.Net Foundation Design Reviews, 22 Sep 2017

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: .NET Design Reviews: GitHub Triage

Description

Notes: https://github.com/dotnet/apireviews/blob/master/2017/09-19-GitHub%20issues.md

A

I'll start recording, okay cut the chitchat. What do you think well, if we say something inappropriate, we're not gonna, publish it. It's okay,.

B

A

C

Right I recorded the last one right, but I didn't.

A

Scrub it yet so let's do it. You know offline again, like it's best effort. So don't worry about that so happy eyes.

A

It is the NE three ready for you there's so the list I just didn't open all of them so that we can pick and choose so.

D

Emo sent an agenda before he left and today some of the core effects lab api's were on the agenda. One of them.

E

Yelled review, it's ready, the other one is not ready. Yeah I'd also like to make sure we get the twenty one. Five three seven today, yeah.

A

So what I would so we have pretty good large backlog. So do you think that what you percent can buy because in it today it's very.

D

Small but I invited us on and I.

F

Say it's gonna.

D

Eat second hour: okay, cuz I thought this is triage.

F

G

Let's do the ones to the high priority for these guys yeah.

D

This yep, it's basic as I, said yeah.

A

I was actually you know one, you know, I'll shoot. You know, thinking that we should maybe do some of the you know, read memory and that stuff, like you know, but it's a priority anyway, let's start with the with the backlog, James Cove was actually poking at this one. Why it's not revealed yet so I think we owe it to him.

A

It's an immutable array and he wants to try to what Julie immutable array he wants it. On the immutable array, builder and I guess immutable dictionary builder on builders. We he basically wants a transfer.

H

Anyone familiar with the immutable classes.

H

Now this is update. Oh, this is older yeah. This is this. Is the latest.

A

Me it seemed to be like why not I understand.

I

What are these convenience methods? I mean? Its functionality already exists. What is it that's ours here? It is.

H

H

Think it likes, you know convenience, as you say,.

D

So this comment says something about: oh it's already an extension method. Are we moving it yeah so.

A

Is it just a move? No, it's I. Don't think that it's a more I, don't think we have the you know he was changing. You know, member methods to extension methods, parolee Bates, on some feedback. We can check it out, I'm, not familiar with the Builder types me. Neither so.

H

The reason this is API comers discussion so actually original.

A

Content was yeah to put it on the directly on the build on the class builder in the builder class, and there was prolly some feedback saying. That's not a good idea.

A

B

hmm Wasn't communal array, a project that you were involved? I was a bit involved, but with Andrew.

D

On either people team reviews but I.

A

Okay, it's not like span it's, not your baby yeah! You should use. This recording is, like you know, yeah. This is why we asked you to fill the full template with the motivation and ever think to.

J

A

I

All the viewers so I'm giving this quickly, it seems like the issue is: there are existing methods that are just named differently mixed methods, and so, when use the extension methods, you get the slower behavior and so he's proposing adding versions named the same thing. So when you compile it, it binds to the more efficient one. I guess is that I.

A

Didn't get that part? That's original.

I

These functions are already there they're just named differently right, are they I mean? Well, they should be I mean. The whole purpose of a builder is to create the thing. So there is a way to go from. You know a builder to an immutable half set so.

B

Is the problem that we have like an ienumerable dot to innumerable array that it's binding to or what I that's.

I

That's my understanding from reading, through the throw from skimming through the thread in like 30 seconds.

B

So a mutable array has an extension, well, maybe extension that that probably extension method yeah over an ienumerable called to innumerable array. So I guess the only thing I can think of is that it's that they're trying to intercept that binding, because people are calling to a new, more bowler to a mutable array when they just meant to call to a mutable yeah. That's that's my understanding as well.

A

B

Calling what today so a mutable array? Uh-Huh, not a mutable array of T but a mutable.

K

E

K

B

Turning it to an enumerable, walking and figuring out the link going back, so it's it has to do it. A rebirth walk when it's already buffered in the Builder, which is why the proposal is too specialized.

B

Api query: yeah: that's eBay.

L

And you get an overload.

B

Of to a mutable array, if one that takes an ienumerable and one that takes a builder but.

A

She should like, because it's a specific type on the Builder, it should pick the right one without any. You know necessary.

B

It cost them things so, but really what they're saying is they wish that when we had made the method or builder dot, two immutable, that we called it to mutable array so.

G

If I the existing code and I recompile it that does two mutable array, it will now become more efficient, but but what Christoph says is true: that I'm gonna hit job and now see I guess I'd see both anyway I already see better already see both you don't see two immutable array.

D

On I obviously do yeah.

B

So looking at the shape, I, a mutable array of T has a nested tech called builder. Yes, so if you have an immutable of T builder, that has an instance method called two immutable, but apparently people are calling that two immutable array which is binding to a mutable array, dot to a mutable array of T, which is the extension method that takes in it takes an enumerable.

