scarlet ridge
Hi,
I'm learning how to use Fiber and I'm trying to find how to pass my SQL database object to my handlers. When I look it up I see a lot of very roundabout solutions for how to do it (wrap handlers in a closure, put your handlers in a struct, etc.). Like for example in Rust's axum you can pass it as an additional parameter in the handler function. I also saw you can just make your database a global variable but I was wondering if there's a proper clean way to do it? Thanks
ocean wave
the two ways you mentioned are fairly canonical, they’re not really roundabout
using methods is the slightly less flexible but more user-friendly strategy in most situations
ocean wave
i’ve seen both used in practice, they’re not that different and i won’t claim that one is definitely used more than the other
using functions returning a closure may be easier to test (since it requires a fixed number of specific parameters, rather than an entire receiver struct where it’s unclear which fields are used), but can be slightly more annoying to change in different places if you need to add a parameter
having access to an entire struct in many handlers may make code less annoying to write initially but leaves you without a strong contract for what the handler requires
scarlet ridge
Is there really not a built-in way to pass it? I've seen mentions of Ctx.Local but haven't seen much about it
ocean wave
you can use contexts, but that leaves you with a very weak contract for what the handler requires, without even a struct field
you may feel like it’s a reasonable trade off for convenience, but testing becomes a lot more fragile (you must set up the context in the same way in every test, without any help from the type system)
scarlet ridge
Might be cause I'm new to the language but this fairly essential thing seems quite overcomplicated lol
ocean wave
closures may introduce some complexity, but receivers for a method are quite core to the language
it’s composing the language’s features outside of the context of the HTTP library, rather than having the HTTP library (and then every other library) provide a one-off solution
pastel scarab
i do it with package level ("global") variables, though ymmv
pastel scarab
you may feel like it’s a reasonable trade off for convenience, but testing becom...
this is a fair compromise and so you should limit the connections you need to set up and what you don't
scarlet ridge
i do it with package level ("global") variables, though ymmv
Do the double quotes when you say global mean anything
Are there different types of globals
pastel scarab
well global is a bit ambiguous of a term
if it's still namespaced to a given package it's not exactly global
scarlet ridge
if it's still namespaced to a given package it's not exactly global
Go has namespaces..?
pastel scarab
a namespace is just a scope that owns / exports things with names
but no Go doesn't have a namespace keyword
bronze agate
i think using structs is the best way, but rather than passing the sql.db instance, is put the db ops in their own interfaces and that allows for proper testing
ive tried using contexts / locals before and it gets messy
i tend to have own struct rather than just one huge struct though so can tell what is actually be injected
pastel scarab
an interface is great for that application, yes.