BUG #6216: Calling PQconnectdbParams from C++ with a char**

On Tue, Sep 20, 2011 at 11:05 AM, Lionel Elie Mamane <lionel@mamane.lu> wrote:

The following bug has been logged online:

Bug reference:      6216
Logged by:          Lionel Elie Mamane
Email address:      lionel@mamane.lu
PostgreSQL version: 9.1.0
Operating system:   Debian GNU/Linux
Description:        Calling PQconnectdbParams from C++ with a char**
Details:

In C++, a "char**" value is not convertible to a "const char**" value,
because that is not safe (see
http://www.parashift.com/c++-faq-lite/const-correctness.html#faq-18.17).

This means one cannot call libpq's PQconnectdbParams and friends passing
them a "char**" value for keywords and/or values, as these arguments are
declared "const char**".

So please apply this patch, which declares them "char const* const*"
instead, which is:
 - equivalent to "const char * const*"; use that one if you prefer
 - still true, since libpq won't write to they keyword/values arrays
 - allows to pass a "char**" to them and be recognised by a C++ compiler as
OK without a cast.

This patch is licensed under a "do whatever you want with it" license.

Thanks in advance.

--- postgresql-9.1-9.1.0.orig/src/interfaces/libpq/fe-connect.c
+++ postgresql-9.1-9.1.0/src/interfaces/libpq/fe-connect.c
@@ -291,8 +291,8 @@ static void freePGconn(PGconn *conn);
 static void closePGconn(PGconn *conn);
 static PQconninfoOption *conninfo_parse(const char *conninfo,
                          PQExpBuffer errorMessage, bool use_defaults);
-static PQconninfoOption *conninfo_array_parse(const char **keywords,
-                                        const char **values, PQExpBuffer errorMessage,
+static PQconninfoOption *conninfo_array_parse(char const* const*keywords,
+                                        char const* const*values, PQExpBuffer errorMessage,
                                        bool use_defaults, int expand_dbname);
 static char *conninfo_getval(PQconninfoOption *connOptions,
                               const char *keyword);
@@ -362,8 +362,8 @@ pgthreadlock_t pg_g_threadlock = default
 * call succeeded.
 */
 PGconn *
-PQconnectdbParams(const char **keywords,
-                                 const char **values,
+PQconnectdbParams(char const* const*keywords,
+                                 char const* const*values,
                                 int expand_dbname)
 {
       PGconn     *conn = PQconnectStartParams(keywords, values, expand_dbname);
@@ -381,8 +381,8 @@ PQconnectdbParams(const char **keywords,
 * check server status, accepting parameters identical to
PQconnectdbParams
 */
 PGPing
-PQpingParams(const char **keywords,
-                        const char **values,
+PQpingParams(char const* const*keywords,
+                        char const* const*values,
                        int expand_dbname)
 {
       PGconn     *conn = PQconnectStartParams(keywords, values, expand_dbname);
@@ -464,8 +464,8 @@ PQping(const char *conninfo)
 * See PQconnectPoll for more info.
 */
 PGconn *
-PQconnectStartParams(const char **keywords,
-                                        const char **values,
+PQconnectStartParams(char const* const*keywords,
+                                        char const* const*values,
                                        int expand_dbname)
 {
       PGconn     *conn;
@@ -4249,7 +4249,7 @@ conninfo_parse(const char *conninfo, PQE
 * keywords will take precedence, however.
 */
 static PQconninfoOption *
-conninfo_array_parse(const char **keywords, const char **values,
+conninfo_array_parse(char const* const*keywords, char const* const*values,
                                        PQExpBuffer errorMessage, bool use_defaults,
                                        int expand_dbname)
 {
--- postgresql-9.1-9.1.0.orig/src/interfaces/libpq/libpq-fe.h
+++ postgresql-9.1-9.1.0/src/interfaces/libpq/libpq-fe.h
@@ -235,14 +235,14 @@ typedef struct pgresAttDesc
 /* make a new client connection to the backend */
 /* Asynchronous (non-blocking) */
 extern PGconn *PQconnectStart(const char *conninfo);
-extern PGconn *PQconnectStartParams(const char **keywords,
-                                        const char **values, int expand_dbname);
+extern PGconn *PQconnectStartParams(char const* const*keywords,
+                                        char const* const*values, int expand_dbname);
 extern PostgresPollingStatusType PQconnectPoll(PGconn *conn);

