Block comments ala Ray

[Bart Schaefer <schaefer@brasslantern.com>]

I've been noodling around with Ray Andrew's oft-repeated suggestion of
some kind of block comment syntax.  After considerable prodding of
gettok() in lex.c, I came up with something that works like this:

A line beginning with !# introduces a block comment.  This
(deliberately) means that NO_BANG_HIST must be in effect, so without
some hoop-jumping this only affects scripts.
I chose this because it seems unlikely to me that anyone would write in a script

!# this is just a silly way to mean false

The comment ends after a line containing #! anywhere in the line.  So
a full block could be:

!# This begins a block comment.
This is merely rambling.
This #! is the last line of the block

Stylistically, of course, I'd recommend putting the #! either at the
beginning or the end of that third line, but tokenizing worked best
just consuming everything up through the newline once #! was
recognized.  I experimented with allowing the block comment to start
or end in the middle of lines, but it was just too weird to be able to
do

print this !# part is a comment
but this part #! is not a comment

(which would print "this is not a comment"), and a solo "!" out of
command position may not be as rare as one at the start of a line.

I tried some other character combinations ... for example I can't find
any circumstance in which it is not a syntax error to write <# so if
it's appealing to match that with #> for symmetry, it's a small edit.
But I wasn't happy that those appear to be redirections.

Having told you all that ... there's one glitch with my
implementation.  To detect whether !# is at the beginning of a line,
in gettok() I test the "isnewlin" boolean.  This works everywhere
(that I've found) except when on the first character of a new script
file.  That means you can't begin a script with a block comment ...
which perhaps is a good thing, since typo-ing "!#/bin/sh" may not be
all that uncommon, and having the entire script turn into a comment
would be a bit startling.  However, I would like to know what (if
anything) can be tested to identify the first character of a new
script?  I've tried combinations of of
  isnewlin  -- this is initialized to false
  isfirstch  -- this is true after any separator
  isfirstln  -- true throughout any single-line command
What else could be examined?

Any other thoughts about this?  Too horrible to consider?  It needs
turning off in emulation modes and I haven't gotten to that yet.