stop ~ in format modifying format string

Currently, the format parser converts ~ or ~~ in a format string into
blank spaces. Since the previous-but-one commit, it only does it in a copy
rather than the original string, but this still defeats the "if the string
is the same don't recompile" mechanism.

Fix this by leaving the ~ alone in the format string, but instead cause
FF_LITERAL to convert '~' to ' ' when appending to the target.

Also, in S_doparseform(), improve the processing of '~~': previously
it only skipped one '~', and processed the second '~' on the next loop;
this happened to work, but it's less unexpected to process both chars at
once.

I've also added some tests, but these don't actually test whether the
format gets re-compiled: I couldn't think of a way to do that short of
checking the output of perl -Df. Instead the tests I added were based
around making sure I didn't break anything related to ~~ formatting.

I also improved the description string for some of the existing tests.

2 files changed, 59 insertions, 11 deletions

diff --git a/pp_ctl.c b/pp_ctl.c
index 8591328abc..79b7d32caa 100644
--- a/pp_ctl.c
+++ b/pp_ctl.c
@@ -604,9 +604,23 @@ PP(pp_formline)
 	case FF_LITERAL:
 	    arg = *fpc++;
 	    if (targ_is_utf8 && !SvUTF8(formsv)) {
+		char *s;
 		SvCUR_set(PL_formtarget, t - SvPVX_const(PL_formtarget));
 		*t = '\0';
-		sv_catpvn_utf8_upgrade(PL_formtarget, f, arg, nsv);
+
+		/* this is an unrolled sv_catpvn_utf8_upgrade(),
+		 * but with the addition of s/~/ /g */
+		if (!(nsv))
+		    nsv = newSVpvn_flags(f, arg, SVs_TEMP);
+		else
+		    sv_setpvn(nsv, f, arg);
+		SvUTF8_off(nsv);
+		for (s = SvPVX(nsv); s <= SvEND(nsv); s++)
+		    if (*s == '~')
+			*s = ' ';
+		sv_utf8_upgrade(nsv);
+		sv_catsv(PL_formtarget, nsv);
+
 		t = SvEND(PL_formtarget);
 		f += arg;
 		break;
@@ -618,8 +632,10 @@ PP(pp_formline)
 		t = SvEND(PL_formtarget);
 		targ_is_utf8 = TRUE;
 	    }
-	    while (arg--)
-		*t++ = *f++;
+	    while (arg--) {
+		*t++ = (*f == '~') ? ' ' : *f;
+		f++;
+	    }
 	    break;
 
 	case FF_SKIP:
@@ -4951,16 +4967,21 @@ S_doparseform(pTHX_ SV *sv)
 	if ( !(!!SvUTF8(old) ^ !!SvUTF8(sv))
 	      && len == SvCUR(old)
 	      && strnEQ(SvPVX(old), SvPVX(sv), len)
-	)
+	) {
+	    DEBUG_f(PerlIO_printf(Perl_debug_log,"Re-using compiled format\n"));
 	    return mg;
+	}
 
+	DEBUG_f(PerlIO_printf(Perl_debug_log, "Re-compiling format\n"));
 	Safefree(mg->mg_ptr);
 	mg->mg_ptr = NULL;
 	SvREFCNT_dec(old);
 	mg->mg_obj = NULL;
     }
-    else
+    else {
+	DEBUG_f(PerlIO_printf(Perl_debug_log, "Compiling format\n"));
 	mg = sv_magicext(sv, NULL, PERL_MAGIC_fm, &PL_vtbl_fm, NULL, 0);
+    }
 
     sv_copy = newSVpvn_utf8(s, len, SvUTF8(sv));
     s = SvPV(sv_copy, len); /* work on the copy, not the original */
@@ -4994,10 +5015,10 @@ S_doparseform(pTHX_ SV *sv)
 	case '~':
 	    if (*s == '~') {
 		repeat = TRUE;
-		*s = ' ';
+		skipspaces++;
+		s++;
 	    }
 	    noblank = TRUE;
-	    s[-1] = ' ';
 	    /* FALL THROUGH */
 	case ' ': case '\t':
 	    skipspaces++;
diff --git a/t/op/write.t b/t/op/write.t
index d436730959..646143d529 100644
--- a/t/op/write.t
+++ b/t/op/write.t
@@ -61,7 +61,7 @@ for my $tref ( @NumTests ){
 my $bas_tests = 20;
 
 # number of tests in section 3
-my $bug_tests = 4 + 3 * 3 * 5 * 2 * 3 + 2 + 6 + 2 + 1 + 1;
+my $bug_tests = 4 + 3 * 3 * 5 * 2 * 3 + 2 + 66 + 2 + 1 + 1;
 
 # number of tests in section 4
 my $hmb_tests = 35;
@@ -542,9 +542,13 @@ for my $tref ( @NumTests ){
 			  "$base\nMoo!\n",) {
 	foreach (['^*', qr/(.+)/], ['@*', qr/(.*?)$/s]) {
 	  my ($format, $re) = @$_;
+	  $format = "1^*2 3${format}4";
 	  foreach my $class ('', 'Count') {
-	    my $name = "$first, $second $format $class";
+	    my $name = qq{swrite("$format", "$first", "$second") class="$class"};
 	    $name =~ s/\n/\\n/g;
+	    $name =~ s{(.)}{
+			ord($1) > 126 ? sprintf("\\x{%x}",ord($1)) : $1
+		    }ge;
 
 	    $first =~ /(.+)/ or die $first;
 	    my $expect = "1${1}2";
@@ -555,12 +559,12 @@ for my $tref ( @NumTests ){
 	      my $copy1 = $first;
 	      my $copy2;
 	      tie $copy2, $class, $second;
-	      is swrite("1^*2 3${format}4", $copy1, $copy2), $expect, $name;
+	      is swrite("$format", $copy1, $copy2), $expect, $name;
 	      my $obj = tied $copy2;
 	      is $obj->[1], 1, 'value read exactly once';
 	    } else {
 	      my ($copy1, $copy2) = ($first, $second);
-	      is swrite("1^*2 3${format}4", $copy1, $copy2), $expect, $name;
+	      is swrite("$format", $copy1, $copy2), $expect, $name;
 	    }
 	  }
 	}
@@ -654,6 +658,29 @@ ok  defined *{$::{CmT}}{FORMAT}, "glob assign";
     $^A ='';
     ::is $format, $orig, "RT91032: don't overwrite orig format string";
 
+    # check that ~ and ~~ are displayed correctly as whitespace,
+    # under the influence of various different types of border
+
+    for my $n (1,2) {
+	for my $lhs (' ', 'Y', '^<<<', '^|||', '^>>>') {
+	    for my $rhs ('', ' ', 'Z', '^<<<', '^|||', '^>>>') {
+		my $fmt = "^<B$lhs" . ('~' x $n) . "$rhs\n";
+		my $sfmt = ($fmt =~ s/~/ /gr);
+		my ($a, $bc, $stop);
+		($a, $bc, $stop) = ('a', 'bc', 's');
+		# $stop is to stop '~~' deleting the whole line
+		formline $sfmt, $stop, $a, $bc;
+		my $exp = $^A;
+		$^A = '';
+		($a, $bc, $stop) = ('a', 'bc', 's');
+		formline $fmt, $stop, $a, $bc;
+		my $got = $^A;
+		$^A = '';
+		$fmt =~ s/\n/\\n/;
+		::is($got, $exp, "chop munging: [$fmt]");
+	    }
+	}
+    }
 }