arc: cpp exceptions don't work

[cooldome]

Compile as nim cpp --gc:arc $1 to get a crash at runtime.

proc myproc = 
  raise newException(ValueError, "Error")

try:
  myproc()
except ValueError as ex:
  echo ex[]

Crashes with stacktrace:

Traceback (most recent call last)
C:\Nim\lib\system\excpt.nim(73) x2
C:\Nim\lib\system\refs_v2.nim(120) nimDestroyAndDispose
SIGSEGV: Illegal storage access. (Attempt to read from nil?)

Works with c backend but not with cpp.

[cooldome]

@Araq,
As far as I got it PR #13065 is supposed to fix this problem.
But looks like the progress has stalled, any serious blockers found a long they way?

[Araq]

The only remaining problem is that tests/iter/tyieldintry.nim fails in my PR and I don't know why. Any help is welcome. Supporting C++ codegen well is more expensive than it looked a couple of years ago, I'm rethinking it. I wanted C++ for the better exception handling but my goto-based exceptions offer reasonable performance so my original motivation to compile to C++ is lost.

[andreaferretti]

my original motivation to compile to C++ is lost.

I guess an important motivation to compile to C++ is to be able to use C++ libraries without C bindings?

[Araq]

@cooldome As a compromise I could merge the PR now disabling the now failing test and later investigate so that gc:arc with nim cpp works better than before.

[cooldome]

@Araq, that would be fantastic solution solution. Much appreciated.
Agree with Andrea Ferretti, major reason for using cpp backend for me is hard requirement to use cpp libraries. I have to import their templates and use their library specific smart pointers and catch their exceptions. Yack, but I have to do it, no choice or possible workarounds this for me.

[Araq]

Yack, but I have to do it, no choice or possible workarounds this for me.

I can imagine. In the long run I hope to have a much better "cpp2nim" tool that wraps the C++ code into DLLs that we then call from Nim, but how to deal with template instantiations is a huge problem. So there might no other way out but to support "nim cpp" for forever. Maybe with a rewrite of the backend things become easier.

[cooldome]

PR #13065 is submitted closing