#!./perl -w
-print "1..76\n";
+print "1..4016\n";
my $test = 0;
my %templates = (
my ($enc, $content, $do_bom) = @_;
my $template = $templates{$enc};
die "Unsupported encoding $enc" unless $template;
- return pack "$template*", ($do_bom ? 0xFEFF : ()), unpack "U*", $content;
+ my @chars = unpack "U*", $content;
+ if ($enc ne 'utf8') {
+ # Make surrogate pairs
+ my @remember_that_utf_16_is_variable_length;
+ foreach my $ord (@chars) {
+ if ($ord < 0x10000) {
+ push @remember_that_utf_16_is_variable_length,
+ $ord;
+ } else {
+ $ord -= 0x10000;
+ push @remember_that_utf_16_is_variable_length,
+ (0xD800 | ($ord >> 10)), (0xDC00 | ($ord & 0x3FF));
+ }
+ }
+ @chars = @remember_that_utf_16_is_variable_length;
+ }
+ return pack "$template*", ($do_bom ? 0xFEFF : ()), @chars;
}
sub test {
for my $nl (1, 0) {
for my $value (123, 1234, 12345) {
test($enc, $value, $value, $bom, $nl, $value);
+ # This has the unfortunate side effect of causing an infinite
+ # loop without the bug fix it corresponds to:
+ test($enc, "($value)", $value, $bom, $nl, "($value)");
}
next if $enc eq 'utf8';
# Arguably a bug that currently string literals from UTF-8 file
# right now, as here we're testing the input filter itself.
for my $expect ("N", "\xFF", "\x{100}", "\x{010a}", "\x{0a23}",
+ "\x{10000}", "\x{64321}", "\x{10FFFD}",
+ "\x{1000a}", # 0xD800 0xDC0A
+ "\x{12800}", # 0xD80A 0xDC00
) {
- # A space so that the UTF-16 heuristc triggers - " '" gives two
+ # A space so that the UTF-16 heuristic triggers - " '" gives two
# characters of ASCII.
my $write = " '$expect'";
my $name = 'chrs ' . join ', ', map {ord $_} split '', $expect;
test($enc, $write, $expect, $bom, $nl, $name);
}
+
+ # This is designed to try to trip over the end of the buffer,
+ # with similar results to U-1000A and U-12800 above.
+ for my $pad (2 .. 162) {
+ for my $chr ("\x{10000}", "\x{1000a}", "\x{12800}") {
+ my $padding = ' ' x $pad;
+ # Need 4 octets that were from 2 ASCII characters to trigger
+ # the heuristic that detects UTF-16 without a BOM. For
+ # UTF-16BE, one space and the newline will do, as the
+ # newline's high octet comes first. But for UTF-16LE, a
+ # newline is "\n\0", so it doesn't trigger it.
+ test($enc, " \n$padding'$chr'", $chr, $bom, $nl,
+ sprintf "'\\x{%x}' with $pad spaces before it", ord $chr);
+ }
+ }
}
}
}