Fix literal rendering of Index values

[leonschoorl]

There was confusion between the size in bits and the type level argument to Index.
Causing it to use way too big numbers in the overflow calculation.

Fixes #2813

Still TODO:

• Write a changelog entry (see changelog/README.md)
• Check copyright notices are up to date in edited files

[leonschoorl]

This fixes the hang that Felix experienced.

But I've still left this as a draft, as the overflow behaviour still isn't right, or at least not the same as in Haskell.

topEntity = 6 :: Index 5

Evaluating this in Haskell produces:

*** Exception: X: Clash.Sized.Index: result 6 is out of bounds: [0..4]

But clash will just generate HDL with the 6 in it:

  assign result = 3'd6;

Should it just produce the same HDL as if one would call errorX?
So:

  assign result = {3 {1'bx}};

or:

  result <= Index_topEntity_types.index_5'(0 to 2 => '-');

[DigitalBrains1]

But I've still left this as a draft, as the overflow behaviour still isn't right, or at least not the same as in Haskell.

Should it just produce the same HDL as if one would call errorX? So:

Let me just note that the contract is "When Clash produces an XException, the HDL can produce any value whatsoever". So it's law-abiding.

I'll not touch the issue whether it's desired.

[leonschoorl]

I've now added the overflow detection too

[DigitalBrains1]

When would k be zero in this code path?

I feel like that only happens for types like Index 1 which shouldn't render in the first place?

[leonschoorl]

I was basically copying what the VHDL backend does:

clash-compiler/clash-lib/src/Clash/Backend/VHDL.hs

Lines 1767 to 1772 in 3b755b9

	expr_ _ (BlackBoxE pNm _ _ _ _ bbCtx _)
	| pNm == "Clash.Sized.Internal.Index.fromInteger#"
	, [Literal _ (NumLit n), Literal _ i] <- extractLiterals bbCtx
	, Just k <- clogBase 2 n
	, let k' = max 1 k
	= exprLit (Just (Unsigned k',k')) i

[leonschoorl]

But I think you're right

[christiaanb]

That code was added in version 0.6.15 6b365dc#diff-d636015ead3ec91b11144d7f18c90be9133810ab83457b3649a91e046cfae73fR647

Which might be from a time before we elided 0-bit values?

[leonschoorl]

Yeah probably.

I had seen that, at least in simple cases clash, removes Index 1 literals because they're 0 bits wide values.
But I'm wasn't sure if there was some code path I didn't foresee. And assumed the author of that code in VHDL.hs knew what they were doing.

You want me to take out the clogBase?

[christiaanb]

LGTM

[DigitalBrains1]

Suggested change

	FIXED: Render overflowed Index literals as errors
	FIXED: Render overflowed Index literals as don't-cares in HDL

?

Maybe that's not good either, but I worry people might not understand what "render as an error" means. I'd also like to emphasise that we're talking about HDL here for clarity.