Обсуждение: ​function arguments are not PG_FUNCTION_ARGS, how to pass Node *escontext

hi. simple question....

The following one works.
------------------------------------------------------------
Datum
test_direct_inputcall(PG_FUNCTION_ARGS)
{
    char    *token  = PG_GETARG_CSTRING(0);
    Datum   numd;
    if (!DirectInputFunctionCallSafe(numeric_in, token,
                                        InvalidOid, -1,
                                        fcinfo->context,
                                        &numd))
    {
        elog(INFO,"convert to cstring failed");
        PG_RETURN_BOOL(false);
    }
    elog(INFO,"%s",DatumGetCString(DirectFunctionCall1(numeric_out,numd)));
    PG_RETURN_BOOL(true);
}
------------------------------------------------------------
--the following one does not work. will print out something is wrong

Datum
test_direct_inputcall(PG_FUNCTION_ARGS)
{
    char    *token  = PG_GETARG_CSTRING(0);
    Datum   numd;
    numd    = return_numeric_datum(token);
    elog(INFO,"%s",DatumGetCString(DirectFunctionCall1(numeric_out,numd)));
    PG_RETURN_BOOL(true);
}

static
Datum return_numeric_datum(char *token)
{
    Datum   numd;
    Node    *escontext;

    if (!DirectInputFunctionCallSafe(numeric_in, token,
                                    InvalidOid, -1,
                                    escontext,
                                    &numd));
        elog(INFO,"something is wrong");
    return numd;
}
------------------------------------------------------------
I wonder how to make it return_numeric_datum works in functions that
arguments are not PG_FUNCTION_ARGS.

To make it work, I need to understand the Node *context, which is kind
of a vague idea for me.
In the top level function (input as PG_FUNCTION_ARGS) the Node
*context can be used to print out errors and back to normal state, I
kind of get it.

I really appreciate someone showing a minimal, reproducible example
based on return_numeric_datum....

static
Datum return_numeric_datum(char *token)
{    Datum   numd;    Node    *escontext;
    if (!DirectInputFunctionCallSafe(numeric_in, token,                                    InvalidOid, -1,                                    escontext,                                    &numd));        elog(INFO,"something is wrong");    return numd;
}

To start with, the semicolon at the end of that if appears bogus. The elog is indented to look like it's conditioned by the if but the semicolon makes it not be.

There are compiler switches in modern gcc at least that help you detect things like this.

cheers

andrew

--
Andrew Dunstan
EDB: https://www.enterprisedb.com

On Mon, Jun 26, 2023 at 7:50 PM Andrew Dunstan <andrew@dunslane.net> wrote:
>
>
> On 2023-06-26 Mo 07:20, jian he wrote:
>
> static
> Datum return_numeric_datum(char *token)
> {
>     Datum   numd;
>     Node    *escontext;
>
>     if (!DirectInputFunctionCallSafe(numeric_in, token,
>                                     InvalidOid, -1,
>                                     escontext,
>                                     &numd));
>         elog(INFO,"something is wrong");
>     return numd;
> }
>
>
> To start with, the semicolon at the end of that if appears bogus. The elog is indented to look like it's conditioned
bythe if but the semicolon makes it not be. 
>
> There are compiler switches in modern gcc at least that help you detect things like this.
>
>
> cheers
>
>
> andrew
>
> --
> Andrew Dunstan
> EDB: https://www.enterprisedb.com


sorry. It was my mistake.

>     Node    *escontext;
>     if (!DirectInputFunctionCallSafe(numeric_in, token,
>                                     InvalidOid, -1,
>                                     escontext,
>                                     &numd))
>         elog(INFO,"something is wrong");

I wonder about the implication of just declaring Node *escontext in here.
In this DirectInputFunctionCallSafe, what does escontext point to.

On Mon, Jun 26, 2023 at 9:32 PM jian he <jian.universality@gmail.com> wrote:
>
> On Mon, Jun 26, 2023 at 7:50 PM Andrew Dunstan <andrew@dunslane.net> wrote:
> >
> >
> > On 2023-06-26 Mo 07:20, jian he wrote:
> >
> > static
> > Datum return_numeric_datum(char *token)
> > {
> >     Datum   numd;
> >     Node    *escontext;
> >
> >     if (!DirectInputFunctionCallSafe(numeric_in, token,
> >                                     InvalidOid, -1,
> >                                     escontext,
> >                                     &numd));
> >         elog(INFO,"something is wrong");
> >     return numd;
> > }
> >
> >
> > To start with, the semicolon at the end of that if appears bogus. The elog is indented to look like it's conditioned by the if but the semicolon makes it not be.
> >
> > There are compiler switches in modern gcc at least that help you detect things like this.
> >
> >
> > cheers
> >
> >
> > andrew
> >
> > --
> > Andrew Dunstan
> > EDB: https://www.enterprisedb.com
>
>
> sorry. It was my mistake.
>
> >     Node    *escontext;
> >     if (!DirectInputFunctionCallSafe(numeric_in, token,
> >                                     InvalidOid, -1,
> >                                     escontext,
> >                                     &numd))
> >         elog(INFO,"something is wrong");
>
> I wonder about the implication of just declaring Node *escontext in here.
> In this DirectInputFunctionCallSafe, what does escontext point to.

gcc -Wempty-body will detect my error.

However, gcc -Wextra includes  -Wempty-body and  -Wuninitialized and others.

i guess in here, the real question in here is how to initialize escontext.

See code for pg_input_error_info and friends for examples.

cheers

andrew

--
Andrew Dunstan
EDB: https://www.enterprisedb.com