This must have been a really old one, but apparently hasn't really
been triggered all the time. Edge case anyway.

With Go 1.10, apparently go vet has become more picky about the
sources and won't be satisfied until Lisp preload and the function
registration data are present. So, simple, build them first.

Also, fix setting unbound variables in face of warnings. That was
a *bad* bug.

Also, new concept of warnings at all.

Also, provide it; this means the l/testing.lisp needed to be renamed
to l/regtests.lisp.

 * prog2: have it in the first place
 * prog1: document that first argument is mandatory
 * prog1/2/n: have test cases

Now we can have the `if' demand a second argument.

Only one, and needs a test case. There is more lurking in that
vicinity, as the ticket says.

Otherwise, results are irreproducible, making test cases difficult.

Also, test cases.

function-definition, function-arglist, function-body, function-docstring
to return the respective properties of functions. In the course,
some cleanup re docstrings and synopses and things.

Does not even compile yet.

The new function readerstring is much closer to those of other
Lisps, e.g. SBCL and CLisp, in particular the latter. More aesthetically
pleasing, too, it seems.

Although I now have no means to get at the function definition
programmatically any more (well, not without tediously dissecting
the readerstring), I can create them.

It has been an alias for normalize-pathname for a while anyway. In
the course also changed get_pathname() to accept arguments other
than strings and ports, such that all functions dealing with pathnames
now accept pathname arguments that are not strings or ports. Their
ValueString is used instead, so symbols or even lists work. Admittedly
it doesn't seem to make much sense with the latter, but why restrict
that?