abuse bitwise ops to check for num vs string to avoid B.pm

[haarg]

B.pm is a rather large module to load, and numbers and strings can be
detected by abusing the behavior of bitwise operators. Using a bitwise
operator is faster as well.

A string & "" results in ""
A number & "" results in 0 (with a warning)

Some variables won't be fully caught by this check alone. A string like
"0.0" that has been used in a numeric context will pass the check as if
it was a number. We detect this by making sure the string is the same
as stringifing the numeric form.

"nan" and "inf" that have been used as numbers will also pass that
check, so we detect them by making sure multiplying by 0 gives 0.

[haarg]

This code also isn't effected by the deprecation added in perl 5.23.6 of using bitwise ops on code points >0xFF. Because one of the strings we are &ing is an empty string, processing stops before it gets to any of the actual characters in the value we're checking. This avoids tripping the deprecation warning, so it should also avoid a future promotion of that deprecation to an exception.

[charsbar]

Thanks for the cool patch. I tweaked it a bit to address the inf/nan issue later, in a different issue (#14, and makamaka/JSON#21 ).

[skaji]

It seems that in perl 5.28, it will be a fatal error if an operand contains a character whose ordinal value is above 0xFF. https://perl5.git.perl.org/perl.git/commitdiff/fac71630f045e7ab325b37548d59b6558735e5c3

In fact, with the current blead, the hack A string & "" does not work anymore.

❯ perl -v
This is perl 5, version 27, subversion 1 (v5.27.1 (v5.27.0-263-g7b3443d31f)) built for darwin-2level

❯ perl -Ilib -Mutf8 -MJSON::PP -e 'JSON::PP->new->encode(["あ"])'
Use of strings with code points over 0xFF as arguments to bitwise and (&) operator is not allowed at lib/JSON/PP.pm line 423.

# minimal case
❯ perl -Mutf8 -e '"" & "あ"'
Use of strings with code points over 0xFF as arguments to bitwise and (&) operator is not allowed at -e line 1.

@haarg What should we do?

[haarg]

That seems like a bug in blead. This PR may have to be reverted, but the current change in blead is taking something that previously didn't warn and makes it fatal, which doesn't follow the deprecation policy.

[skaji]

the current change in blead is taking something that previously didn't warn and makes it fatal

Indeed.

❯ perl -Mutf8 -Mwarnings -E 'say $^V; print "" & "あ"'
v5.26.0
❯ echo $?
0

❯ perl -Mutf8 -Mwarnings -E 'say $^V; print "" & "あ"'
v5.27.1
Use of strings with code points over 0xFF as arguments to bitwise and (&) operator is not allowed at -e line 1.
❯ echo $?
255

[charsbar]

@haarg, thanks for starting a thread in p5p ML.

[skaji]

It seems that the behavior was restored for a while.
https://perl5.git.perl.org/perl.git/commitdiff/b50535da2f4aaf97d13e96cda0069755fb6bbf76

Thanks, @haarg !