@neauoire @z I have on occasion said something similar about software. We should talk less in terms of product X using programming language Y and library Z, and more about doing task A using mechanisms B and C.

Or imagine if we didn't compare programming languages but spoke in terms of programming language mechanisms. Ooh, I think I just figured out another reason https://github.com/jwilk/python-syntax-errors rubbed me the wrong way.

@neauoire @alderwick Can I ask what the motivation is for the optimizations?

Idea: distinguish between optimizations and _idioms_. If someone's writing something in an unncessarily convoluted way, a linter is a good way to remind them of features.

But programs are often fast enough. Nagging about optimizations might seem heavy handed. Show that sort of advice outside of build.sh, only when asked. "My app is slow, do you have any suggestions?" That way also lies a profiler.

@neauoire Turning "JSR2 JMP2r" into "JMP2" (tail call elimination) does feel like it makes the code a little harder to read. There's something reassuring about the way you have a function call per line at the moment.

@neauoire @alderwick That makes sense 👍 I just feel defensive with most linters 😄

@neauoire @alderwick There's an interesting discussion about tail call elimination in Forth at https://wiki.c2.com/?TailCallOptimization. Neither it nor Scheme provide for pragmas to disable optimizations, I believe 🤔

@d6 @neauoire @alderwick My sense is that the term has enough meanings in this neighborhood to justify further dilution 😀

@neauoire @d6 @alderwick Can you elaborate on what get-ruler does?

@neauoire @d6 @alderwick Oh. The first execution of JMP2r returns from `@get-bound`, and the second returns from `get-bounds`.

Wow.

Maybe this is like an inner function? Like this pseudocode?

fn get-bounds(a, b) return @get-bound(a), @get-bound(b) end

Of course, that ignores the interesting mechanism you're pointing out. Perhaps what you want to do is focus on the mechanism rather than the result.

@neauoire @d6 @alderwick Tail, body. Depends on how long it is and how many legs it has, right? 😄

@neauoire @d6 @alderwick Ah, I see.

So you have two routines sharing a single body. Maybe both together are a single multi-headed routine?

There's also some similarity with optional arguments. Like if you had a flag that decided whether or not you run the first part of a function.

@neauoire @d6 @alderwick Oh one candidate is 'trampoline'.

https://en.wikipedia.org/wiki/Trampoline_(computing)

How to avoid side effects

Pay attention to all effects. Then they won't be _side_ effects.

(Reaction after reading https://arslan.io/2019/07/03/how-to-write-idempotent-bash-scripts)

@xiroux @neauoire Check out the Readme at https://github.com/akkartik/mu, and feel free to ask me more questions after.

What platform are you running on?

@xiroux Yes, I was already watching this thread. Especially if you prefer something more interactive than a book, I'd be happy to answer questions.

Some caveats, just in the spirit of full transparency:

- Mu only supports 32-bit x86, but I'd be happy to help you look up answers to questions about 64-bit as well.

- I'm more knowledgeable about the processor instruction set than quirks of Assembly notation. But I can try to answer questions about Assembly.

Thanks @neauoire.

@xiroux Cool! What OS?

@ekaitz_zarraga What processor do you tend to run it on (even if emulated)? @opfez @royniang

@s_ol I think I see the constraints now.

A pseudoinstruction might be an option. Represent auipc+jalr altogether, and mention the label just once.

This whole RISC-V feature makes me nervous. Are there any special boundaries where auipc being on one side and jalr being on the other would lead to subtle problems?

Can you -1 more in auipc and keep offset positive?

CISC looks pretty good right now 🙂

You haven't really said anything about your goals? I get the sense they're different from mine.

@s_ol The basics make sense, though I don't understand the sign bit. Is the instruction grabbing 11 LSBs and 1 MSB?

The slicing operators seem find, but I'm not sure how I feel about the `label+4` syntax. One heuristic I used in SubX was to not try to use all the available power of the instruction set. Are you sure you need far jumps to unaligned addresses? Maybe organize code in pages, with cross-page jumps restricted to the start. A dispatch table or something.

@s_ol Oh interesting! I see in my browser history that I actually searched for 'aiupc', which is probably why I got that page.

Thanks for pointing it out. I get annoyed when people spell Gandhi as Ghandi 🙂

@s_ol Yeah that aiupc construct is a bit strange: https://stackoverflow.com/questions/55314747/risc-v-user-level-reference-or-reference-implementation

Even outside of our weird notations 🙂 I'm struggling to understand how a compiler would use it. Are they expecting that functions are aligned at 4KB boundaries or something?

Oh, this is only for far jumps. So that alignment constraint might actually be a reasonable choice.