++ed by:

5 PAUSE users
10 non-PAUSE users.

Tim Bunce


This is all fairly random.

double/tripple query execution on items! and with/without me. (SQL differs by prefix on pk.)

Things to do, or at least consider, before we stop being alpha:

    Explore using serializer and adaptor objects.

    Add some docs re choosing ActiveModel vs JSONAPI vs HAL even if just links
    Eg see http://discuss.emberjs.com/t/should-we-keep-using-ember-data/7405/10

    Add tests and docs for RouteMaker make_routes_for args and options

    TODOs for JSON API:
    - provide "links" separately to prefetch logic
    - support collection urls, e.g. /photos/1,2,3
    - support relationship urls, e.g. /photos/1/links/photographer
    - support include= in addition to prefetch=
    - support multiple sorts, e.g. sort[posts]=-created,title&sort[people]=name
    - support the "name[type]" style of field= param value
    - support creating resources via the application/vnd.api+json media type
    - support updating resources via the application/vnd.api+json media type
    - support updating relationships via the application/vnd.api+json media type

    Test string keys, including unicode
    Look into at least basic support for documentation links in HAL

    Tweak test schema so FK field names don't match the PKs
    Test using a non-pk indentity eg genre name
    Add tests for links being present, correct, and working

    Implement ::SetInvoke
    (Ponder /foo/invoke/bar/drill/down/in/results)

    * warn at build-time if a resource has attributes that clash with reserved keys
    eg id (that's not the unique key being used for id), type, href, or links.
    * add optional resultset-class to jsonapi-type name mapping
    * explore https://github.com/json-api/browser

    Document authentication and authorization
    Add eg _build_initial_search eg for 'show only records owned by current user'
    Mention https://metacpan.org/pod/DBIx::Class::Schema::RestrictWithObject

    Review authorization handling and explore alternatives.
    Document 'basic auth on internet needs https', see also:
    Also http://stackoverflow.com/questions/17530992/authentication-and-or-https-with-plack-psgi-poet-application/24808547#24808547

    Review code for DBIx::Class sanity and best practice

    Refactor param handling to abstract ~json both for requests *and* constructing
    link urls (where json data isn't fully handled at the moment, other than pass-thru).

    Any security issues with allowing arbitrary json in parameters?
    Should we allow use of relationships in conditions that aren't exposed as routes?
    (perhaps make that optional per class and default off, maybe allow via regex on relname)

    Allow (and prefer?) prefetch=me over prefetch=self?

    Adopt Log::Any for logging and debug?
    Note relevant env vars, eg WEBAPI_DBIC_DEBUG, PATH_ROUTER_DEBUG

    Review code and docs for XXX / TODO /TBD markers

    Better test coverage, especially for roles


    WebAPI::DBIC::Resource::Role::DBICException finish_request() figures out
    invalid fields by doing regex matching against database error messages.
    The current regex works for Pg and SQLite, probably won't work against other
    fREW has some code on L<https://github.com/frioux/DBIx-Exceptions|github>
    for matching the error messages. That plus L<DBIx::Introspector> could do the trick.
    (Another alternative would be to build a valid parameter list while we're
    introspecting the schema, but that wouldn't catch many other kinds of exceptions.)

Other items:

    Refactor transaction handling.

    Consider viability of reworking to be able to stream results using callbacks
    Would presumably require a streaming JSON encoder
    (or encode row-by-row and fake a closing "]" at the end)

    Better error reporting (in HAL format)
    Reconsider and/or polish HTTP::Throwable::Role::JSONBody
    Handle and report parameter validation and db constraint errors in a way that
    allows sufficient information to be returned to the caller e.g., to mark which
    fields of a web form hold invalid values.

    Review Moo usage for best practice
    especially Roles and before/after/around

    Profiling, especially re the cost of constructing relationship links
    for each item. Much work could be cached to make it much cheaper.

    improve partial response support for prefetch

    Consider how to handle multi-field primary keys and foreign keys

    Reconsider using or removing unused code from ::Util
    (I think only create_header is used at the moment)

    use HAL CURIE prefix in link keys?

    make it easy to limit fields returned by default so new apps can adopt that
    as best practice - add to the code on demand, so we know the
    fields are actually being used and the inverse, that we can remove/change a
    field that's not being exposed and know that no users of the api will be affected.
    Perhaps allow DBIC resultset classes to define what default set of fields to use.
    Note effect on caching.

    Support multiple $schema's? With cross-schema linking?

    Consider async?

    Look for further ideas in

HTTP Caching

Explore of HAL item caching using ESI or SSI


The cached urls would need to include extra params like fields=...

So it's possibly better to use a surrogate key approach:


which would need some support in WebAPI::DBIC to pass some kind of cache key through the code to include in a response header.

HAL Browser

To update:

    git subtree pull --squash --prefix hal-browser git@github.com:mikekelly/hal-browser.git master

Other Test Schemas

See the run-chinook-demo.sh script for a quick way to run the code with the Chinook sample database from http://chinookdatabase.codeplex.com