You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
1.fromErrorHooks vs .map(fromErrorHook):
Initially there were only fromErrorHooks which were backwards compatible with feathers 4 error hooks - they all run in a single catch() call, so if one of them throws too then others do not run. I added fromErrorHook as a utility that person can use to get a single catch() hook. So if you do .map(fromErrorHook) then you get a bunch of catch() calls, not one as before.
I'm ok with this breaking change, but I assume that you are not and that it might have been overlooked. (If it was deliberate decision than all is good.)
2. Who should go first afterHooks or beforeHooks?
Small thing, but I think it is better for beforeHooks to run before afterHooks since there is no point to call an after hook if a before hook throws, it will just add noise to a trace. (I'm talking about this line.)
3.type should be required in runHook internal util:
It was optional for fromHooks and required for fromHook since there are no fromHooks now, type should be non-optional argument and if (type) can be simply removed. (This of course depends on the resolution for first point, if fromHooks will be back than this one needs no change.)
4. Propagation of bad practice with support for returning {...context}:
As I said in #1443 and #2462 supporting for returning new context object from a regular hook is a wrong choice (no upside beside backwards compatibility for obscure feature and a lot of downside). Now it tries to migrate from feathers into hooks. Additional example for why it is not right way: in a hook manager there is props() method with which you can set additional properties on a context, but props() does not simply Object.assign passed properties, it copies property descriptors, allowing for readonly and getters/setters things. Support for practice of {...context} goes against this, since all of those descriptors will be lost and will lead to bugs when those props get used before a cloned context is returned. (Thought about it when was looking how to migrate context.statusCode into context.http.statusCode with getters/setters.)
The text was updated successfully, but these errors were encountered:
1.
fromErrorHooks
vs.map(fromErrorHook)
:Initially there were only
fromErrorHooks
which were backwards compatible with feathers 4 error hooks - they all run in a singlecatch()
call, so if one of them throws too then others do not run. I addedfromErrorHook
as a utility that person can use to get a singlecatch()
hook. So if you do.map(fromErrorHook)
then you get a bunch ofcatch()
calls, not one as before.I'm ok with this breaking change, but I assume that you are not and that it might have been overlooked. (If it was deliberate decision than all is good.)
2. Who should go first
afterHooks
orbeforeHooks
?Small thing, but I think it is better for
beforeHooks
to run beforeafterHooks
since there is no point to call an after hook if a before hook throws, it will just add noise to a trace. (I'm talking about this line.)3.
type
should be required inrunHook
internal util:It was optional for
fromHooks
and required forfromHook
since there are nofromHooks
now,type
should be non-optional argument andif (type)
can be simply removed. (This of course depends on the resolution for first point, iffromHooks
will be back than this one needs no change.)4. Propagation of bad practice with support for returning
{...context}
:As I said in #1443 and #2462 supporting for returning new context object from a regular hook is a wrong choice (no upside beside backwards compatibility for obscure feature and a lot of downside). Now it tries to migrate from
feathers
intohooks
. Additional example for why it is not right way: in a hook manager there isprops()
method with which you can set additional properties on a context, butprops()
does not simplyObject.assign
passed properties, it copies property descriptors, allowing for readonly and getters/setters things. Support for practice of{...context}
goes against this, since all of those descriptors will be lost and will lead to bugs when those props get used before a cloned context is returned. (Thought about it when was looking how to migratecontext.statusCode
intocontext.http.statusCode
with getters/setters.)The text was updated successfully, but these errors were encountered: