Message of the day:
Anything GraphQL and SilverStripe related.
likes GraphQL. The same API, but not the hassle of REST or SOAP
If the tests are green and @tractorcow thinks it's wrapped up now, I should merge it and tag a new version indeed
Time things. Wibbly wobbly timey wimey things prevented me from mergirg
Not much, thanks for the reminder 🙂
https://github.com/Firesphere|@Firesphere would love to discuss this with you before merging; See you on slack maybe?
Overview of changes in general order of most significant to least
• Return type of all validate / create mutations has been consolidated and normalised. The standard response is now a
MemberToken type with these fields:
• Member (the member)
• Token (string token)
• Valid (bool valid flag)
• Status (enum status code of the token)
• Code (int HTTP code that matches the status)
• Message (string message of humanised status code)
• Multiple authenticator tokens per user; Allows multiple devices to have their own tokens.
• Anonymous authentication is now shifted up into a custom Authenticator class; Support injectable authenticators into createToken mutation. Removed
• When logging in anonymously, create a
real member object to better integrate with silverstripe security logging / auditing practices.
• Bump minimum dependency to php 7.1: strict_types are everywhere too.
• Updated tests and documentation of course
• Support multiple schemas (graphql 3.0 and above).
• Relative path supported for JWT tokens
• Updated renewal process to support separate token / refresh expiration dates
• I gave every file my Love. Green ticks everywhere.
https://github.com/Firesphere|@Firesphere if you merge, I suggest branching master -> 1 first, as this is a new 2.x major change.Hide attachment content
I have recently run into a use case where mid-development of a feature I made a pull from the upstream repository.
Unbeknownst to me at the time some GraphQL fields had been added to a query, however being that this was defined in
yaml things were destined not to work until a
flush had taken place.
As I was not aware I simply loaded the CMS without a flush... and nothing happened.
Literally nothing. Half of it loaded (HTML templates), but everything depending on a GraphQL query as the source of data just hung with a loading screen. There was no hint of an error besides a great lack of data. My console was quiet of errors, every XHR response was
200 OK, and there were no UI updates to notify me of an issue.
Luckily a colleague informed me of the alteration, so I was quickly back on track, but if he had not been available at the time I would have been left to laboriously run through each XHR and check the output matched my expectations manually.
To this I see a few different approaches to solving this problem:
HTTP response codes
As I had asked the server for something it didn't know about, the problem lies with me and my request. I feel like there should be something in the 4xx range (like
422) that is a far more appropriate response.
However with rendering a 4xx response this resolves with most handlers as an error.
This could stop our UI from updating in the bad case... an argument against HTTP error codes - in which I make a counter case we should still update it to let the user (developer - I'm in
dev mode) know.
In a good case we could still render what data we did receive (if any) with a caveat that some is potentially missing.
In any case with the
admin module I think there's still plenty of space to render a global toast style notification that an issue has happened with our data request no matter the UI or section that is open at the time.
tl;dr there needs to be at least something to let a developer know that things have gone awry.
Something more than a hung screen and a message buried deep within an "everything's fine" response.