I'm porting a number of programs from Informix embedded SQL to
postgreSQL, and ran into the following problem. The optim_accounts.ec
program included the following:

<begin code>
EXEC SQL BEGIN DECLARE SECTION;
EXEC SQL INCLUDE acct_plan_web.h;
EXEC SQL END DECLARE SECTION;
<end code>

The acct_plan_web.h file is:

<begin code>
#ifndef H_ACCT_PLAN_WEB
#define H_ACCT_PLAN_WEB

typedef struct acct_plan_web_st {
long acct;
long plan;
long min_deposit;
long max_deposit;
} *user_t;

#endif
<end code>

The results are:

<begin output>
$ make optim_accounts.c
ecpg optim_accounts.ec
../acct_plan_web.h:3: ERROR: syntax error at or near "#define
H_ACCT_PLAN_WEB
"
<end output>

If I delete the header guard from acct_plan_web.h, it preprocesses
cleanly.

If I eliminate the declare section and use a more conventional

EXEC SQL INCLUDE acct_plan_web;

then I get:

<begin output>
$ make optim_accounts.c
ecpg optim_accounts.ec
optim_accounts.ec:482: ERROR: invalid datatype 'struct
acct_plan_web_st'
<end output>

Any ideas? Does ecpg not like typedefs? Am I going blind and missing
something obvious?

--
Al Balmer
Sun City, AZ

Al Balmer <albalmer (AT) att (DOT) net> wrote:
[...]

I played around, and found that when you

EXEC SQL INCLUDE mumble;

ecpg will look for 'mumble' before looking for 'mumble.h'.

Is there maybe a file 'acct_plan_web' around?

Other than that, it should work.
I'll append some sample code that works on my PostgreSQL 8.1.4 system.

Yours,
Laurenz Albe


-------------------- begin ecpg.pgc ---------------------
#include <stdio.h>
#include <stdlib.h>

EXEC SQL BEGIN DECLARE SECTION;
EXEC SQL INCLUDE ecpg;
EXEC SQL END DECLARE SECTION;

int main(int argc, char** argv) {
EXEC SQL BEGIN DECLARE SECTION;
my_t my = (my_t)malloc(sizeof(struct mystruct));

EXEC SQL END DECLARE SECTION;

EXEC SQL CONNECT TO DEFAULT;

EXEC SQL SELECT 1, 2 INTO :my;

EXEC SQL DISCONNECT;

printf("my = { %ld, %ld }\n", my->a, my->b);

return 0;
}
------------------------ end ----------------------------

-------------------- begin ecpg.h -----------------------
typedef struct mystruct {
long a;
long b;
} *my_t;
------------------------ end ----------------------------

$ ecpg ecpg
$ gcc -Wall -g -o ecpgtest ecpg.c -I/magwien/postgres/include \
-L/magwien/postgres/lib -lecpg -Wl,-rpath,/magwien/postgres/lib
$ ./ecpgtest
my = { 1, 2 }

On 06 Oct 2006 09:18:39 GMT, Laurenz Albe <invite (AT) spam (DOT) to.invalid>
wrote:

That's a good thing to know, though it doesn't seem to be the problem
here. I notice that the include file in your sample doesn't have a
header guard, and my problem seems to go away if I remove the header
guard.

Searching the archives, I found some hints that there was a bug
involving #define statements and typedefs, but that it was fixed some
time ago. I know that ecpg won't process #defines, but I think it
should ignore them.

Anyway, thanks for the reply. It's good to know that someone is using
ecpg <g>. We're porting some ancient Informix_se programs, and so far
we like postgreSQL very much. Our support people love the fact that
they can use pgAdmin to maintain the customers' Unix databases.
--
Al Balmer
Sun City, AZ

Al Balmer <albalmer (AT) att (DOT) net> wrote:
Ah, good, I thought there was another problem besides the #ifdefs.
You could report it as a bug on pgsql-bugs that the header guards make
ecpg choke. Maybe somebody will fix it some day.

Yours,
Laurenz Albe

On 09 Oct 2006 07:38:15 GMT, Laurenz Albe <invite (AT) spam (DOT) to.invalid>
wrote:

Dang! I figured out what it was. If the struct definition is used in a
declare section, the definition itself has to be in a declare section,
else the preprocessor doesn't recognize it. The problem only seemed to
go away because the preprocessor stopped complaining (a bug in itself,
probably.)

The modified include file is:
<begin code>
#ifndef H_ACCT_PLAN_WEB
#define H_ACCT_PLAN_WEB

EXEC SQL BEGIN DECLARE SECTION; <--- this line added
typedef struct acct_plan_web_st {
long acct;
long plan;
long min_deposit;
long max_deposit;
} *user_t;
EXEC SQL END DECLARE SECTION; <--- this line added
#endif
<end code>

--
Al Balmer
Sun City, AZ

Al Balmer <albalmer (AT) att (DOT) net> wrote:
So you think that the example I posted contains a bug that does not get
reported by ecpg? Or do I misunderstand you?
If you surround the EXEC SQL INCLUDE with a DECLARE SECTION, the struct
definition _is_ inside the DECLARE SECTION.

Do I get it right that you removed the EXEC SQL BEGIN/END DECLARE SECTION
surrounding the EXEC SQL INCLUDE in the .c file?

That would work. Take care that the header file is never #included in
regular C code.

Yours,
Laurenz Albe

On 11 Oct 2006 07:25:11 GMT, Laurenz Albe <invite (AT) spam (DOT) to.invalid>
wrote:

program, and the EXEC SQL INCLUDE was not in a declare section, but
the include file was the same. Loking back, the original program did
have the EXEC SQL INCLUDE in a declare section, so the mystery
remains. Now I have to go back to the original program and verify that
the modified header works with that (or remove the include from the
declare section there, too.)

I realize I'm ignorant on this subject. I've been writing C since the
mid-70's, but this is the first time I've had to deal with embedded
SQL. Quite likely the last - it's a tiny part of a large port, and
once everything builds and runs, I'll probably never have to deal with
this part again.

--
Al Balmer
Sun City, AZ