Re: libpq bug? - Mailing list pgsql-general

Hi,

On Sat, Dec 29, 2018 at 1:37 AM patrick keshishian <pkeshish@gmail.com> wrote:
>
> On Fri, Dec 28, 2018 at 5:40 PM Igor Korot <ikorot01@gmail.com> wrote:
>>
>> Hi, Patrick,
>>
>> Here is my new code:
>>
>> int PostgresDatabase::GetTableOwner (const std::wstring
&schemaName,
>> const std::wstring &tableName, std::wstring &owner,
>> std::vector<std::wstring> &errorMsg)
>> {
>>     int result = 0;
>>     std::wstring query = L"SELECT u.usename FROM pg_class c, pg_user
>> u, pg_namespace n WHERE n.oid = c.relnamespace AND u.usesysid =
>> c.relowner AND n.nspname = $1 AND relname = $2";
>>     char *values[2];
>>     values[0] = NULL, values[1] = NULL;
>>     int charlength1 = schemaName.length() * sizeof( wchar_t ),
>> charlength2 = tableName.length() * sizeof( wchar_t );
>>     values[0] = new char[schemaName.length() * sizeof( wchar_t ) + 1];
>>     values[1] = new char[tableName.length() * sizeof( wchar_t ) + 1];
>>     memset( values[0], '\0', schemaName.length()  * sizeof(
wchar_t ) + 1 );
>>     memset( values[1], '\0', tableName.length() * sizeof(
wchar_t ) + 1 );
>>     strcpy( values[0], m_pimpl->m_myconv.to_bytes(
schemaName.c_str()
>> ).c_str() );
>>     strcpy( values[1], m_pimpl->m_myconv.to_bytes( tableName.c_str()
>> ).c_str() );
>>     int len1 = (int) schemaName.length() * sizeof( wchar_t );
>>     int len2 = (int) tableName.length() * sizeof( wchar_t );
>>     int length[2] = { len1, len2 };
>>     int formats[2] = { 1, 1 };
>>     PGresult *res = PQexecParams( m_db, m_pimpl->m_myconv.to_bytes(
>> query.c_str() ).c_str(), 2, NULL, values, length, formats, 1 );
>>     ExecStatusType status = PQresultStatus( res );
>>     if( status != PGRES_COMMAND_OK && status !=
PGRES_TUPLES_OK )
>>     {
>>         result = 1;
>>         std::wstring err = m_pimpl->m_myconv.from_bytes(
>> PQerrorMessage( m_db ) );
>>         errorMsg.push_back( L"Error executing query: " + err );
>>         PQclear( res );
>>     }
>>     else
>>     {
>>         owner = m_pimpl->m_myconv.from_bytes( PQgetvalue(
res, 0, 0 ) );
>>     }
>>     return result;
>> }
>>
>> The charlength2 variable contains the value of 8 and I'm still getting
>> the same error.
>
>
> I was hoping someone more versed in C++ would jump in to answer
your question. I haven't used C++ in at least a decade.
> You need to convert the wchar_t  data that wstring stores into
UTF-8. Personally, I would use iconv (common enough).

But that is very weird.
When I check what is stored in the values[1] array, I see the same
byte sequence as what I got from the database
information_schema.tables..
Maybe I should just upgrade the libpq and try the latest release?

>
> I assume the PostgresDatabase class is your own (?) I would add a
helper function to do the conversion. Here is a very rough template
for you to adapt if you think it helps you.

Yes, PostgresDatabase is my class.
I will look at that later today, but it would definitely be
interesting to get someone with the current C++
experience (especially with C++11), because I believe that I am doing
a conversion into UTF8.

The m_convert variable is declared as:

std::wstring_convert<std::codecvt_utf8<wchar_t> > m_myconv;

and so I think it is converting to the UTF8.

Thank you.