B

So they see in intellisense two immutable into a mutable array and they write the one with the longer name, which is the slower function. Like that's the hint. It's got a longer name, there's more code there so.

A

Binary compound we're not gonna help them, but you know next time they recompile. They will at least pick the efficient ball, which makes sense like so bit.

D

Of success is what it's intending to do this feature, but I worried I, wonder whether it's going to accomplish what it intends to come. Why would it not accomplish what.

L

Problems, it's.

B

Both methods already show up now it doesn't matter which one you call you still get the fast behavior, because.

A

She sharp will pick two right over old for you and the road will do the right fast. Think it's not gonna to casting explicitly to ienumerable. They just want this extension method. The forward.

M

M

B

No I assume I mean yes, otherwise this makes no sense. Yeah. Okay,.

B

Yes, I list, so yeah beautiful array of T. That builder is an I collection and ienumerable, an ienumerable of T and I list and I read only collection and I. We don't know I.

B

Believe do okay, immutable array more easily.

A

Any comments are we, okay approving that seems reasonable to me.

H

Okay, so API approve this. This right.

H

Pull your future face there.

H

N

And it's gonna work. Yes,.

H

H

And that looks like a program should be surely straightforward and I bet that James is gonna jump on that. So.

H

N

Legs right earth.

H

H

A

The API to get an assembly name that doesn't throw on a native binary.

A

So it's adding the try at assembly name which returns false instead of catching exception, but image format exception. So.

G

What what's the problem with the try catch code? It's try catch code, it's reverse and also throws an unnecessary.

D

Exception so we had it. Try try better for performance reasons, come.

J

On we had it try we had it watch trial, Potter, ok,.

D

The either guideline says in throwing exceptions, is too slow for your scenarios. You should consider tripod parry, it's not to fix syntax of exception, handy, like I, don't think we should be doing this well. Such a slippery slope, like all API, is basically that catch exceptions. Now we're gonna have tried e ice and we back to error codes like let's not. The purpose of type error. I agree with that, but but.

G

Is it true, there's no cases where this is a performance problem.

K

Needy bugs everywhere.

G

K

All the toils you do crack open. My every G look at their versions and you basically can't distinguish whether there's a native or managed.

D

Seriously, like exceptions are slow, but only for, like you know, we're just loading assembly and throwing one exception.

G

I think try is reasonable to also.

D

Not the reason, because basically, availability is an issue, then again it applies to literally every single API. In that net framework, like we shouldn't be, throwing except then basically.

G

Says we should inverse.

D

G

I stand for when you use them as intended. Most of the time throw. That is not the case right, I think.

D

Majority of our exceptions, just throw I mean there are so there are precondition- exceptions that, yes, you can like you, you can kind of. You know, know how to use the API in that personal if it checks for an all. So yes, those I agree with you, but many other exceptions where they are not like precondition. Checks like you can get right in real programs, so I kind of don't like the trend of us, saying we don't like how c-sharp syntax looks. Therefore, we gonna, like I, think.

G

The thing is, the c-sharp Center I think that point may have been conceded. My my point is active out. However. I keep hitting this should.

D

Also, like you know so, we should have a meeting like I will fixing the problem with the departed, or we are throwing away exceptions in that night like we are not using them anymore, except for preconditions like one of those tops to is a like. You know, self-consistent a you know, guidance, but I, wouldn't like one off like somebody. You know this day. Somebody doesn't like an exception here. So where does it stop or.

G

It does it I mean it is possible for us to define something between its two extremes, right where a method is often called many times and in the normal course of events will often throw we could maybe say to use Mahalo trying. So.

D

We went through this like when, like lots of people rather basically said exactly what she said and the problem is I can give you kind of a history of the guidance around exceptions. It's very difficult to define depends on software. Like you know, people were saying like out of member. Is you know not normal? Well, if you like secret, very throughout all memory on the machine. The next year out of memory is very normal and like.

D

We I would basically say we should change guidance and not changing. Ap is one by one I.

B

Guess my question on this would be do: do we have an API that you can identify that something isn't as simply as as a boolean, without involving a throw like? Could you write this to avoid the throat you can't do that.

O

Because you had the crack open, dissemble you're, basically gonna do really nice only twice, but maybe that way that's.

G

True, but there would avoid a try, would you become what we had is assembly, but it.

D

First so like, if we are talking about the specific issue, I would say first, whoever filed the issue. Eric should say: is it syntax that his he doesn't like on his performance that he doesn't like performing what if I get like, because they should has long been state, why he doesn't like the code that he I.

P

D

So so one by one, if it's in that I would say, C sharp team, and actually that is my position by the way. I do think that our syntax for exception pending is really. The crux of this is the fact you have to catch the exception and you have to or you do the performance so anyway. So one is: when is syntax, one is performance, but for performance we have a problem like if it's performance, then we have tripod. It is designed for it.

