On 2/14/2018 9:35 AM, Francesco Riosa wrote: > > On 14/02/2018 05:37, Robin H. Johnson wrote: >> On Tue, Feb 13, 2018 at 09:32:32PM -0500, Brian Evans wrote: >>> I have a plan I would like some eyes on... >>> >>> I want to gradually *BAN* the use of virtual/mysql and >>> virtual/libmysqlclient as dependencies. >> Overall I agree, but there's some slight concerns I have. >> >>> To accomplish this, force dev-db/mysql-connector-c to be the only souce >>> of libmysqlclient.so. >>> >>> Packages that choose to support libmariadb.so instead can include a >>> libmariadb USE to hook up to dev-db/mariadb-connector-c that will be >>> introduced (and they can live side-by-side). The motivation for this >>> could be licensing with libmariadb being LGPL instead of GPL. This is >>> similar to ffmpeg/libav, except the libraries can co-exist. >> Have all the concerns about using slightly different libmysqlclient.so >> builds been resolved? Esp for pre-built binaries (I don't know if there >> are any left in the tree). >> >>> The current providers of virtual/mysql would get a new USE flag that is >>> MASKED for all users for the transition period and pull in the lib >>> package(s) when that USE is disabled. >>> >>> virtual/mysql would become a server reference for USERS only. It would >>> be a QA warning violation to depend directly on virtual/mysql as it can >>> live anywhere. >> This part worries me slightly. I do understand that mysql-embedded is >> retired entirely, but apps that spun up their own local mysqld instance >> would still be affected this this change. > > Checked a random desktop install and found these packages depending on > virtual/mysql: > > vivo@Monfi ~ $ equery d virtual/mysql > * These packages depend on virtual/mysql: > dev-db/mariadb-10.2.12 (server ? ~virtual/mysql-5.6[embedded=,static=]) > dev-libs/apr-util-1.6.1 (mysql ? =virtual/mysql-5*) > dev-libs/cyrus-sasl-2.1.26-r11 (mysql ? virtual/mysql) > dev-libs/redland-1.0.17-r1 (mysql ? virtual/mysql) > kde-apps/akonadi-17.12.2 (mysql ? virtual/mysql) > net-analyzer/net-snmp-5.7.3_p3 (mysql ? virtual/mysql) > net-mail/mailutils-3.4 (mysql ? virtual/mysql) > sci-mathematics/glpk-4.63 (mysql ? virtual/mysql) dev-libs/apr-util-1.6.1 (mysql ? dev-db/mysql-connector-c:=) dev-libs/cyrus-sasl-2.1.26-r11 (mysql ? dev-db/mysql-connector-c:=) dev-libs/redland-1.0.17-r1 (mysql ? dev-db/mysql-connector-c:=) kde-apps/akonadi-17.12.2 (mysql ? dev-db/mysql-connector-c:=) net-analyzer/net-snmp-5.7.3_p3 (mysql ? dev-db/mysql-connector-c:=) net-mail/mailutils-3.4 (mysql ? dev-db/mysql-connector-c:=) sci-mathematics/glpk-4.63 (mysql ? dev-db/mysql-connector-c:=)
>
> It would be interesting to know how their ${*DEPEND} should look after
> the change.
>
> Honestly I'd also like to understand the rationale better, why this
> change is needed?
> Are the implementations of Oracle mysql and Mariadb so different that
> it's not possible anymore to consider those equivalent (for what
> packages see)?
This is needed for a few reasons:
1) Virtuals cannot rebuild dependencies when there are subslot changes
2) dev-db/mysql-5.7 and dev-db/percona-server-5.7 are prevented from
entering the tree simply due to the client library SOVERSION change. It
is difficult for the user to swap between server versions.
3) libmariadb.so is a pthread_once library which many other packages are
not expecting. So they may have to write things two different ways
signature.asc
Description: OpenPGP digital signature
