switch to destructors

[ringabout]

legacy modules which are too important to change(need more think) or they should be deferred
maybe we a new IO modules

• File Handle
• Socket Handle
• Streams

Less important modules
provide a wrapper for them with destructors

• Lock/RLock(reference: https://github.com/xflywind/concurrent/blob/main/mutexes.nim)
• Cond
• Threads

New modules
easy to add

• TempFile

Unstable modules
can we remove deinit**** procs and use destructors?

• sharedlists
• sharedtables

[timotheecour]

also:

• add dynlib.LibHandle to the list?
• some destructor issues (that shouldn't be too hard to fix) should maybe be fixed before, or at least should be re-prioritized, eg destructor for seq[T] elements not called with gc:refc upon setLen 0 nim-lang/Nim#17186, regression(0.20) destructor not called in VM nim-lang/Nim#11612
• destructors are only for objects, not ptr|pointer|ref, so for things like handles (file, socket, stream, dynlib handles etc), we'd have 2 options:

option 1

a wrapper object eg:

type FileObj* = object
  handle: File
proc openFile*(path: string): FileObj = FileObj(handle: open(path))
proc `=destroy`(a: var FileObj) =
  if a.handle != nil: close(a.handle)

• we also should allow UniquePtr/ShadredPtr to pass ownership of those wrapper objects around and make sure it works in various settings
• and we should make sure this has 0 impact on performance compared to using handles directly and doing manual management of init+deinit (eg open close)

option 2

extend the rules to allow destructors for types other than objects; maybe only allowing it for distinct types?

I haven't thought of implications of this though, this potentially cannot work

[ringabout]

Latest experiment is here: nim-lang#18640