D

In this particular case, I would like to see a proof that it's actually impacting performance at all, because exception, throw is like hundred cycles for simple exceptions and login DLL and analyzing is like its million.

G

So, for you, adding and adding an API like is assembly would be an anti-pattern, because now I'm, not a hundred cycles. I needed is a bit more about the like. It is super small in comparison to long in it, so you wouldn't care coding is assembly net load assembly, especially if only a fractal is also actually assembly.

K

Know you would care cuz, you don't want in fact, no stranger than twice every time. Yeah.

D

The cost is seen actually opening the file loading. You know, that's not your course. Yes, we don't. We don't want to do that twice. Yeah.

K

D

So you know it can be syntax, it can be performance, it can be debug ability, I would say like if it's one or of the others. We need to first decide which one and then think about what is the solution for our platform, not one by one kind of an API. Well.

K

In these cases, I think it is one by one why the case is because it's this API is performing. It has a performance performance I agree because we already.

D

Have a guidance for it, I'm saying I would like, because the exception syntax yes, I, would be really shocked if it's performance problems like it's loading stuff from disk like what, but it shows up, tons and tons, we're loading every cracking and designer into a change exceptions.

O

Were even in a ballpark or floating stuff from this, then so, so how would you.

B

If, if he had data here, that said that this would make like asp.net things start up, 50% faster, sufficient 50%.

D

Source, okay, I'm, saying like if there's anything, I touch your Apple. Yes, that's what try is for I'm saying, but it's not for fixing c-sharp, syntax or debug ability of I. Don't.

G

Think anyone thinks this is a performance issue as written yeah.

K

I would disagree, I think the EMAs build team in the compiler team will definitely think it's it's a performance issue.

D

That was there many years.

G

Except shion's are being thrown from time to time. That's right as much as I'd like sales performance I mean it's like.

K

There was basically big jihad about normal cases. Of watching. Vi should never have first chance exceptions. It comes.

G

K

I was, it is like when.

G

They had a new edit, suddenly thousands of exceptions when you start up and I mean that was maybe a person, but it is a debugger all if you think I'm not saying it. Try is the answer that, by throwing exceptions in.

D

Okay, so if it's.

G

Is so if it's, if.

D

It's not syntax, okay, awesome! So it's not a c-sharp team. It's a debug ability! Then we should meet with the debugger team and think what we do about it because again otherwise we like we basically are saying exceptions. Internet are broken, don't use them because they practical ability. I got over that. So like.

B

Well, it's one thing to say that you know never use exception, it's another to say that there should. You should always be capable of determining it is question, but the class.

D

You cannot you can before you open the file, you cannot check whether the file exists and then you know yo and file would succeed. So there are basically these I'd, like not preconditioning, issues where you cannot check the problem.

G

Is problem is exceptions? Don't live up to the name, it's like they're supposed to be exceptional, but actually what they're used for is the API can't do its job, which all too often is not exceptional, and even if we talk to the debugger team, the problem is like I'm sitting in window bug most of the time that's not kind of help or vs coat and stuff like we, you know we visual studio is not our only debugger I.

D

Isn't necessarily.

G

My answer, but I think that it's not so silly.

D

What you're saying is portrays your position is: we cannot fix the ecosystem debugging in your system, therefore, exceptions that are broke and therefore we should not use them, and it's.

F

D

Considered a exception.

A

In the past, we use that, like exception, should not be used for the main code path right, you know or main control or control flow or whatever, whatever we.

D

Call it exactly everybody likes States, it slightly differently, yeah and there's nobody ever. It defines what okay. Yes, people have been saying this and we've been saying: please write the guidance that people can understand when to use exceptions, and then people couldn't come up with it. So we have guidance that we have another I'm saying like we should do something, because I think there are significant problems in our error handling model both in the language there are times but I think it's.

D

There is the certain.

G

Pieces of work, they don't immediately come to mind where we've added is blah API to avoid throwing an exception. Now, in this case, you you don't want, you don't think, that's a solution because you crack the file twice. There's certainly other cases where that's a reasonable thing to do. Well,.

D

So I don't think it's even about cracking the file try twice, because the is pattern works for cases where there's no race right, like the become this one, won't be able to accomplish petitions. The here is also a race. So, like it's not a race, this will not be rich. What do you mean? You're.

K

Asking well in our file manager simply and then you go on open it up as a medicine but have been overwritten yeah. Then you've.

D

G

Especially begins loading, it.

G

K

I was trying to think of a comparison. One of the persons that we have is like the zip library when you go open up a zip file and if you it's not actually a zip file with over invalidated or something trapped like that.

K

But generally you don't open it up unless you have some kind of confidence as a zip, zip extension or something right, this API they're all DLL extensions, so you can't really even kind of make sort of like prejudgments like that are cheap before you open it, so that I feel like there's something we can do here. Whether or not it's a try pattern may be another overload or something maybe even change the API. The return null if it's not yeah. Somebody named if managed that was hurting nor was was a proposal there.

