todo

listing format errors: ignore whitespace of input

documentation: print supported directives

---------------------------------------
I was not able to locate a Macro-11 language reference manual any more
recent than for RT11 version *3*, so I used that plus my recollection
of more modern features.  It was enough to get the RT11 V5.4 kernel
built, plus a significant chunk of our own code.

The biggest missing feature is full featured listings.  The .LIST and
.NLIST directives are ignored, as is .SBTTL.  No table of contents is
accumulated or printed.  No symbol cross referencing is done (most
likely I'll just write a CTAGS file, not a cross reference listing).

Many errors still go unchecked.  Off the top of my head, I recall that
object and listing file output errors are ignored.

.FLT4 format may be inaccurate in the low bits.  This is because IEEE
64 bit format has two fewer mantissa bits than 64 bit PDP-11 format.
Without writing soft-float routines, there's not much I can do abbout
it.

Expression math is done in native width, almost certainly 32 bits.
Truncation to 16 bits is done only for listing and output.  This may
make some output differ in the presence of 16-bit overflows and
underflows.  I don't think this needs fixing.

.REM blocks containing code can screw up .MACRO, .REPT, .IRP, .IRPC.
read_body in macro11.c would need to be able to parse and ignore .REM
blocks.

Need to search a path for the .INCLUDE directive.  Right now it only
takes a complete file name.  And most likely, existing code will have
RT-11 style file names; I don't know what to do about that, except put
in a device name parser.

Possible enhancements:

It would be very simple to make macro11 resolve internal symbols with
more that 6 significant characters.  Even so, only the first 6 would
be used for external symbols, and you have to be wary of existing code
that used (for example) .LOOKU rather than .LOOKUP, since these two
would become distinct.

SYM = 0
MOV SYM(R0),R0	; macro11 could optimize SYM(R0) into just (R0)

I dream of automatically fixing branches out of range.  Easy when the
destination is backwards, difficult when it's forwards.  I have this
idea: during the first assembly pass, all branches generate a long
branch if the target symbol is undefined, otherwise an "optimized"
branch (short or long) if the target is defined.  Then keep a
128-instruction FIFO of generated instructions.  Each FIFO entry is
tagged with context and symbol definition as they are pushed to the
FIFO.  When an instruction gets pulled from the FIFO because it's more
than 128 words away, the FIFO is searched for long branches that point
to this location; any such are shortened, and any symbols defined
following their location in the stream are adjusted.  In the second
assembly pass, the FIFOs aren't used because all jump distances are
known, and the right sized branch (JMP or Bcc) can be generated.