 /* Synchronous (blocking) */
 extern PGconn *PQconnectdb(const char *conninfo);
-extern PGconn *PQconnectdbParams(const char **keywords,
-                                 const char **values, int expand_dbname);
+extern PGconn *PQconnectdbParams(char const* const*keywords,
+                                 char const* const*values, int expand_dbname);
 extern PGconn *PQsetdbLogin(const char *pghost, const char *pgport,
                        const char *pgoptions, const char *pgtty,
                        const char *dbName,
@@ -413,8 +413,8 @@ extern int  PQsetnonblocking(PGconn *conn
 extern int     PQisnonblocking(const PGconn *conn);
 extern int     PQisthreadsafe(void);
 extern PGPing PQping(const char *conninfo);
-extern PGPing PQpingParams(const char **keywords,
-                        const char **values, int expand_dbname);
+extern PGPing PQpingParams(char const* const*keywords,
+                        char const* const*values, int expand_dbname);

 /* Force the write buffer to be written (or at least try) */
 extern int     PQflush(PGconn *conn);

I agree with the rationale. Note query parameter arguments are
already declared this way.

merlin

On 21/09/2011 12:05 AM, Lionel Elie Mamane wrote:

The following bug has been logged online:

Bug reference: 6216
Logged by: Lionel Elie Mamane
Email address: lionel@mamane.lu
PostgreSQL version: 9.1.0
Operating system: Debian GNU/Linux
Description: Calling PQconnectdbParams from C++ with a char**
Details:

In C++, a "char**" value is not convertible to a "const char**" value,
because that is not safe (see
http://www.parashift.com/c++-faq-lite/const-correctness.html#faq-18.17).

This means one cannot call libpq's PQconnectdbParams and friends passing
them a "char**" value for keywords and/or values, as these arguments are
declared "const char**".

So please apply this patch, which declares them "char const* const*"
instead, which is:
- equivalent to "const char * const*"; use that one if you prefer

Yeah, this really annoyed me when I was working with C libraries from C++.

+1 from me; this should be applied.

Lionel: Can I get you to add the patch to the commitfest app?

https://commitfest.postgresql.org/

I'll accept review of it once you do and flag it as ready for committer
so we can streamline its inclusion.

As for wording: my *personal* preference is "const char * const" but I
don't know what the opinions of those who work with the code day-to-day are.

--
Craig Ringer

Craig Ringer <ringerc@ringerc.id.au> writes:

As for wording: my *personal* preference is "const char * const" but I
don't know what the opinions of those who work with the code day-to-day are.

+1. Isn't the other ordering deprecated by recent C standards?
(Or maybe I'm just thinking of where you're supposed to put "static",
but in any case "char const *" looks pretty weird to me.) Also,
the existing usages in libpq-fe.h look like that, and there's no good
reason for these to be randomly different.

regards, tom lane

ghtOn Wed, Sep 21, 2011 at 07:29:23AM +0800, Craig Ringer wrote:

On 21/09/2011 12:05 AM, Lionel Elie Mamane wrote:

Bug reference: 6216
Logged by: Lionel Elie Mamane
Email address: lionel@mamane.lu

In C++, a "char**" value is not convertible to a "const char**"
value, (...)

This means one cannot call libpq's PQconnectdbParams and friends
passing them a "char**" value for keywords and/or values, as these
arguments are declared "const char**".

Lionel: Can I get you to add the patch to the commitfest app?

https://commitfest.postgresql.org/

Sure.

Topic "performance" is not right, but only value I was allowed to put,
and it wanted a value...

I added my initial patch, and as far as I understand, I have to send
the revised patch to the list before I can register it at the
commitfest. So here is my revised patch, that uses "const char *const * "
like elsewhere in the same file instead of "char const* const*".

--
Lionel

On 21/09/2011 4:25 PM, Lionel Elie Mamane wrote:

I added my initial patch, and as far as I understand, I have to send
the revised patch to the list before I can register it at the
commitfest. So here is my revised patch, that uses "const char *const * "
like elsewhere in the same file instead of "char const* const*".

Yep, I'm happy with that. It does what it says and no more.

I have *NOT* done any detailed testing, as I'm in the middle of moving
house, but I find it hard to see how it can break anything that isn't
its self incorrect. You can still pass a `char *' to a `const char *
const' param, so it won't break any call sites.

--
Craig Ringer

Craig Ringer <ringerc@ringerc.id.au> writes:

On 21/09/2011 4:25 PM, Lionel Elie Mamane wrote:

I added my initial patch, and as far as I understand, I have to send
the revised patch to the list before I can register it at the
commitfest. So here is my revised patch, that uses "const char *const * "
like elsewhere in the same file instead of "char const* const*".

Yep, I'm happy with that. It does what it says and no more.

I went ahead and committed this, since there seems no very good reason
to make it wait for the next commitfest.

regards, tom lane