O

Eric said he was trying I, don't know, probably.

K

A degree of a breaking change.

K

A

Example, try gettin it.

K

Assembly name or no that's what you.

C

G

The name right so, okay, so I I think that the next step is to defer this and do two things. One is be more explicit about what we actually care about, although I think we know in this case and.

D

Then I so convenient.

F

It's convenient or debug ability.

D

Either debug ability or syntax one of.

G

Those two things, no one is yes,.

D

G

And use a live debugging through this exception, um and so I suggest you confer this and then have some separate time where we discussed this larger problem, which you write. We should have a principle rather than doing it, one bit at a time which that menu is probably here.

B

Maybe the you know, other versions, meeting right so.

D

H

D

Started to talk with him about the same taxi show, especially with Jared, because on Midori we had quite convenient syntax for many things like this. Basically, something like the note propagating thingy, where you could just call an API and he throws with.

G

Like bang bang, now on the.

D

Executing as a clause in your language, so we could address the syntax using something like this. The debugger issue is very power. I, don't know it's it's harder.

D

G

Mean I don't know if anyone happens to know, but it seems like in the latest editions of the ass like the default, the manage to buggery don't really catch anything first chance like it used to be that most of the first chances get into checked off in our installer. Yes, it doesn't break in anything I, wonder if that's a symptom, so they did that.

D

Yeah I I think in general, this whole feature of Christians exceptions could be improved to actually make it useful, because if it's on it, it's like it finds you too much. If it's off it bugs you naughty naughty nice.

G

Life are many verses and russiaís exceptions in just one cut I first John's acceptance to me something like seriously I was thinking about something like this way. You can kind of say like well.

D

You interested, in those exceptions.

G

D

Chance exceptions.

G

To me, or a wonderful, wonderful thing, having debugged a lot in the native code, yes, where it's like somewhere in this million line, function, someone's not taking a trip yeah, but to bring only first John's exceptions is probably the issue or maybe the next one yeah, but so.

D

In native code, because, like 32 API, there's no difference correct because basically there are error, codes and exceptions are kind of like that. You know the the houses on fire thingies in that net. For again, for historical reasons, we chose exceptions to use for everything we want to have error codes, which also exceptions and.

G

Destination, my point was that it sucks in the native code, because there's no quick way to break and envious is like 1 billion lines to load up a solution, whereas in managed code it feels like it used to be in the good old days you break in and like the second or third exception, would probably be the adulation.

A

All right, so is the second item would be what we convert it. We should discuss general guidance on exceptions from BCL or does the capture what we wanted to I wonder if it's slightly more specific.

D

That, okay, it could be useful because it comes up from time to time. I could give like short presentation or, like the history of exception, like even like what the guidance r2 is today, because it's United in.

G

Each of those contexts in each of this context: okay, that's.

H

Good and Christo signed up yeah.

D

I could kind of organize something and get the C sharp guys in the room. You got visual studio, debugger people in the room, cuz it it's in, buy money for you. So.

A

Again, we should keep in mind that XPS debugger is not the only debugger. You know. That's unlocked my.

G

Crisis yeah, sometimes you don't even using a managed to program yeah.

J

J

So back to its work.

F

It's a ship of squash today.

A

So this will be the longer on, but we had it for a while. Then you mouth ate the ice. If it shows up it's wrong in my laptop.

H

You're more api's.

D

So question about the CPI: how will it be implemented? Looks like it's gonna be using C++ CRT, so we're gonna just define.

K

Voc to CRT yeah, some of Matt's what we do put a lot of it: existing Maps API, so.

F

A

Rather than asking for what I'm trying to kind of leave, you know ask people to note that in the API proposals, like you know, I assume that you're looking at it from the point of view B, you know you like the API is assuming that we can actually bring Earthman assets. Well,.

D

I'm actually, fine yeah, it's because some of the methods are called X m1, which is not very that much in a sharp, but the main thing I was thinking about is some of these may be complicated and they're, not like this lots of code yeah and if you cannot reuse an existing binary that we already have. It depends on like CRT, for example,.

P

D

It will just blobs the runtime potential.

N

D

Then you know like maybe we should have these operations in some other. You know XML DLL, that's! Okay, like you know,.

A

Like really advanced right, yeah, that's a bullet point like you know, asking for the for the implementation later or in a condition like a proof, assuming that you know it doesn't blow the runtime and with more dependencies and stuff on platforms, a.

G

Dc, w user who's been very helpful in numeric stuff in the past, he's I think it's. His job has a bunch of interesting points about some of these are good matches for the framework and some are, and some are statistically oriented in some a numeric analysis, kind of oriented.

A

And then he's he's kind of providing data for or again.

G

Having this the easy.

A

G

And some not well that and also maybe there should be in different namespaces huh yeah.

