* made a few changes in order to further simplify the setup for
developers of the Groupware backend
* add STALWART_DOMAIN to deployments/examples/opencloud_full/.env
* adapt the Stalwart configuration file to not set server.hostname and,
instead, pick it up from /etc/hostname, which is set by Docker
Compose as we can use default values for STALWART_DOMAIN there, in an
analogous fashion to the other containers in that project
* add config/keycloak/clients/groupware.json to avoid requiring manual
configuration of Keycloak via the admin web UI
* Stalwart container:
- listen for SMTPS on :1465
- remove the stalwart-logs volume, not needed (logs are going to
stdout)
* updated services/groupware/DEVELOPER.md:
- refer to a variable OCDIR to make instructions more copy-pasteable
- remove manual Keycloak configuration section as it is now obsolete,
replaced by provisioning a configuration file instead
Minor: be more Go idiomatic: just use a function to pick the attachment
from an Email's attachment list instead of using an interface with
multiple iplementation structs.
* the JMAP error handling was not working properly, fixed it and added
error definitions accordingly
* add operations to retrieve mailbox roles and mailboxes by role for
all accounts
There was really no reason to go with "Messages" as far as the
vocabulary of the Groupware API goes, since the objects those APIs serve
are "Emails", to stick with the wording of the JMAP specification.
* add URL to retrieve all the mailboxes for all the accounts of a user,
as a first use-case for an all-accounts operation, as
/accounts/all/mailboxes
* add URL to retrieve mailbox changes for all the mailboxes of all the
accounts of a user, as a first use-case for an all-accounts
operation, as /accounts/all/mailboxes/changes
* change the defaultAccountId from '*' to '_', as '*' rather indicates
"all" than "default", and we might want to use that for "all
accounts" operations in the future
* refactor(groupware): remove the accountId parameter from the logger()
function, as it is not used anyways, but also confusing for
operations that support multiple account ids
* refactor some pkg/jmap and groupware methods to make more sense from
an API point-of-view
* add path parameter documentation, but automate it by injecting their
definition into the OpenAPI YAML tree that is extracted from the
source code using go-swagger as it is too cumbersome, repetitive and
error-prine to document them in the source code; wrote a TypeScript
file apidoc-process.ts to do so
* add generating an offline HTML file for the OpenAPI documentation
using redocly, and injecting a favicon into the resulting HTML; wrote
a TypeScript file apidoc-postprocess-html.ts to do so
* move the logging of the username and session state away from pkg/jmap
and into services/groupware
* introduce more decoupling for the session cache, as well as moving
the implementation into groupware_session.go
* remove the baseurl from the JMAP client configuration, and pass it to
the session retrieval functions instead, as that is really the only
place where it is relevant, and we gain flexibility to discover that
session URL differently in the future without having to touch the
JMAP client
* move the default account identifier handling from the JMAP package to
the Groupware one, as it really has nothing to do with JMAP itself,
and is an opinionated feature of the Groupware REST API instead
* add an event listener interface for JMAP events to be more flexible
and universal, typically for metrics that are defined on the API
level that uses the JMAP client
* add errors for when default accounts cannot be determined
* split groupware_framework.go into groupware_framework.go,
groupware_request.go and groupware_response.go
* move the accountId logging into the Groupware level instead of JMAP
since it can also be relevant to other operations that might be
worthy of logging before the JMAP client is even invoked
* implement more metrics, in a more streamlined fashion
* use concurrent-map to store SSE streams instead of a regular map with
one big lock that will not scale when it grows, causing too much
contention on that one lock
* while testing error metrics, noticed a few bugs with error handling
when Stalwart is down: fixed
