Обсуждение: BUG #13462: Impossible to use COPY FORMAT BINARY in chunks through libpq

The following bug has been logged on the website:

Bug reference:      13462
Logged by:          Korenberg Mark
Email address:      socketpair@gmail.com
PostgreSQL version: 9.3.8
Operating system:   Linux
Description:

If I use PQputCopyData() to pass parts of the binary protocol (as documented
in http://www.postgresql.org/docs/9.3/static/sql-copy.html ), postgres
binary socket data gets corrupted (i.e. protocol is broken).

It is unexpected that libpq inserts \d\0\0\0 on every PQputCopyData() call
even in BINARY COPY mode. So, it is impossible to write a program that
generate tuples in binary form dynamically without merging them into one big
buffer.

Part of program:
====================================
   static const struct {
        char magic[11];
        int32_t zero1;
        int32_t zero2;
    } __attribute__ ((packed)) bin_header = {"PGCOPY\n\377\r\n",0,0};

    if (PQputCopyData (conn, (const char *) &bin_header, sizeof
(bin_header)) != 1)
    {
        fprintf (stderr, "PQputCopyData failed\n");
        return 4;
    }

    struct
    {
        int16_t fields;
        int32_t ip1len;
        uint32_t ip1;
        int32_t ip2len;
        uint32_t ip2;
    } __attribute__ ((packed)) binary_line =   {
        2, 4, 0x01020304, 4, 0x05060708
    };


    for (i = 0; i < repeats; i++)
    {
        if (PQputCopyData (conn, (const char *) &binary_line, sizeof
(binary_line)) != 1)
        {
            fprintf (stderr, "PQputCopyData failed\n");
            return 4;
        }
    }

    static const int16_t copy_end = -1;
    if (PQputCopyData (conn, (const char *) ©_end, sizeof (copy_end)) !=
1)
    {
        fprintf (stderr, "PQputCopyData failed\n");
        return 4;
    }

    res = PQgetResult (conn);
    if (PQresultStatus (res) != PGRES_COMMAND_OK)
    {
        fprintf (stderr, "COPY failed: %s\n", PQerrorMessage (conn));
        PQclear (res);
        return 3;
    }
    PQclear (res);
===============================================
will generate:

"d\0\0\0\27PGCOPY\n\377\r\n\0\0\0\0\0\0\0\0\0d\0\0\0\26\2\0\4\0\0\0\4\3\2\1\4\0\0\0\10\7\6\5d\0\0\0\6\377\377X\0\0\0\4"

but actually should generate (not sure about tail)

"d\0\0\0\27PGCOPY\n\377\r\n\0\0\0\0\0\0\0\0\0\2\0\4\0\0\0\4\3\2\1\4\0\0\0\10\7\6\5\377\377X\0\0\0\4"

socketpair@gmail.com writes:
> If I use PQputCopyData() to pass parts of the binary protocol (as documented
> in http://www.postgresql.org/docs/9.3/static/sql-copy.html ), postgres
> binary socket data gets corrupted (i.e. protocol is broken).

I think your test program is buggy; specifically, it doesn't appear to
have any provision for dealing with data endianness, which means it
will not work on a little-endian machine.  Per the COPY binary format
spec:

    Headers and data are in network byte order.


            regards, tom lane

Thanks, but still not works.... How to debug ? strace:

sendto(3,
"d\0\0\0\27PGCOPY\n\377\r\n\0\0\0\0\0\0\0\0\0d\0\0\0\26\0\2\0\0\0\4\4\3\2\1\0\0\0\4\10\7\6\5d\0\0\0\6\377\377X\0\0\0\4",
59, MSG_NOSIGNAL, NULL, 0) = 59

2015-06-23 4:08 GMT+05:00 Tom Lane <tgl@sss.pgh.pa.us>:
> socketpair@gmail.com writes:
>> If I use PQputCopyData() to pass parts of the binary protocol (as documented
>> in http://www.postgresql.org/docs/9.3/static/sql-copy.html ), postgres
>> binary socket data gets corrupted (i.e. protocol is broken).
>
> I think your test program is buggy; specifically, it doesn't appear to
> have any provision for dealing with data endianness, which means it
> will not work on a little-endian machine.  Per the COPY binary format
> spec:
>
>         Headers and data are in network byte order.
>
>
>                         regards, tom lane



--
Segmentation fault

Where I can see example how to generate COPY FORMAT BINARY ? Seems,
I'm doing something wrong, but cannot figure out what exactly.
According to strace, library act as follow:

sendto ("COPY.....FROM STDIN (FORMAT binary)")
recv (...) => receive some response (that copy starts succesfully)
next, debugging and logging of PQputCopyData calls
logging of calling  PQgetResult, that reports IMMEDIATELLY, that COPY failed.
and AFTER that, library sends "\d\0\0\0\27PGCOPY\n\377\r\n\0\0\....."

WTF?

If my example will work, I want to contribute that example to
documentation (as canonical example of that stuff), since I cannot
find working example of generating binary copy data.


2015-06-23 4:08 GMT+05:00 Tom Lane <tgl@sss.pgh.pa.us>:
> socketpair@gmail.com writes:
>> If I use PQputCopyData() to pass parts of the binary protocol (as documented
>> in http://www.postgresql.org/docs/9.3/static/sql-copy.html ), postgres
>> binary socket data gets corrupted (i.e. protocol is broken).
>
> I think your test program is buggy; specifically, it doesn't appear to
> have any provision for dealing with data endianness, which means it
> will not work on a little-endian machine.  Per the COPY binary format
> spec:
>
>         Headers and data are in network byte order.
>
>
>                         regards, tom lane



--
Segmentation fault