# Expose mysqlnd API through all MySQL extensions

- Thread: https://externals.io/message/60270
- Original archive: https://news-web.php.net/php.internals/60270
- Started by jpauli on 2012-04-24T11:35:12+00:00
- Messages: 8
- Reply by email to internals@lists.php.net
- More for agents: search with https://externals.io/search.md?q={query} · JSON API at https://externals.io/api · documentation at https://externals.io/llms.txt

---

## #60270 — jpauli — 2012-04-24T11:35:12+00:00

URL: https://externals.io/message/60270#60270

Hi all,

I'm actually putting my head into ext/mysqlnd and I noticed weird
things about its API.

As you should know, ext/mysqlnd is a replacement for libmysql, that
is, all our 3 MySQL-related extensions may use it, and it is done by
default for 5.4 branch.
However, I've been surprised that ext/mysqlnd expose its API through
ext/mysqli only...

- `mysqli_get_client_stats()`
- `mysqli_get_connection_stats()`
- mysqli_get_cache_stats()

Those three functions are ext/mysqlnd related, they are #ifdef 'ed in
the source code of ext/mysqli and not defined if ext/mysqli has been
compiled against libmysql.
There is a mismatch in the naming and the place here, those 3
functions should be part of ext/mysqlnd API, and be named
mysqlnd_****().

What if I choose to only compile pdo_mysql (against mysqlnd) and not
mysqli ?? I then would be able to use mysqlnd, but I would have access
to its statistics, as its functions are part of ext/mysqli... Strange
as well as embarrassing isn't it ?

I then suggest to expose ext/mysqlnd API through itself
(http://lxr.php.net/xref/PHP_5_3/ext/mysqlnd/php_mysqlnd.c#33) in
php_mysqlnd.c or create a new php_mysqlnd_fe.c
For the actual functions, please refer to
http://lxr.php.net/xref/PHP_5_3/ext/mysqli/mysqli_nonapi.c#392

Second, I would suggest to add a function that would reset the global
stat counter. The only way to do it actually is by passing in MINIT(),
so by restarting PHP ; it may be useful, and not very hard to develop,
to create a mysqlnd_stats_reset($someOptionsIfNeeded) function.

Any ideas, comments ?

Cheers,

Julien

---

## #60271 — johannes.schlueter@oracle.com — 2012-04-24T12:06:42+00:00

URL: https://externals.io/message/60270#60271 · In reply to: #60270

> - `mysqli_get_client_stats()`
> - `mysqli_get_connection_stats()`
> - mysqli_get_cache_stats()
> 
> Those three functions are ext/mysqlnd related, they are #ifdef 'ed in
> the source code of ext/mysqli and not defined if ext/mysqli has been
> compiled against libmysql.

With libmysql we don't have the statistics, therefore they are useless.

> There is a mismatch in the naming and the place here, those 3
> functions should be part of ext/mysqlnd API, and be named
> mysqlnd_****().

Until 5.4 we couldn't do "mysqlnd_get_connection_stats" as there was no
way to get from an arbitrary zval (PDO or mysqli object, mysql resource)
to the corresponding MYSQLND*.

The other reason is that one idea for mysqlnd was that it doesn't expose
anything to the userland while ext/mysql is legacy and get's no
additions and PDO would allow user-defined functions,while that's not
really nice.

> What if I choose to only compile pdo_mysql (against mysqlnd) and not
> mysqli ?? I then would be able to use mysqlnd, but I would have access
> to its statistics, as its functions are part of ext/mysqli... Strange
> as well as embarrassing isn't it ?

You simply shouldn't do that ;-)

> Second, I would suggest to add a function that would reset the global
> stat counter. The only way to do it actually is by passing in MINIT(),
> so by restarting PHP ; it may be useful, and not very hard to develop,
> to create a mysqlnd_stats_reset($someOptionsIfNeeded) function.
> 
> Any ideas, comments ?

Feel free to open a ticket in the bug tracker and assign it to us (mysql
user).

johannes
-- 
Johannes Schlüter, MySQL Engineering Connectors and Client Connectivity
ORACLE Deutschland B.V. & Co. KG, Riesstr. 25, D-80992 München

---

## #60272 — jpauli — 2012-04-24T12:27:18+00:00

URL: https://externals.io/message/60270#60272 · In reply to: #60271

2012/4/24 Johannes Schlüter <johannes.schlueter@oracle.com>:
> 
> > […]
>
> With libmysql we don't have the statistics, therefore they are useless.

Right, NP

>
> > […]
>
> Until 5.4 we couldn't do "mysqlnd_get_connection_stats" as there was no
> way to get from an arbitrary zval (PDO or mysqli object, mysql resource)
> to the corresponding MYSQLND*.

