Mirror of forum.nim-lang.org

4489 :: Using var object in a proc that is the object's property

• lqdev (orginal) [2018-12-20T19:13:11+01:00] view original

  Hello,

  I'm trying to do something like this:

  type
    Events = object
      onload, onfail: proc (name: string)
  
  type
    SomeWrapper = object
      events: Events
      onload, onfail: proc (name: string)
  
  proc registerEvents(wrapper: var SomeWrapper) =
    wrapper.events.onload = proc (name: string) = echo "loaded " & name; wrapper.onload(name)
    wrapper.events.onfail = proc (name: string) = echo "failed to load " & name; wrapper.onfail(name)
  Run

  But the following error is thrown:

  
  $ nim c mutability.nim
  Hint: used config file '/home/daknus/.choosenim/toolchains/nim-0.19.0/config/nim.cfg' [Conf]
  Hint: system [Processing]
  Hint: mutability [Processing]
  mutability.nim(11, 65) Error: illegal capture 'wrapper' of type <var SomeWrapper> which is declared here: mutability.nim(10, 20)
  Run

  So far I understand the problem (when wrapper is destroyed, events may still persist as a reference somewhere, and then the implementations of events.onload and events.onfail would access illegal memory, and cause a segmentation fault, which in my real use case is very unlikely as the lifetime of the hypothetical events never exceeds the lifetime of the wrapper), but I don't know how to fix it.

---

• Araq (orginal) [2018-12-20T19:33:45+01:00] view original

  proc registerEvents(wrapper: ptr SomeWrapper) =
    wrapper.events.onload = proc (name: string) = echo "loaded " & name; wrapper[].onload(name)
    wrapper.events.onfail = proc (name: string) = echo "failed to load " & name; wrapper[].onfail(name)
  Run

  I'm not saying it's the best solution, but it works.

---

• lqdev (orginal) [2018-12-20T19:49:59+01:00] view original

  How do I pass the ptr SomeWrapper to registerEvents? I never used pointers in Nim before.

---

• 2vg (orginal) [2018-12-20T22:25:07+01:00] view original

  using addr.

  var someObj = SomeWrapper()
  
  registerEvents(addr someObj)
  Run

  why not using ref object? performance?

---

• lqdev (orginal) [2018-12-20T22:53:24+01:00] view original

  Oh wait, it's a pointer! Mistook it as a reference (those are what I never used, I use pointers on a daily basis because of OpenGL).

---

• 2vg (orginal) [2018-12-21T02:30:05+01:00] view original

  if using ref object,

  type
    Events = object
      onload, onfail: proc (name: string)
  
  type
    SomeWrapper = ref object
      events: Events
      onload, onfail: proc (name: string)
  
  proc registerEvents(wrapper: SomeWrapper) =
    wrapper.events.onload = proc (name: string) = echo "loaded " & name; wrapper.onload(name)
    wrapper.events.onfail = proc (name: string) = echo "failed to load " & name; wrapper.onfail(name)
  
  var s = SomeWrapper()
  registerEvents(s) # work
  Run

---

• moerm (orginal) [2018-12-21T10:04:20+01:00] view original

  No, it doesn't. It just seems it does but as soon as a call like s.events.onload("some event" is made it segfaults with Illegal storage access. (Attempt to read from nil?)

  Please note that the approach of @Araq also fails because in the OP wrapper.onload was left nil which leads to the segfault.

  Adding something like wrapper.onload = proc (name: string) = echo "wrapper loaded & name (plus onfail()) solves the problem for both the ptr and the ref approach.

---

• 2vg (orginal) [2018-12-21T15:53:52+01:00] view original

  proc registerEvents(wrapper: var SomeWrapper) =
    wrapper.events.onload =
      proc (name: string) =
        echo "loaded " & name
        wrapper.onload(name)
  Run

  why call onload recursively?

---

• 2vg (orginal) [2018-12-21T15:57:51+01:00] view original

  ah sorry, missed

---

• moerm (orginal) [2018-12-21T16:09:47+01:00] view original

  P.S. @Araq and team

  If it's not too much trouble it might be a good thing to differentiate between illegal memory access and calling a nil proc. Would probably be helpful and errors like the above would be easier to see/understand for new Nim users.

---

• twetzel59 (orginal) [2018-12-21T16:21:34+01:00] view original

  That would be nice to have. It would perhaps make sense to include it only in debug mode, as I believe it would require a branch for a nil check before each call, though. The Segfault isn't caused by userland checking for NULL pointers, but rather by the protected memory system getting angry that you're trying to access a low memory address (0).

  (Please correct me if I'm wrong. I'm sorta new to low-level stuff).

---