D

There's this issue of like: what's our strategy format, is it like? We should show everything into the math class or we have a separate binary for you know kind of more advanced mathematics. Besides the very basic arithmetic operations that are in math today,.

D

Okay, where does it start like we keep going like everything is in.

K

One math class in math: well, we have a lot of advanced math stuff in the math. There are all basic trigonometry stuff and already listen. Yeah.

B

But I got it I, don't know that I would expect, like the error, reducing function to be.

G

Or the gamma function, looking for math.abs and I found the gamma function. Oh.

B

Well, you know what, if you want to compute a factorial, this lowly are just.

G

Looks like fifteen years a grade-school between those two things, yeah.

B

Because I mean I, like certainly you know the CRT if they found useful to add these things. It's probably you know in holistically speaking to useful to add these things. I just.

B

N

I mean there's a lot of math. You can.

B

Do but if, like maybe more statistics see things should be, you know, system numeric,.

G

Statistics thing would be helpful to have Eric Molina read.

D

Like so, we have system to melt the other correct something like this extrude it to a dll or like non core operations and is.

B

It question of chimeric is part of the well.

K

Complexin rationals big integer they're, not and system numeric vectors has been moved into big integers, not part of the shared framework.

B

Pill, it might be came to my first part of for, but it's not part of poor little sure well, I get. My question is: if we think that the implementation of this is going to be heavily based off of calling into system libraries on all our platforms, then we probably don't want it to be a free-form nougaty thing, but if, if we already have system numerix as part of the core install set where it's easy for it to carry, rather they like PM oaks and native dependencies and whatnot in trailer.

K

We don't you booked into the CRT outside of Portland more directly. You.

G

Can have sub namespaces on math as well I mean if I you can have nested.

B

Types yeah: well, my name is Cora Liban, where.

K

Did we put Matt that can't.

B

Have a sub class in a different assumption or.

N

Sub class, yes yeah, so we got.

B

B

Yeah so so I would say, like statistic: C functions that maybe we get a class of things under system, tumeric statistics, whether that be a namespace for a class and.

L

D

What another thing related is so eric is working on the tensor, a GIS. Oh there, you.

L

M

I just bring him very good rocket.

C

D

Tensor is kind of like.

G

So things that come up so far is: should these be a math? Should these be here at all and also we're gonna, stop they all likely to be CRT pass through so they're gonna blow our code and I think there's a general sort of theme across the different core lips to move more stuff into CRT, so I'm make it faster and more accurate. Is that not I mean.

P

G

Numerical operations- that's not the case. Okay, I.

Q

G

And I guess, presumably, then performance isn't an issue when you're doing removing that way. Well, ideally, we can potentially.

D

Improve it sometimes they interrupt colas like if the operation is super, cheap danger of call is slower than death.

Q

Yeah, there's also like we've, had a lot of problems where there's different behavior on different platforms in the CRT and also there's not consistent behavior like between, like versions of Windows. They might change the implementation of some of these functions that we call. So if we actually have a management limitation, it's much easier to get the same.

D

Sorry, but not CRT,.

Q

To avoid CRT, that's what we prefer.

D

I think so yeah, because these especially some of these, they I assume they are so cheap that actually they cost yeah.

Q

Another thing we're really hoping is with Hardware intrinsic forward.

D

Q

Can like some of the things that are not possible to like do outside of the CRT like use, Cindy, intrinsics and there's other stuff like that, shouldn't be possible in c-sharp soon, so, there's even less of a reason to go down into this CRT. So.

G

Do you think these are reasonable? On math I mean some of them like awk, sign and stuff that she's hyperbolic out sign? That seems reasonable on math, I guess, but hypotenuse I think.

Q

Ipod is pretty useful, um I, don't know about cubic square or the cubic root. Well, they only use they're all useful right. So.

D

Q

D

Like should they be on the bath right like do they, you know, are they in the same, we.

Q

Already have like arc sine and R cosine and our tangent on there right um do your shine and caution. Kenna.

G

Yeah guys that's ten age.

D

So what so, you're saying, basically that you believe the implementation for these types would be managed. Oh I,.

Q

Think, ideally, they will eventually be I, don't know if who proposed this? Oh.

R

Q

Know Tanner's already done a bunch of optimizations of some of these something existing math guys and some cases he has just called into the CRT like he implement all the existing method functions when I existing all the new ones that he added were in math and he did a bunch of like work in the. In course, he'll are to make them column to the CRT etcetera, see.

D

There's actually a comment at the end, it says that he wants to use articles any physics, yeah.

Q

D

Intrinsics that I assume this is there, then you feature and of course that's what.

Q

Copy and paste from CRT now that caused this calling an authority of calls where you call into the run I'm in the run time with this or I thought its we take cold copy and pasting to the rounded I think it could be either, but I might be wrong. It's.

P

Definitely calling in at their own time, I guess.

