More UTF-8 patches from Inaba Hiroto.

- The substr lval was still not okay.
- Now pp_stringify and sv_setsv copies source's UTF8 flag
  even if IN_BYTE.  pp_stringify is called from fold_constants
  at optimization phase and "\x{100}" was made SvUTF8_off under
  use bytes (the bytes pragma is for "byte semantics" and not
  for "do not produce UTF8 data")
- New `qu' operator to generate UTF8 string explicitly.
  Though I agree with the policy "0x00-0xff always produce bytes",
  sometimes want to such a string to be coded in UTF8.
  I can use pack"U0a*" but it requires more typing and has
  runtime overhead.
- Fix pp_regcomp bug uncovered by "0x00-0xff always produce bytes"
  change, the bug appears if a pm has PMdf_UTF8 flag but interpolated
  string is not UTF8_on and has char 0x80-0xff.

TODO: document and test qu.

p4raw-id: //depot/perl@8439

1 files changed, 6 insertions, 1 deletions

diff --git a/pp_ctl.c b/pp_ctl.c
index 07545dc28a..5490221d0b 100644
--- a/pp_ctl.c
+++ b/pp_ctl.c
@@ -116,9 +116,14 @@ PP(pp_regcomp)
 	    pm->op_pmflags = pm->op_pmpermflags;	/* reset case sensitivity */
 	    if (DO_UTF8(tmpstr))
 		pm->op_pmdynflags |= PMdf_DYN_UTF8;
-	    else
+	    else {
 		pm->op_pmdynflags &= ~PMdf_DYN_UTF8;
+		if (pm->op_pmdynflags & PMdf_UTF8)
+		    t = bytes_to_utf8(t, &len);
+	    }
 	    pm->op_pmregexp = CALLREGCOMP(aTHX_ t, t + len, pm);
+	    if (!DO_UTF8(tmpstr) && (pm->op_pmdynflags & PMdf_UTF8))
+		Safefree(t);
 	    PL_reginterp_cnt = 0;		/* XXXX Be extra paranoid - needed
 					   inside tie/overload accessors.  */
 	}