fixes #25306; Dangling pointers in stack traces with -d:nimStackTraceOverride

[ringabout]

fixes #25306

type
  StackTraceEntry* = object ## In debug mode exceptions store the stack trace that led
                            ## to them. A `StackTraceEntry` is a single entry of the
                            ## stack trace.
    procname*: cstring      ## Name of the proc that is currently executing.
    line*: int              ## Line number of the proc that is currently executing.
    filename*: cstring      ## Filename of the proc that is currently executing.
    when NimStackTraceMsgs:
      frameMsg*: string     ## When a stacktrace is generated in a given frame and
                            ## rendered at a later time, we should ensure the stacktrace
                            ## data isn't invalidated; any pointer into PFrame is
                            ## subject to being invalidated so shouldn't be stored.
    when defined(nimStackTraceOverride):
      programCounter*: uint ## Program counter - will be used to get the rest of the info,
                            ## when `$` is called on this type. We can't use
                            ## "cuintptr_t" in here.
      procnameStr*, filenameStr*: string ## GC-ed alternatives to "procname" and "filename"

[arnetheduck]

maybe add =copy...{.error.} somewhere?

[ringabout]

=copy...{.error.}

I can add it for nimStackTraceOverride . Though I hope it won't break something

[arnetheduck]

I can add it for nimStackTraceOverride . Though I hope it won't break something

I think there's more value adding it always to remove as many differences as possible between override and non-override mode

[ringabout]

That would potentially break external libraries that use StackTraceEntry.

type
  StackTraceEntry* = object
    when defined(nimStackTraceOverride):
      programCounter*: uint ## Program counter - will be used to get the rest of the info,
                            ## when `$` is called on this type. We can't use
                            ## "cuintptr_t" in here.
      procnameStr*, filenameStr*: string ## GC-ed alternatives to "procname" and "filename"
    else:
      procname*: cstring      ## Name of the proc that is currently executing.
      filename*: cstring      ## Filename of the proc that is currently executing.

when defined(nimStackTraceOverride):
  proc `procname`*(s: StackTraceEntry): cstring =
    s.procnameStr.cstring

  proc `filename`*(s: StackTraceEntry): cstring =
    s.filenameStr.cstring

Ideally, we only provide accessors for procname and filename for nimStackTraceOverride. The differences over filename and procname with/without nimStackTraceOverride already makes these options diverge

[arnetheduck]

only provide accessors

come to think of it, this is tricky because in current versions, libbacktrace has to set these values to a valid pointer as part of constructing the trace entry.

another way I thought of was to implement an actual =copy implementation so the compiler would inject copies in the right places - that copy operator would do the necessary pointer adjustments - not sure if that's possible at this early stage of the nim compilation pipeline though

[Araq]

You don't have to copy anything here, all you need to do it to patch every entry after an add operation.

[arnetheduck]

strictly speaking, in this particular case, one would want to move the existing entries...

[ringabout]

add operation triggers a copy, which is marked {.error.}. Perhaps I need to use nodestroy or something?