G

With their own time, because this.

Q

Not only was right, but it only one cost direct call.

G

It is worth observing, though, that if we do it ourselves, there's actually a lot of subtlety steps. More than was necessarily handled long ago in the original implementations right, I've seen a bunch of Piazza over the last year where people like ECW's and tan are doing up like fixing like weird cases with you know, yeah tiny epsilon and these old things.

G

B

If you guys have done that agrees with Eric statement that we do have sine cosine and tangent, the hyperbolic functions already on math so arc sine or cosine and arctangent those belong. There is the inverse of the existing function.

G

They do I now it makes becomes clear to me that these feedback was referring to a previous version of this and way down at the bottom. Tannaz extracted that now, so he was objecting to things like. I think he had way more remainder of quotient and, like the log gamma function, that's probably hard to par now.

D

So so, let's so, let's run through one possibility, so I can see like the hyperbolic cosine sine tangent being there, because we already have the non hyperbolic ones. Exponent of two looks good to me. Long looks good to me. What would we do with something like X, m1 yeah like I? Just don't think I like this to me, scissors are so much.

Q

Worth it because of the X yeah I, think it's just like combined operation, that's something.

B

Commented that it's more accurate, if you know you're doing it together, cuz, you know you can drop off rounding yeah, but.

D

Do you see like I think, like you know, can I.

B

D

Like this starts, like you know, some very corner case api's and especially like I, don't know what the name would be.

G

Three of like math to pine, for example: oh.

Q

G

But I mean, were you the graphics guy like? Quite often, you have a console like motorized or pie, and these are that potential.

Q

G

Pi over 2, we don't have where I like to say. We don't have right.

B

To PI you're gonna mean that most you know a bit inaccurate at PI over 2 you're gonna be at most of it, and it.

G

B

G

Mean that nobody have them in a concert have to do it every time so to multiply or whatever we already have, mass or PI. We could have math or 2 pi.

A

Sorry basically saying to be like them, except the xm1 and lock one pink I mean and.

D

I'm also concerned with the names of computer roads, yeah.

M

D

Mean we did call square roots word.

M

B

I mean the real question is like this cube: root belong on yeah. It seems like like square rooting.

Q

Is actually less.

B

Common than like this square.

Q

Right, like the.

B

Math class, for the most part, contains things you would have on a calculator in you know when you're taking like trigonometry and ice mmm I, don't recall my calculator ever having a cube root button had to do the manually go, take something Jeana. Keough blue button is quite common, but no it's.

G

2001, maybe.

M

With a little three yeah, absolutely yeah, we fancier Calculon.

M

I

M

Then the like headaches go to me 90. My.

D

Calculator had the same things that you're alright, my daughter's got later. It connects she graph.

F

G

A bunch of math today there wouldn't be on a calculator, did where my Tripoli remainder things like that.

G

Round something truncate: okay,.

M

So here's the code, this expand one windows, windows, calculator in Windows 10- does.

B

Not contain a cube root, but.

L

G

Your baseline did you switch to the I mean scientific one, are you know, man I would say, be m1 and change it to minus.

B

And Google X to the Y yeah, that's so you'd be like you know: I want eight to the 0.3 and that's how you do a cube root to.

G

The point three point: three three t2 to t3 yeah right, so you're, not getting it accurate, is the button.

O

What about this? That's the square root? Is it just quit? We should send immediate operator just downstairs right. Yes, the only one is not even raising to.

K

The power1, by the way, you know that it's basically taking the exponential yours.

N

And we're up to.

K

That subtracting one so yeah yeah.

G

And what's the base Apple that one's marginal on is it yes.

K

We use the the thing you're raising your reason. What.

G

Every person to eat that power comment when assume it's a numeric thing: how common is X minus one and Lord 1p I.

Q

Know that in the recent refactoring we did for complex log 1p and maybe log two showed up and also hypotenuse at the bottom. So I know that those who use those might have been sort of edit just because of that. But it seems like they're fairly common in like algorithms, so.

D

Why don't we go log log plus one.

G

Just expanding yeah I think that's a no-brainer. We were gonna. Have it that's what you'd call that right.

Q

I think the only like the reason there as is, is because that's the names that are in like see the standard libraries see it. We.

G

Q

On 100 places right.

G

Q

G

We don't do strollin and stressed and Worcester a stereo.

Q

B

So you know I, just googled log on P and X m1 or whatever that, if, if those like sort of the what the gold standard here be, if that was a immediate hit on Wikipedia of like this is the name of this formula then like that would be fine, but just because the CRT called it something doesn't mean it really does. But it's not usage, they do yeah. Okay,.

G

So I'm looking on the JavaScript math they have, they have all of.

P

G

Have CDR tea did I have ONC e plus one in the head expert, all.

G

X/X x x, down one, isn't it CL, z, 32x and one that's Lord, one P yeah. Well, they all super random on that as well. So cognitive.

