It was an intentional change made in scope of MDEV-13132. Does it cause any troubles?

Hi,

I think it will cause troubles to many peoples.

Doctrine ORM (PHP) just gone crazy to update a schema structure
SequelPro (Mac) see nullable field with default to "NULL" instead of NULL

That's tools I use everyday but I'm pretty sure that make troubles in many other tools

Yes, we're painfully aware of that

See MDEV-13132 — it was a genuine bug, because someone has misread the standard years ago. And in 10.2 the problem has become critical. MariaDB (and MySQL) violates the SQL standard and behaves differently from all other databases that we've tried (explained in MDEV-13132).

We've tried to reduce the damage by doing this change only in 10.2 (not in 10.1 or 10.0 or 5.5) and by keeping old behavior of SHOW COLUMNS (because it's non-standard, so there're no rules for it). Applications can switch to SHOW COLUMNS easily. Or adapt to use the new behavior, because old one cannot describe default values unambiguously.

Hi Sergei, make sense.

just released a P/R for doctrine: https://github.com/doctrine/dbal/pull/2825 if anyone face the same issue.

Hi Sergei,

I digged a bit into implications by fixing doctrine... One thing that I would like you to be aware:

> We've tried to reduce the damage by doing this change only in 10.2 (not in 10.1 or 10.0 or 5.5)

Two things:
1. In 10.2.6 (stable) it was working... yet I understand. But this fix have a lot of implications
2. Packaging issue, the official ppa set a version like '5.5.5-10.2.8-MariaDB-10.2.8+maria~xenial-log'. Which means that from PHP, we cannot trust the `mysqli->server_version` (returns 50505), but need to parse 'mysqli->server_info' with so many packaging differences (10.0.15-MariaDB-1~wheezy, ...). This is really painfull. I know this is not directly related, but if the packaging version pretend to be compatible with 5.5.5, it must be the case. Otherwise you should repackage with '10.2.8-MariaDB-10.2.8+maria~xenial-log' or '10.2.8-MariaDB+maria~xenial-log

Just to let you know... but mariadb will probably suffer from this.

What client library are you using? All mariadb client libraries strip "5.5.5-" prefix from the version (it was a hack to fix mariadb-mysql replication, where mysql code only looked at the first digit of the version, so 10.0 was below 3.23, and rejected as "too old").

Thanks for pointing out, I understand better now... But still it's not ideal...

On ubuntu xenial, the libmysqlclient package I have is libmysqlclient20 with proposed versions:

• 5.7.19-0ubuntu0.16.04.1 (from xenial update - this is what I have on my test system)
• 5.7.11-0ubuntu6 (from xenial)

See here.

I don't remember installing it (generally just using: "sudo apt install mariadb-server libmariadbclient-dev" with the 10.2 ppa). But I'm pretty sure it might have happened when I installed 'mysql-workbench-community'. Needs more tests to be sure. Anyway my php is using 'mysqlnd', see below:

FYI tested PHP versions are PHP "7.1.8-2+ubuntu16.04.1+deb.sury.org+4" and "7.2.0beta3" (both from ondrej ppa).

Important, for PHP7.1 the client is 'mysqlnd 5.0.12-dev - 20150407' (not dynamically linked to libmariadb-dev). Pretty usual.

Anyway, to be sure I attempted installing 'libmysqlclient18' (from mariadb -10.2.8+maria~xenial). As expected it does not help.

So if I understand, it would only work if the mysqli extension is linked to lib<maria>-client (not mysqlnd) ?

Not so simple, mmm.

Let me know, if there's something that could help.

PS:
1. BTW, "select version()" gives a nice '10.2.8-MariaDB-10.2.8+maria~xenial-log'. But I don't want to add an extra query to doctrine.
2. I guess the problem should be solved by using the libmysql-client from mariadb instead of mysqlnd. In another project I was seeing improvements over mysqlnd. Not only with speed but also detection of metadata: see 'here'. Anyway this kind of setup is pretty rare in my experience. Most use mysqlnd.

Right. mysqlnd doesn't strip the prefix.

There's a pull request, but it wasn't accepted — https://github.com/php/php-src/pull/1767

I see... so I've stripped 5.5.5 prefix on my side too, hope you don't increment later