Notices by pinskia (pinskia@hachyderm.io)

@dalias oh that is good to know.

I would have been more helpful except I am getting ready for a party on Saturday so I have not been in front of the computer or even next to the phone most of today

@dalias

So it looks like you need a new constraint that will be used in call_symbolic/call_value_symbolic patterns.

This is what aarch64 has:
(define_constraint "Usf"
"@internal Usf is a symbol reference under the context where plt stub allowed."
(and (match_code "symbol_ref")
(match_test "!(aarch64_is_noplt_call_p (op)
|| aarch64_is_long_call_p (op))")))

aarch64_is_noplt_call_p does:
```
bool
aarch64_is_noplt_call_p (rtx sym)
{
const_tree decl = SYMBOL_REF_DECL (sym);

if (flag_pic
&& decl
&& (!flag_plt
|| lookup_attribute ("noplt", DECL_ATTRIBUTES (decl)))
&& !targetm.binds_local_p (decl))
return true;

return false;
}
```
aarch64 still has the same/similar predicate as the sparc backend still. So even if combine does the recombining, the RA comes around and forces it to a register.

@dalias

if combing is doing recombining, then the predicate on the instruction (pattern) is accepting more than a register. (You might need to make sure the constraints reject the thing you don't want too).
For a memory location you might need to change the legitimize_memory hook for the target.
Without any further hints on what the md pattern looks like or the rtl looks like, it is hard to know the full solution here.

@inthehands

It is why things like opendoc/pink would be better if it was adapted when it was there. Would have provided the tinkering to the masses even more as it provided a high level way of doing embedding of different apps all at the same time.
But Steve Jobs killed it.

@dalias @mxshift

Even gcc before dco, accepted small bug fixes without a copyright assignment.
Small/non copyrightable changes didnt happen that much; maybe 10 over 20 years.
We(GCC) have been getting more recently (this year) but with dco there is less worry. And we just point to the dco documentation.

@dalias @kkarhan @vtrlx @hipsterelectron

Died in 2007 or so (so almost 20 years ago). The PS3 gaming experience was download updates on day 1 even and that was bad; especially for 2007. Many games had updates that would download on day 1 of their release too.

There was a version of the PSP Vita which was download only. That was 2011/2012.

I didn't keep up with consoles post leaving Sony but it was as far as I know not much as changed.

I think Steam has the same issue on PC gaming too. I bought some "physical games" back in 2010 or so and they just turn out to be steam codes.

@inthehands yea even pi has 6 9s in a row.

@dalias @mirabilos @equinox I think it was my fault since I replied (on accident being followers only and then everyone else did).
And the way followers only works in fedi is just "broken".

@dalias @equinox Note CFI as implemented in clang/LLVM actually requires LTO. The reduced form KFCI does not require LTO and uses a hash to check for compatibility (though it does not follow C compatibility rules; it is not enough for the Linux kernel).

This does remind me I need to review the KFC patches as stage 1 GCC 17 is about to start soon.

@dalias @navi

Yes I have one case where I was moving from one callback to another and I forgot to change the context object.

I only noticed when I reviewed the code. It was able to compile; stopping at compile would have saved me time.

In this case; the original call back was checking a hashset to see if an edge was in that set and we added all edges into that hashset except for the latch.
the new callback only checked one edge. So I was trying to speed up the code :).

@dalias @hipsterelectron It might be a hachyderm.io level block. Since I am blocked too.

@mattly so does that mean the dependency of rustc, llvm is off-topic?

@skinnylatte
I am not so sure.
Take east palo alto. (It has been gentrified in the early 2000s, when IKEA moved in and such). But it was always not the nice part of the bay. Even heard that after the gentrification like 8 years ago.
Now Oakland and Hayward has been the not so nice parts. There are also not so nice parts of San jose.
And I hate hearing that but I have heard from non whites too.

Note I have not lived in the bay area for 8 years but I lived in San Mateo for 4 years and Sunnyvale for 8 years. I still have friends and family that live there too.

@dalias I agree with the replies to a post that is public should be always public or the reply only goes to the op followers only and not your followers. I dont know if that is implementable.

Posts that are followers only should stay to the followers of the original poster only.

This will limit the harassment and also allows the op followers to easier see what is going on.

There are times, I rather keep to smaller group and random people not replying sometimes.

@inthehands I don't know if folks know there was at least one big name games made with hypercard: Myst.
Now there are things that should be improved over hypercard. (supercard did some of those).
I always find folks saying hypercard being limited is the reason why we should not look towards it to see what can be learned is just a way of excusing the complexity of the current computing.

@dalias Oh yes the Reiser case.
It is rather interesting how Reiserfs is was only finally being removed from Linux at the same time as you wrote that toot. I still don't know why it was not removed back in 2008.

@dalias I totally fucked up the other joke dealing with those 2 numbers.

Why 6 is afraid of 7, because 7 ate 9.

@skinnylatte Which is only 1/3 of the fine California imposed on PG&E for the fire in 2021:
https://www.cpuc.ca.gov/news-and-updates/all-news/cpuc-approves-45-million-penalty-in-settlement-with-pge-for-dixie-fire-2024

And 6.8x the amount paid to tribes for that same time.

@thomasfuchs made by insurance companies so they dont have provide money from insurance you paid into.

@thomasfuchs OO programming was supposed to simplified the complexity of objects instead people made layers of complexity out of objects. Templates/generics were supposed to simplify the complexity of generic objects instead it complexity even worse.

Concepts and type traits are most likely going to do the same.