Build linux/amd64, linux/arm64, and linux/arm/v7 (arm 32-bit) targets
all at once.
This still requires Gradle and JDK on the local machine (i.e. outside of
the Docker build container), because it seems pointless to build the
.jar 3 times, once for each architecture, but tags all the images with
the same tag, rather than using a tag for each architecture. This allows
clients to all use the same image tag, but pull down the architecture
that's right for them.
I'd like to have it build the .jar file in a container (so the host
doesn't need JDK and Gradle), but couldn't think how to do that
efficiently (i.e. only once), with this approach.
Update Dockerfile to use a multi-stage build and run Gradle *within* the
builder container, rather than on the host, so the host needn't have
Gradle or JDK installed, then copy the build .jar from the builder
container into the final container.
Update build.gradle's dockerBuild target to not build the jar, but
instead do it within the Docker build process. This results in some
double-handling because it requires Gradle and JDK both on the host
system *and* within the container, and runs two instances of Gradle
during the build, but is added for backwards compatibility. The better
approach (rather than running `./gradlew dockerBuild`) is to manually
run `docker build -t ma1ua/ma1sd .`.
I've tested this process on both a Debian amd64 system and a Raspberry
Pi 3 running Raspbian, and it seems to work (though the RPi is pretty
slow).
In AD forest samAccountName (or uid) may not be unique in the
entire forest and userPrincipalName contains "@" symbol
disallowed in Matrix User Identifiers.
This commit reflects changes in ldap_auth_provider that adds
mxid generation logic for Active Directory.
Signed-off-by: Yuri Konotopov <ykonotopov@gnome.org>
These changes complement #11 where locally saved e-mail address can be "name.surname@example.com", but e-mail address in LDAP can be "Name.Surname@example.com". They are treated as two different e-mail addresses and user gets 2 invitation notification e-mails. We change ThreePid model's address property to convert all info to lowercase and [be915ae](be915aed94) can do it's job better.
The downside of this is that all medium addresses get converted to lowercase, not only e-mails. For now I can't think of any examples where medium values need to stay case sensitive.
Recently Synapse dropped unused (by Synapse itself) table "room_names" which brakes room name retrieval for ma1sd. There is a table "room_stats_state" from which we can retrieve room name by it's ID. Note that people to people conversations do not contain room names, because they are generated on-the-fly by setting other participants names separated by word "and". That's why this query will only get names for rooms where room names are set during creation process (or changed later) and are the same for all participants.
Link to Synapse code where it drops "room_names" table: https://github.com/matrix-org/synapse/blob/master/synapse/storage/data_stores/main/schema/delta/56/drop_unused_event_tables.sql#L17
I use LDAP for user store. I have set up "mail" and "otherMailbox" as threepid email attributes. When people get invited to rooms, they receive 2 (sometimes 3) invitation e-mails if they have the same e-mail address in LDAP "mail" and "otherMailbox" fields. I think it's a good idea to check identity store for duplicates before sending invitation e-mails.
Disable v2 by default.
Add migration to fix the accepted table (due to sqlite unable to change constraint, drop table and create again).
Fix displaying the expiration period of the new token.
Remove duplicated code.
Use v1 single lookup when receive the request with `none` algorithm and the only one argument.
Hide v2 endpoint if v2 API disabled.
