This follows the style of wai-middleware-auth package and
makes the JWT parsing a middleware.
Co-authored-by: Wolfgang Walther <walther@technowledgy.de>
Getting the value for a header GUC on PostgreSQL v14 is done using `current_setting('request.headers')::json->>'name-of-header'` and in a similar way for `request.cookies` and `request.jwt.claims`
PostgreSQL versions below 14 can opt in to the new JSON GUCs by setting the `db-use-legacy-gucs` config option to false (true by default)
BREAKING CHANGE
Not gzipping the output more than doubles the throughput for a simple "do nothing" RPC function.
Postgrest's gzip support was surprisingly slow, lacked in configurability, was undocumented and untested.
It was possible to achieve this before by not sending `Accept-Encoding: gzip`, but that turned into a bit of an undocumented gotcha. Most browsers do send it and by default a proxy pass would forward this header. Better to not support gzip at all, further focusing on ensuring Postgrest is a minimal layer over Postgres.
Nginx, or whatever proxy, provides critical options such as min length, vary headers and so forth, and will likely have much better performance as well.
The request.spec GUC contains the schema cache structure in json.
It's only available when the root endpoint(/) is requested and when
db-root-spec is not empty.
Also correct db-root-spec to accept a schema.
* Use ExceptT to avoid 'staircasing' case analysis in App.hs
* Split large function in App.hs into individual handler functions
* Adapt API of Auth.hs, OpenApi.hs etc. to simplify the use of those modules in App.hs
* Split optional rollback functionality into Middleware
* Unify SimpleError and ApiRequestError into one Error type, so it can be used across modules
* secret-is-base64 -> jwt-secret-is-base64
* role-claim-key -> jwt-role-claim-key
* max-rows -> db-max-rows
* pre-request -> db-pre-request
* root-spec -> db-root-spec
* db-schema -> db-schemas
This is not a breaking change, because aliases are added as well.
refactor: sorted all config keys alphabetically where applicable
* remove configQuiet from Config
configQuiet was not an end user setting.
The logging setup is now an internal parameter.
* move proxy uri validation to Private dir
The good part is that protolude 0.3.0 builds with GHC 8.10.
The bad part is that this change is a bit painful:
- the default `toS` has changed to no longer convert to and from ByteString
- similarly, `show` no longer outputs ByteString
The changes here are pretty much minimal to keep things compiling; I didn't
see a nice way to work with the new ConvertText class, even though `toUtf8`
seems like it might help if used besides `toS` at just the right spots.
* Update config default; Copy non-defaults
* Update .stylish-haskell config version to match pgrst
* Apply stylish haskell to all files
* CircleCI config
* Remove redundant import.
What is used from Network.HTTP.Types.Headers is also exported by Network.HTTP.Types.
* Grouped imports
* Show un-styled files on CircleCI failure
* Fix styling imports
* Apply adhoc standard correctly
Fix#828, breaking change: computed columns now only work if they are on
the config schema.
Fix#835, tests now not depend on the search_path of the
postgrest_test_authenticator.