B

What's the word, I mean he, you know, crypto math uses random all the time. It's clearly belongs right. There, man thought I need a random. They have square root of Hafiz long.

B

Q

Great opponent.

D

A

There's section ongoing stuff: looked it Morgan like or no you mean function or under narrator, no I function, Simmons right, yeah,.

R

A

Know community is working on that and Morgan was just like move.

R

Too, encouraging so.

G

Much expel more next I want more grumpy system that random and then you all those ones all the proposals seem to exist already of both java.lang math and also JavaScript math, in the exact naming sending them 1 and 1 Peter really m11.

R

G

Q

R

Q

R

Turner is here for the first time you meet him and see him in flash, hey, Bill, hey.

Q

Argumentation aiming now.

G

D

G

It fair to say we were convinced all this should be on.

D

My TV recording I, don't think that's good. Yes, so that depends of the names I give the names are beautiful. You can create a car out that way. It's not like, but.

G

Not sufficient I think I.

B

Completely agree with with a cosh and and friends, because we have not only do we already have cosine and R cosine. We already have hyperbolic cosine, so we should have its inverse. That makes sense the rest of the stuff I, don't know that it necessarily belongs like you should make another factory class and not not make math. Be this long.

G

What about the argument of consistency with Java and JavaScript and C++ eater? Well, it's just energy.

D

So on, what's diapers see in JavaScript as a.

G

D

Timenado but they.

Q

G

B

Like are we gonna always always grow, or do we want to try breaking things like you know in Excel, if you want formulas, you can say I want a statistics formula. Now it starts giving you things, you statistics say: I want trig formulas. It gives you trait like. Should we keep growing math? Should we start carving out like what kind of math spaces we have and put it somewhere on your parents? So as.

P

For ease of use in today's implementation, probably putting it math is the easiest just because that's where all the other app calls are routed through right now, I know that you can route F calls fairly easily to other places, but also for visibility. That's probably I think the best place, at least for the core functions here.

P

I did leave off some of the C runtime functions that aren't as well used like the gamma functions in the error correction, but when the symptom trinsic proposal goes in and we're actually able to implement these in c-sharp rather than having the F calls, I think it would be significantly easier to happen in other places. Oh.

D

So let's go stock like what is the right thing to do because, like sometimes you know like if the cost is 10x and I, think it plays a role but I don't think it's 10x yeah.

P

I think the cost of implementing these using air calls today be biggest cost as writing the tests. So there's another way. So, but let's.

D

Say we wanted to have them somewhere else like I? Don't think the cost is would be underneath yeah. No, it was.

P

You're about creating.

D

Blue entries, so I was saying like let's go like let's talk like what is that I think to do with one or we talk about.

P

The cost so I think that for the hyperbolic and cube root, those ones I think should be in the core math class. So the.

D

Hyperbolic we already, as Jeremy said, I completely.

Q

Agree I think that matches that, just because it squares us all so there yeah.

P

The other exponential and logarithmic those are more leash, even so than the hyperbolic it they're really they're, just because I Triple E hazard, behavior with and really big numbers and those ones probably could exist elsewhere, and the people who need them would probably be pulling in a numerous library already the same for hypotenuse.

P

That one exists because you're able to do some computations at infinite precision and then round down, rather than rounding it it's each intermediate operation and that one's probably exclusive to people who are already doing things like pulling in system numerous factors and complex. So let's say we had spent.

D

This whole space of pneumatics- and we had more not like things where we, where will be put I.

Q

D

And I don't know.

K

What they would.

Q

Be called and then I fear we'd have like a bunch of smaller siloed things that are hard to discover and and while the real.

A

Question is like, would they have to be in coral, ivory and Emma scarlet, or is it something that we kind of pull it out entirely out of PCL horrifies? We have the hardware.

Q

P

They have to be part of Courland because their app calls yeah.

A

Oh people don't approve them until you know, we basically make them block on same thing.

Q

A

That's the idea, the question is how much harm is downlight and by doing them by I, don't know a year or something I mean I.

Q

Guess they could be like okay sort of like guinea pigs for the heart or intrinsic put.

L

Q

Lab to start with, and then optimize them see how that pot works and then promote that more or more into domestic effects: extensions: okay, let's split up just because they're more nation! hmm Well,.

B

I mean most people will miss them. A little bit of it does depend on so like. If all of these are mostly now like a processor instruction, then then that's, then they can go wherever once we get. You know more processor level.

P

Capability, none of them will be singular, there'll be multiple, but and and the other limitation on that is there's no so be the most performant implementations of the core math functions and the C runtime currently come from AMD, Intel and Microsoft, and none of those are open source, so we're either going to have to talk either internally or externally to see if we can agree to get some of that open source as a community effort or we'll have to go and reimplementation ever so.

