/backend/utils/fmgr/funcapi.c b/src/backend/utils/fmgr/funcapi.c
index b9efa77291..c76c16f799 100644
--- a/src/backend/utils/fmgr/funcapi.c
+++ b/src/backend/utils/fmgr/funcapi.c
@@ -1167,10 +1167,11 @@ get_func_arg_info(HeapTuple procTup,
     {
         arr = DatumGetArrayTypeP(proargmodes);    /* ensure not toasted */
         if (ARR_NDIM(arr) != 1 ||
-            ARR_DIMS(arr)[0] != numargs ||
             ARR_HASNULL(arr) ||
             ARR_ELEMTYPE(arr) != CHAROID)
             elog(ERROR, "proargmodes is not a 1-D char array");
+        if (ARR_DIMS(arr)[0] != numargs)
+            elog(ERROR, "proargmodes is not %d elements long", numargs);
         *p_argmodes = (char *) palloc(numargs * sizeof(char));
         memcpy(*p_argmodes, ARR_DATA_PTR(arr),
                numargs * sizeof(char));

Alvaro Herrera <alvherre@alvh.no-ip.org> writes:
> Perhaps we can improve these error messages like below.  (Or maybe just
> keep it one message "proargmodes is not a 1-D char array of %d
> elements"?)  There are about 5 places to change I think.

I doubt that it's really worth expending more code on this.
Certainly I see no reason why that particular test is more likely
to fail than the others, in the presence of corrupt data --- and
I don't want to add individual elog's for each one.

Adding the expected length to the error message might be OK though.

            regards, tom lane

On Tue, Nov 17, 2020 at 10:32 AM Tom Lane <tgl@sss.pgh.pa.us> wrote:
> Adding the expected length to the error message might be OK though.

Certainly seems like we should do at least that much. The current
message is just wrong, right?

-- 
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company

Robert Haas <robertmhaas@gmail.com> writes:
> On Tue, Nov 17, 2020 at 10:32 AM Tom Lane <tgl@sss.pgh.pa.us> wrote:
>> Adding the expected length to the error message might be OK though.

> Certainly seems like we should do at least that much. The current
> message is just wrong, right?

It's incomplete, for sure.  Doesn't mention nulls either.

            regards, tom lane

On 2020-Nov-17, Tom Lane wrote:

> Robert Haas <robertmhaas@gmail.com> writes:
> > On Tue, Nov 17, 2020 at 10:32 AM Tom Lane <tgl@sss.pgh.pa.us> wrote:
> >> Adding the expected length to the error message might be OK though.
> 
> > Certainly seems like we should do at least that much. The current
> > message is just wrong, right?
> 
> It's incomplete, for sure.  Doesn't mention nulls either.

So let's go with this one.

Alvaro Herrera <alvherre@alvh.no-ip.org> writes:
> So let's go with this one.

WFM.

            regards, tom lane

On 2020-Nov-23, Tom Lane wrote:

> Alvaro Herrera <alvherre@alvh.no-ip.org> writes:
> > So let's go with this one.
> 
> WFM.

Thanks, pushed.