On Wed, Nov 21, 2018 at 01:28:59AM -0500, George Koehler wrote:
> There's a bug in the OpenBSD port of devel/pcre, when built with
> ports-gcc, that prevents the build of databases/mariadb. The patch
> below seems to fix pcre, but there is another problem in mariadb,
> probably with big-endian platforms.
>
> mariadb detects the pcre bug, then rejects the system pcre and wants
> to use its internal pcre, but the port told mariadb to use the system
> pcre, so mariadb fails in the configure phase with the error,
> "system pcre is not found or unusable".
>
> The error appeared in the second-to-last sparc64 bulk (gcc 4.9):
> http://build-failures.rhaalovely.net//sparc64/2018-11-01/databases/mariadb.log
>
> (The last sparc64 bulk failed devel/cmake and didn't reach mariadb.)
>
> The same error appeared on my powerpc machine (gcc 6.4). The patch
> just copies the fix from mariadb's internal pcre to our system pcre.
> I built devel/pcre with the patch on amd64 and powerpc and ran
> `make test` on both platforms: all 5 tests passed.
>
> With this fix, my powerpc machine gets mariadb through the configure,
> build and fake phases, but fails in the package phase. I'm attaching
> the log, but in summary, it packages mariadb-client and then fails
> to package mariadb-server because 3 "mroonga" files don't exist.
> mariadb's WRKSRC/server/mroonga/CMakeLists.txt says, "Mroonga doesn't
> support on big-endian", but I have a big-endian powerpc. I don't have
> a fix for this problem.
Nice analysis ! It's not so uncommon to have arch-specific files in
packages, check the PFRAG.${ARCH} mechanism, check x11/vlc or lang/rust
for examples.
I'll put your patch for pcre in the next sparc64 bulk.
Landry
