FMSLogo Code

Finish fixing Bug #616 (Empty words created by DLLCALL now act like the empty word).

The earlier fix missed that ldllcall() converts the buffers from Unicode to bytes after the function call.
This would have been okay except for two things. First, Null_Word is not a STRING but CASEOBJ and so
getstrptr returns something that was not intended to be a string pointer. Second, there's an off-by-one
in this conversion logic modifies one byte more than the length. This was previously safe because
make_buffer allocates one more character than needed to pad it with a NUL character, but is not safe for
Null_Word where getstrptr returns a pointer to who-knows-what?

The follow-up fix is to skip the Unicode-to-byte translation if the buffer NODE* is Null_Word.

Fix Bug #616; Empty words created by DLLCALL now act like the empty word.

I chose to fix this in DLLCALL, making make_buffer return Null_Word like make_strnode does.

An alternative fix could have been to find all places which assume that Null_Word is the only
empty word and fix those to look at the string's length. That would have been a more robust
fix, but it also would have been more invasive and undone whatever the original authors were
tying to accomplish by making Null_Word a singleton.

Update the expected test log to include the output of some new tests, as well as some bug fixes.

Cleanup: Remove the special case in butfirst() and llbutlast() to return Null_Word.
make_strnode already has this logic, so there's no need to duplicate it.

Add tests that FIRSTS, BUTFIRSTS, and BUTLAST take ownership of any node :erract provides them.

Fix signficant regressions in LAST introduced by recent refactoring.

Document the calling convention of logofunc, as I understand it.

Fix a recent regression in FIRST where it could leak a value provided by :erract.

Fix a recent regression in BUTFIRST where it could return a freed NODE*.
Calling gcref(arg) could inadvertantly free val if it were a list.
This change also adds some more comments and an assert around the memory
management contact of validate_argument.

Change BUTFIRSTS so that, on bad input, it considers its entire input bad, not just one element.
This is consistent with how other procedures work.
The logic was adpated from lfirst() and lfirsts().