Aha, I like when an internals guy teach me internals stuff :)
So, OK for 5.3 , what about supporting that for >=5.4 ?

>
> The other reason is that one idea for mysqlnd was that it doesn't expose
> anything to the userland while ext/mysql is legacy and get's no
> additions and PDO would allow user-defined functions,while that's not
> really nice.

That was my first question : Is it better to provide mysqlnd API
through all 3 MySQL extensions, or through a common mysqlnd_***() API
?
I understand your thoughts, but I disagree as I think it would be much
more clean to expose it via mysqlnd_***() API than through each MySQL
ext (and for PDO that would be even more dirty). That would as well
add more and more #ifdef MYSQLND inside their source...
I was thinking the same way libxml2 API is provided inside ext/libxml
instead of through each PHP XML extension API.

>
> > […]
>
> You simply shouldn't do that ;-)

;-) OK, but that can be done easily. More and more people use
pdo_mysql and could be able to not compile other MySQL extensions (I
got some machines where only pdo_mysql is compiled in)

>
> > […]
>
> Feel free to open a ticket in the bug tracker and assign it to us (mysql
> user).

More, I'm gonna work on a patch to feed the bug report with ;-)
Stats are not the hardest part of ext/mysqlnd :)

Julien.P

---

## #60274 — Ulf Wendel — 2012-04-24T12:34:31+00:00

URL: https://externals.io/message/60270#60274 · In reply to: #60272

Am 24.04.2012 14:27, schrieb jpauli:
> I understand your thoughts, but I disagree as I think it would be much
> more clean to expose it via mysqlnd_***() API than through each MySQL
> ext (and for PDO that would be even more dirty). That would as well
> add more and more #ifdef MYSQLND inside their source...
> I was thinking the same way libxml2 API is provided inside ext/libxml
> instead of through each PHP XML extension API.

A fourth MySQL API is not my goal. There's enough confusion with three 
of them already.

Ulf

---

## #60275 — johannes.schlueter@oracle.com — 2012-04-24T12:41:00+00:00

URL: https://externals.io/message/60270#60275 · In reply to: #60272

> Aha, I like when an internals guy teach me internals stuff :)
> So, OK for 5.3 , what about supporting that for >=5.4 ?

Maybe. ;-)

> > […]
> 
> That was my first question : Is it better to provide mysqlnd API
> through all 3 MySQL extensions, or through a common mysqlnd_***() API
> ?
> I understand your thoughts, but I disagree as I think it would be much
> more clean to expose it via mysqlnd_***() API than through each MySQL
> ext (and for PDO that would be even more dirty). That would as well
> add more and more #ifdef MYSQLND inside their source...
> I was thinking the same way libxml2 API is provided inside ext/libxml
> instead of through each PHP XML extension API.

Well, as said, we couldn't before 5.4. I gave the "historical" reason
for the current state.

> > […]
> 
> ;-) OK, but that can be done easily. More and more people use
> pdo_mysql and could be able to not compile other MySQL extensions (I
> got some machines where only pdo_mysql is compiled in)

All distros I recently checked put all three mysql extensions in the
same package. People who build PHP themselves should be able to fix this
easily. So this does not seem to be an issue. (while that's irrelevant
for the mysqli vs. myslqnd question)

johannes

---

## #60278 — jpauli — 2012-04-24T13:32:46+00:00

URL: https://externals.io/message/60270#60278 · In reply to: #60274

> Am 24.04.2012 14:27, schrieb jpauli:
>
> > […]
>
>
> A fourth MySQL API is not my goal. There's enough confusion with three of
> them already.

I agree with that, but I still think having mysqli_***() function that
is only related to mysqlnd stuff is a bad idea.


Julien

---

## #60279 — jpauli — 2012-04-24T13:33:59+00:00

URL: https://externals.io/message/60270#60279 · In reply to: #60275

2012/4/24 Johannes Schlüter <johannes.schlueter@oracle.com>:
> 
> > […]
>
> Maybe. ;-)

I can start working on ideas/patches

>
> > […]
>
> Well, as said, we couldn't before 5.4. I gave the "historical" reason
> for the current state.

Got it, got it.

>
> > […]
>
> All distros I recently checked put all three mysql extensions in the
> same package. People who build PHP themselves should be able to fix this
> easily. So this does not seem to be an issue. (while that's irrelevant
> for the mysqli vs. myslqnd question)

Makes sense.


Julien

---

## #60285 — Andrey Hristov — 2012-04-24T17:08:41+00:00

URL: https://externals.io/message/60270#60285 · In reply to: #60278

Hi,

> 
> > […]
>
> I agree with that, but I still think having mysqli_***() function that
> is only related to mysqlnd stuff is a bad idea.
>

you probably hasn't seen `mysqli_stmt_get_result()` and result::fetch_all() :)

>
> Julien
>

Andrey