Mirror of forum.nim-lang.org

3176 :: Providing writable string buffer to a foreign C-function

• olwi (orginal) [2017-09-18T17:23:13+02:00] view original

  Being new to nim, I try to use some external DLL written in C via nim's FFI. Several of this DLL API calls look roughly like this:

  int do_something(..., char *errMsg);
  Run

  (I replaced all the parameters that are not relevant to my question with ...)

  do_something() can succeed returning some non-negative retcode, or it can fail and then it returns negative retcode and puts text error message into errMsg buffer that the caller is supposed to provide. The message is known not to exceed 256 chars.

  Currently, I call this function like this:

  proc do_something(..., errMsg: cstring): cint {.importc, dynlib: "doSmth.dll".}
  
  var
    errBuf = newStringOfCap(256).cstring
    retCode = do_something(..., errBuf)
  if retCode < 0:
    echo "ERROR: ", $errBuf
  Run

  This works, but I'm not sure if this a proper way to do it.

  I'm not even sure that the anonymous newStringOfCap(256) string that backs my errBuf won't be garbage-collected from under my feet...

  Also, $errBuf makes string from errBuf performing a copy of the errBuf content. This is not that important in this particular case, but is it possible to make it into string without content copying?

  Maybe there are some better/more idiomatic alternatives to the above?

---

• Araq (orginal) [2017-09-21T17:25:47+02:00] view original

  Use something like:

  var
    errBuf = newStringOfCap(256)
    retCode = do_something(..., errBuf.cstring)
  if retCode < 0:
    # if the API doesn't tell us about the size ... but it really should!
    errBuf.setLen(len(cstring(errBuf))
    echo "ERROR: ", errBuf
  Run

---

• Udiknedormin (orginal) [2017-09-22T18:57:19+02:00] view original

  Maybe add a single proc to update a string if its cstring is modified separately? It would make this example much easier and the implementation seems easy assuming we cstring's memory is still within the string capacity. It seems to me it could be pretty nice to have such a proc in stdlib, actually.

---