more m4p bugs

Since you continue to patch things as fast as I report them (now
testing with commit a4cc0f3c):

* m4p is not draining diversions at zero-arg undivert, but saving them
till EOF (and m4exit is correctly skipping the implicit undivert that
otherwise happens at EOF):

$ m4
divert(1)a divert(2)b divert undivert.
 b a .
m4exit
$ m4p
divert(1)a divert(2)b divert undivert.
 .
m4exit

* m4p is not allowing undivert to take multiple arguments:

$ m4
divert(1)1divert(2)2divert

undivert(2,1)
21
$ m4p
divert(1)1divert(2)2divert

undivert(2,1)
m4:stdin:2: Warning: excess arguments to builtin `undivert' ignored
2

* While modulo math works for +, -, *, and you've already fixed some
bugs for / and %, there are other cases where doing math wide and
truncating only later still shows observable differences:

$ m4
eval(0x80000000%3) eval(1<<31>>1)
-2 -1073741824
$ m4p
eval(0x80000000%3) eval(1<<31>>1)
2 1073741824

But you're in good company: bsd-m4 from
https://github.com/chimera-linux/bsdm4.git claims to try and be GNU
compatible, but has its own quirks in this area:

$ m4-bsd
eval(0x80000000%3)
m4: numeric overflow in expr:
-2
eval(1<<31)
-""

and in general, since C17 says that many aspects of shifting a signed
integer when the sign bit is involved are undefined, your
implementation may still manage to comply with current POSIX.

* Your latest tweaks to builtins being aliased is still not right:

$ m4
dumpdef(`defn')
defn:	<defn>
define(`a', defn(`defn'))dnl
dumpdef(`a')
a:	<defn>
dumpdef(a)
a:	<defn>
$ m4p
dumpdef(`defn')
defn:	<defn>
define(`a', defn(`defn'))dnl
dumpdef(`a')
m4:stdin:4: undefined macro `'

dumpdef(a)
a:	<a>


-- 
Eric Blake, Principal Software Engineer
Red Hat, Inc.
Virtualization:  qemu.org | libguestfs.org