Closed austinjp closed 6 months ago
Hi @austinjp . Thank you for the clarification! I think it makes sense to add something like this.
I'm a bit unsure about the API design, however. There are three main approaches I see:
Send
return a Message
, like you've proposed in #41.Send
return an ID
.Send
accept a pointer to a message instead, where the ID can be updated if the send succeeds.I'll have to think about what approach I like the most. The last one is arguably the least idiomatic, but I kinda like it anyway. Hmm. I'll get back to this.
Excellent, and no worries about taking time! I'm using my hacked version for now.
If it's of interest, my preference might also be your third option. When hacking on your code I initially returned an ID, then thought I'd prefer the Message
instead in case of subsequent changes to the DB/schema/etc. Perhaps a Returning
type might be useful, in addition to Message
?
I'm going to close PR #41 because it contains a bug: the 'common' dir is inside the 'internal' dir, meaning the 'common' items are unavailable to code using my PR. I've fixed this in the latest commit to my fork.
@austinjp I'm trying out a change in #43. Will think about it a bit. I'm worried the pointer design would be too magic (non-obvious) for my taste.
I'm worried the pointer design would be too magic (non-obvious) for my taste.
Yeah, I can see that. My thinking about returning an object (rather than an ID) was that it would hide the database implementation details to some degree, and both the object and the database schema could be extended or updated in the future without breaking changes (hopefully).
Okay, the pointer solution is in #44. Which one is the most intuitive to the caller?
Notes:
Send
, but the ID update feels a bit too magicID
's around in Extend
/Delete
, but will often not be used.I'm worried the pointer design would be too magic (non-obvious) for my taste.
Yeah, I can see that. My thinking about returning an object (rather than an ID) was that it would hide the database implementation details to some degree, and both the object and the database schema could be extended or updated in the future without breaking changes (hopefully).
The ID
is already used elsewhere. It's just an opaque string and not something like a UUID anyway, so I think that's fine.
Which one is the most intuitive to the caller?
In all frankness, I may not be qualified to make that call :nerd_face: Coming from Python, I have little experience with pointers so #43 feels intuitive to me. But someone with experience of Go may prefer #44, so I have to plead ignorance :shrug:
Whichever you prefer! As long as I can meet my use-case, I'm happy :innocent:
Your responsiveness is much appreciated, by the way.
Just a quick update: I haven't forgotten about this, I just can't decide. 😅
Just a quick update: I haven't forgotten about this, I just can't decide. 😅
No worries my dude, no rush. I haven't come to any firm conclusions myself :thinking:. I guess, in #43 the id, err := q.Send(...)
approach looks like idiomatic Go to me, while the pointer does seem a tad magical :mage_man: as you say. Coming from Python, I'm a believer that explicit is better than implicit so on balance maybe I prefer #43.
Yeah, I like explicit better as well. I'm just a bit sad that in the arguably main send-and-forget approach, there's additional clutter on the client side, because the id
won't be used in the common case: _, err := q.Send(…)
.
Maybe a SendAndGetThatIDBack
would work better? :D
Indeed!
Hmm. Public functions could be something like err := q.Send(…)
and id, err := q.SendReturningID(…)
. This is nice because AFAICT this wouldn't be a breaking change to your API.
Internally, code duplication might be minimised by calling some 'private' third function e.g. q.SendOptionallyReturningID(…)
which might use structs for func options and return values which would/wouldn't have an ID field as appropriate.
That way:
q.Send(…)
and q.SendReturningID(…)
is minimalAt this point, though, it's becoming about how best to hide clutter, and from whom! :laughing:
@austinjp third iteration in #45. :D
I like it. Looks cleanest to me in terms of client-side clutter, API consistency, impact on code-base, etc.
@austinjp I renamed the method SendAndGetID
, but otherwise I've stuck to the implementation you saw last week. Released in https://github.com/maragudk/goqite/releases/tag/v0.2.3 ! Thank you for all the feedback. 😊
@markuswustenberg Fantastic, thanks for all your efforts! :clinking_glasses:
Hi there. With reference to PR #41, it would be great if goqite could handle a particular use case I have.
I'm building a REST-ish API. Users submit jobs which are long-running (minutes/hours). Upon submission the API immediately responds with a '202 Accepted', and jobs are added to a goqite queue. According to received wisdom, the response sent to the user should include a 'Location' header with a URL where the user can check job progress, e.g.
Location /task/123
where123
is the ID of the job.However, since the
Send
method ofgoqite.Queue
does not return anything, I cannot reliably discern the ID of the item the user submitted. Therefore I can't build the/task/123
URL, and the user can't check progress.I guess I could embed some identifying information into the job body, but this would need to be generated by the client whereas that's clearly a server-side responsibility. I'd also need to parse many large BLOBs in the database to look-up any item, which would be inefficient.
Consequently, I believe it would be useful if
Send
could return the ID of the row it inserts. This is trivial in SQL with areturning
clause, supported in SQLite since version 3.35.0 (2021-03-12).PR #41 provides this functionality. However, implementing it resulted in a cascade of changes throughout the code-base, which I appreciate isn't ideal. Happy to discuss alternatives! :smile: