[debian-mysql] mariadb-10.5 builds regressed

[Daniel Black]

On Wed, Oct 7, 2020 at 7:55 AM Helmut Grohne <helmut at subdivi.de> wrote:
>
> Hi Otto,
>
> On Tue, Oct 06, 2020 at 06:14:48PM +0300, Otto Kekäläinen wrote:
> > Build status: https://buildd.debian.org/status/package.php?p=mariadb-10.5
>
> All of the builds that failed thus far miss -latomic on some linker
> invocation. This applies at least to armel, mipsel, m68k and riscv64.

> > Most likely one of these caused it:
> > - https://salsa.debian.org/mariadb-team/mariadb-10.5/-/commit/061acd336f7cdaa16ff0271feafce4a2551ab903
>
> This one does things about atomics.

Seems the removal of the rocksdb parts in the patch was a mistake.

I'd remove those parts of the patch under storage/rocksdb

unittest/mysys/./unittest/mysys/my_atomic-t.c

I need a solution to this today
will probably be a form of
https://github.com/MariaDB/server/pull/979/files#diff-d4fd018cd9bae50dc4204b5ce6f75752


> > - https://salsa.debian.org/mariadb-team/mariadb-10.5/-/commit/2920d282168f0986430f2feb65273af04ecb2240
>
> This one is very unlikely, because the changed code is conditional to
> x86.

It needs sse4.2

ref: https://jira.mariadb.org/browse/MDEV-20386?focusedCommentId=167857&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-167857


> > - https://salsa.debian.org/mariadb-team/mariadb-10.5/-/commit/f0ba31e1565feaca9c2a6f2534e55bfc29af7021
>
> This patch is still broken due to refactoring. After my remark, you only
> added the libedit-dev build dependency back. libgnutls28-dev and
> libncurses5-dev are still missing and therefore mariadb-10.5 still fails
> to cross build.
>
> Once you fix the ftcbfs, I guess that you can reproduce the native armel
> ftbfs using a cross build from x86. Doing so allows you to figure out
> which patch is causing the failure by temporarily reverting each in
> local builds. It'd start with the first one about atomics.
>
> Helmut
>