Do not re-set iconv converter state when the input encoding may use

byte-order from a BOM, because some iconv implementations (including
win_iconv, Apple libiconv) forget the byte order on re-set. Partially
reverts 85476.


git-svn-id: https://svn.r-project.org/R/trunk@87222 00db46b3-68df-0310-9c12-caf00c1e9a41

src/gnuwin32/system.c

@@ -460,8 +460,12 @@ FileReadConsole(const char *prompt, unsigned char *buf, int len, int addhistory)
  err = (res == (size_t)(-1));
  /* errors lead to part of the input line being ignored */
  if(err) {
+ /* Should re-set with a stateful encoding, but some iconv
+ implementations forget byte-order learned from BOM. 
+
  Riconv(cd, NULL, NULL, &ob, &onb);
  *ob = '\0';
+ */
  printf(_("<ERROR: re-encoding failure from encoding '%s'>\n"),
  R_StdinEnc);
  }

src/main/sysutils.c

@@ -793,11 +793,16 @@ attribute_hidden SEXP do_iconv(SEXP call, SEXP op, SEXP args, SEXP env)
  /* it seems this gets thrown for non-convertible input too */
  /* EINVAL returned for invalid input on macOS with system
  libiconv */
+ /*
+ Should re-set with a stateful encoding, but some iconv
+ implementations forget byte-order learned from BOM.
+
  res = Riconv(obj, NULL, NULL, &outbuf, &outb); 
  if (res == -1 && errno == E2BIG) {
  R_AllocStringBuffer(2*cbuff.bufsize, &cbuff);
  goto top_of_loop;
- } 
+ }
+ */
  if(fromUTF8 && streql(sub, "Unicode")) {
  if(outb < 13) {
  R_AllocStringBuffer(2*cbuff.bufsize, &cbuff);

src/unix/sys-std.c

@@ -1025,8 +1025,12 @@ Rstd_ReadConsole(const char *prompt, unsigned char *buf, int len,
  err = res == (size_t)(-1);
  /* errors lead to part of the input line being ignored */
  if(err) {
+ /* Should re-set with a stateful encoding, but some iconv
+ implementations forget byte-order learned from BOM.
+
  Riconv(cd, NULL, NULL, &ob, &onb);
  *ob = '\0';
+ */
  printf(_("<ERROR: re-encoding failure from encoding '%s'>\n"),
  R_StdinEnc);
  strncpy((char *)buf, obuf, len);