RFC: Future[T] leak or unattended Future[T] tracking. about nim-chronos HOT 8 CLOSED

Capturing and carrying around all these string values is a bit expensive btw. I suggest the following optimisation:

We can create a reusable module for capturing "source locations". A good place for it would be std_shims.

type
  SourceLocation* = object
    file*: cstring
    line*: int

proc `$`*(loc: ptr SourceLocation): string =
  result.add loc.file
  result.add ":"
  result.add $loc.line

proc sourceLocationIMPL(file: static string, line: static int): ptr SourceLocation =
  var loc {.global.} = SourceLocation(file: cstring(file), line: line)
  return loc.addr

template getSourceLocation*: ptr SourceLocation =
  sourceLocationIMPL(instantiationInfo(-2).filename, instantiationInfo(-2).line)

It's used like this:

import
  source_locations

type
  Future[T] = object
    createdAt: ptr SourceLocation

proc newFuture(T: type, loc: ptr SourceLocation): Future[T] =
  result.createdAt = loc

template newFuture(T: type): auto =
  newFuture(T, getSourceLocation())

var f = newFuture(int)
echo f.createdAt

The main idea is that every usage of newFuture effectively allocates a single global variable associated with the line of code where newFuture was used. A pointer to this global variable is all that needs to be passed around at run-time to track the origin location. There is no run-time memory allocation, string copying and other more expensive operations.

from nim-chronos.

@zah its very nice idea, but i don't see any reasons to put this small part of code as dependency std_shims.

Also i'm curious how it will work for Future[T] objects which was created using macro code?

from nim-chronos.

The main benefit is that it can be used in other projects as well for similar purposes.

from nim-chronos.

With recent nimble problems i wish to avoid external dependencies, also i dont think 13LOCs deserve its own module name. But i will add your copyright because it nice idea :)

from nim-chronos.

could this be a finalizer for the future type? ie assert on future that's released without being actioned?

from nim-chronos.

@arnetheduck you mean with being read()?

from nim-chronos.

https://nim-lang.org/docs/system.html#new%2Ctypedesc - new takes a proc that is run whenever the gc gets to deallocating something. we could use that (probably in debug mode) to catch stragglers

from nim-chronos.

Looks like great minds think alike :)

I implemented this: nim-lang/Nim#11181

from nim-chronos.