1
0
mirror of https://github.com/catseye/SixtyPical.git synced 2024-11-28 12:51:10 +00:00
SixtyPical/TODO.md
2018-08-26 14:35:28 +01:00

3.7 KiB

TODO for SixtyPical

low and high address operators

To turn word type into byte.

Trying to remember if we have a compelling case for this or now. The best I can think of is for implementing 16-bit cmp in an efficient way. Maybe we should see if we can get by with 16-bit cmp instead though.

The problem is that once a byte is extracted, putting it back into a word is awkward. The address operators have to modify a destination in a special way. That is, when you say st a, >word, you are updating word to be word & $ff | a << 8, somelike. Is that consistent with st? Well, probably it is, but we have to explain it. It might make more sense, then, for it to be "part of the operation" instead of "part of the reference"; something like st.hi x, word; st.lo y, word. Dunno.

Save multiple values in single block

As a shortcut for the idiom

save a { save var {
    ...
} }

allow

save a, var {
    ...
}

Save values to other-than-the-stack

Allow

save a to temp_a {
    ...
}

Which uses some other storage location instead of the stack. A local static would be a good candidate for such.

Make all symbols forward-referencable

Basically, don't do symbol-table lookups when parsing, but do have a more formal "symbol resolution" (linking) phase right after parsing.

Associate each pointer with the buffer it points into

Check that the buffer being read or written to through pointer, appears in appropriate inputs or outputs set.

In the analysis, when we obtain a pointer, we need to record, in contect, what buffer that pointer came from.

When we write through that pointer, we need to set that buffer as written.

When we read through the pointer, we need to check that the buffer is readable.

Table overlays

They are uninitialized, but the twist is, the address is a buffer that is an input to and/or output of the routine. So, they are defined (insofar as the buffer is defined.)

They are therefore a "view" of a section of a buffer.

This is slightly dangerous since it does permit aliases: the buffer and the table refer to the same memory.

Although, if they are static, you could say, in the routine in which they are static, as soon as you've established one, you can no longer use the buffer; and the ones you establish must be disjoint.

(That seems to be the most compelling case for restricting them to static.)

An alternative would be static pointers, which are currently not possible because pointers must be zero-page, thus @, thus uninitialized.

Question "consistent initialization"

Question the value of the "consistent initialization" principle for if statement analysis.

Part of this is the trashes at the end; I think what it should be is that the trashes after the if is the union of the trashes in each of the branches; this would obviate the need to trash values explicitly, but if you tried to access them afterwards, it would still error.

Tail-call optimization

More generally, define a block as having zero or one gotos at the end. (and gotos cannot appear elsewhere.)

If a block ends in a call can that be converted to end in a goto? Why not? I think it can. The constraints should iron out the same both ways.

And - once we have this - why do we need goto to be in tail position, strictly? As long as the routine has consistent type context every place it exits, that should be fine.

"Include" directives

Search a searchlist of include paths. And use them to make libraries of routines.

One such library routine might be an interrupt routine type for various architectures. Since "the supervisor" has stored values on the stack, we should be able to trash them with impunity, in such a routine.