>
>
> #include <err.h>
>
> #include <stdlib.h>
>
> #include <string.h>
>
> #include <iostream>
>
> #include <string>
>
>
> #include <iconv.h>
>
>
> class PGDB {
>
> public:
>
>         // your stuff ...
>
>         iconv_t         ic;
>
>
>         PGDB(void) {
>
>                 setlocale(LC_CTYPE, "");
>
>                 ic = iconv_open("UTF-8", "wchar_t");
>
>                 if ((iconv_t)-1 == ic)
>
>                         errx(1, "iconv_open");
>
>         }
>
>         ~PGDB() {
>
>                 iconv_close(ic);
>
>         }
>
>         // caller should free()
>
>         char    *wchar2utf8(std::wstring const &ws) {
>
>                 char    *in, *buf, *out;
>
>                 size_t  bufsz, inbytes, outbytes;
>
>
>                 in = (char *)ws.data();
>
>                 inbytes = ws.length() * sizeof(wchar_t);
>
>                 outbytes = inbytes;
>
>                 bufsz = inbytes + 1; // XXX check for overflow
>
>
>                 buf = (char *)calloc(bufsz, 1);
>
>                 if (NULL == buf)
>
>                         err(1, NULL); // or throw something
>
>
>                 out = buf;
>
>                 if ((size_t)-1 == iconv(ic, &in, &inbytes, &out, &outbytes))
>
>                         errx(1, "iconv"); // or throw ...
>
>
>                 // TODO ensure inbytes is 0 (meaning all input consumed)
>
>                 return buf;
>
>         }
>
> };
>
>
> // demo using above PGDB class/code
>
> int main(int argc, char *argv[])
>
> {
>
>         char            *str;
>
>         size_t          i, n;
>
>         std::wstring    tab;
>
>         PGDB            pg;
>
>
>         tab = L"ºabcß";
>
>         str = pg.wchar2utf8(tab);
>
>
>         n = strlen(str);
>
>         for (i = 0; i < n; ++i) {
>
>                 printf("%02hhx ", str[i]);
>
>         }
>
>         printf("\n");
>
>         printf("->%s<-\n", str);
>
>
>         free(str);
>
>         return 0;
>
> }
>
>
>
> So in GetTableOwner() you'd call wchar2utf8() for the input wstring data and you don't do the new/memset/strcpy. Just
rememberto free() the returned utf8 string pointers after call to PQexecParams(). 
>
> cheers,
> --patrick
>
>
>
>>
>> Any idea?
>>
>> Thank you.
>>
>> On Fri, Dec 28, 2018 at 5:40 PM patrick keshishian <pkeshish@gmail.com> wrote:
>> >
>> > On Fri, Dec 28, 2018 at 3:07 PM Igor Korot <ikorot01@gmail.com> wrote:
>> >>
>> >> Hi,
>> >>
>> >> On Fri, Dec 28, 2018 at 4:51 PM patrick keshishian <pkeshish@gmail.com> wrote:
>> >> >
>> >> >
>> >> > On Fri, Dec 28, 2018 at 2:00 PM Igor Korot <ikorot01@gmail.com> wrote:
>> >> >>
>> >> >> Hi, ALL,
>> >> >> Following code:
>> >> >>
>> >> >> int PostgresDatabase::GetTableOwner (const std::wstring &schemaName,
>> >> >> const std::wstring &tableName, std::wstring &owner,
>> >> >> std::vector<std::wstring> &errorMsg)
>> >> >> {
>> >> >>    int result = 0;
>> >> >>     std::wstring query = L"SELECT u.usename FROM pg_class c, pg_user
>> >> >> u, pg_namespace n WHERE n.oid = c.relnamespace AND u.usesysid =
>> >> >> c.relowner AND n.nspname = $1 AND relname = $2";
>> >> >>     char *values[2];
>> >> >>     values[0] = NULL, values[1] = NULL;
>> >> >>     values[0] = new char[schemaName.length() + 1];
>> >> >>     values[1] = new char[tableName.length() + 1];
>> >> >>     memset( values[0], '\0', schemaName.length() + 1 );
>> >> >>     memset( values[1], '\0', tableName.length() + 1 );
>> >> >>     strcpy( values[0], m_pimpl->m_myconv.to_bytes( schemaName.c_str()
>> >> >> ).c_str() );
>> >> >>     strcpy( values[1], m_pimpl->m_myconv.to_bytes( tableName.c_str()
>> >> >> ).c_str() );
>> >> >>     int len1 = (int) schemaName.length();
>> >> >>     int len2 = (int) tableName.length();
>> >> >>     int length[2] = { len1, len2 };
>> >> >>     int formats[2] = { 1, 1 };
>> >> >>     PGresult *res = PQexecParams( m_db, m_pimpl->m_myconv.to_bytes(
>> >> >> query.c_str() ).c_str(), 2, NULL, values, length, formats, 1 );
>> >> >>     ExecStatusType status = PQresultStatus( res );
>> >> >>     if( status != PGRES_COMMAND_OK && status != PGRES_TUPLES_OK )
>> >> >>     {
>> >> >>         result = 1;
>> >> >>         std::wstring err = m_pimpl->m_myconv.from_bytes(
>> >> >> PQerrorMessage( m_db ) );
>> >> >>         errorMsg.push_back( L"Error executing query: " + err );
>> >> >>         PQclear( res );
>> >> >>     }
>> >> >>     else
>> >> >>     {
>> >> >>         owner = m_pimpl->m_myconv.from_bytes( PQgetvalue( res, 0, 0 ) );
>> >> >>     }
>> >> >>     return result;
>> >> >> }
>> >> >>
>> >> >> when ran with the call of
>> >> >>
>> >> >> GetTableOwner( "public", "abcß", owner, errorMsg );
>> >> >>
>> >> >> returns:
>> >> >>
>> >> >> ERROR: Invalid byte sequence for encoding UTF8.
>> >> >>
>> >> >> Does this mean I found the bug in the library?
>> >> >
>> >> >
>> >> > The bug is in your C++ code. "abcß" as tableName.lenght() (wstring) returns 4 (as in four characters) not
numberof bytes required to represent the intended string: 61 62 63 c3 9f 
>> >> > Since the last character is a 2 bytes in length. Therefore, your call to PQexecParams() specifies a shorter
lengthand hence an invalid UTF-8 sequence. 
>> >> >
>> >> > Furthermore, your value[] array allocation is in error since wstring::length returns number of characters, not
numberof bytes. so you will end up with buffer-overflows. 
>> >>
>> >> So I should use
>> >> https://stackoverflow.com/questions/9278723/how-can-i-get-the-byte-size-of-stdwstring
>> >> in both places?
>> >
>> >
>> > size() also returns 4. If you multiply it with sizeof(wchar_t) you will end up with maximum buffers size necessary
tohold the string (minus terminating \0), but not the correct length you are after. I am unsure of the "correct" C++
solution.
>> >
>> > Sorry,
>> > --patrick
>> >
>> >
>> >>
>> >> Thank you.
>> >>
>> >> >
>> >> > HTH,
>> >> > --patrick
>> >> >
>> >> >
>> >> >>
>> >> >> Any idea what I can do?
>> >> >>
>> >> >> Thank you.
>> >> >>