B

Yeah, so the the other position would be if we think that this is always going to be calling the CRT or on UNIX systems calling in to lift math or whatever a so that we make it look. We always give the same answers that a native program will be given and B, because they can, they can do things like they'll already have written the oh. If you support this special instruction that I'll, do it one instruction or I'll, do it in software 15?

B

Well, that's what this event allows and right, but I was gonna for if we're going to end up calling in to library functions anyway, then I mean it's not that we need to put it in a mask or lip or that low in core, but we probably don't want it to be a standalone nuga package because now we're doing complicated, like oh I'm jumping into live math as a peon book which read split packages of nougat are annoying.

D

Okay, so let's assume that we have the intrinsics and this and this and we can put them into a separate library, I still don't understand like what would the library be and with a title like, we would have a new-type code well,.

P

I would think something like it would go into system numerix along with complex in big integer and what time possibly math extensions or something, and if c-sharp ever gets extension everything. Then you could make that B extensions on the actual system, math class, if you wanted, but that would also probably be a location where half precision if that gets supported or quad precision, Triple E floating point, if that's ever supported overdoing.

B

It fresh right, now.I I'd, say that we want you, know, system, numerix, trigonometry system, numerix whatever, and it's each class of functions isn't a helper type of its of its appropriate naming fact that we have this big jumble right now called math things that if we want to make a second one, we're probably just going to make a big jumble, but it would be nice if we could, like whatever you know like you know, maybe we need a system numeric logarithms that there's just utility functions for doing logarithmic things so log to log with an arbitrary base like anything else whatever at and yes, it'll look silly right now when it has two functions on it, but as it as we have more need for doing things and as we have more need for helping people by writing the doing processor level intrinsic stuff.

B

That you know is hard to get right, then, okay, so why did.

L

B

L

That log two is not.

D

Used commonly like.

B

I think I I, really love to is actually quite useful. It tells you things that tells you the upper bound of Hamming weight of a values.

P

I think because more commonly people or even the compiler, like.

B

G

P

Will convert that to a pop counterfeit test and just return the highest bit rather than actually doing a large the algorithm so.

D

There's basically.

P

D

That, if I use look to for like the purpose of you, know doing it, checking and manipulation I would just not use this function like if it's a power of two, it's not gonna, be faster right. So, let's still do the same exact thing: yeah attention, yeah, okay, yeah in this case I- would just not have those like you kind of have to understand what this thing is for so.

A

Honestly, if you actually thinking about like having math extension or something like I, think that we get another good thing inside our collections into the core effects, extensions and basically have it internally community-driven, because we should not be gatekeepers or, like you know how many cool crazy things in math. There should be. There's no need for us to do.

J

B

Get modulo some implementation details like if it's all, if it's all the things that we're going to try routing through system, libraries, I, don't think any I, don't think it makes sense to be a standalone movie from.

A

The reasons it's not possible or it's extremely hard because bite the bullet for.

B

Extremely hard and it's painful to the to the consumer of a package, why? For one it makes package publication difficult because, like the building, you now have the problems that we had with the core 100 of like. Oh, you can't build this and copy it to different systems that you have to go build about.

P

The darkest of the Linux runs.

B

And the NuGet packages are bigger and when you hit a platform like if we've moved on to a new platform, but that one hasn't, but now it's just a dent like there's a dead tree, you just can't go down that so I think. If, if we see things being just oh, this would be a live math, and this would be the CRT like. That's that should be in the court issue not in front.

L

Of convenience, of course, but yeah, because it's just it's a paint Elementary's.

B

In the intrinsic said, we don't I just don't know how many things to CRT defines that are helping for Standardization, as opposed to like, because the processor can do this in an optically.

P

Well, most of the math functions except for square root. Aren't can do at any single instruction. They are generally 100, plus Cindy instructions with special paths and they exist because people use math all the time and I know one of the big pain points for math was we didn't have minute half, but now we do and now the big pain point that is, it doesn't exist on desktop.

P

So you, if you want to use single precision math you have to target an accord or create a wrapper class and caste etc, and I could see the same thing happening with this at least as long as it run time dependent in assembly. Instructions will be like that for a while as well absently.

A

Well, the desk desktop versus core angle is true for any. You.

P

Know API we will have to buy double with with with api's that are not dependent on the or runtime users, at the very least, can take the source code and copy it and know what it in practice. No, but it does that well, but I.

D

Think it's that's one thing, but basically, if it doesn't Bend on a particular runtime, then we have an option of shipping it. As a you know, particle out open component, then anybody can use it immediately. Yeah, like I'm.

R

D

Thread now with Steve, because I don't know what to do with some of the expand processing. Api swear, like you know, I would like, for example, people who write address decays to be able to use them, but they need to be portable.

D

You know all the versions of that that framework and the moment we put them in the car. We exactly run into the same issue and API such as cure il implementations. They may as well be out to abandoned and anybody can use them today. So this is kind of a very similar trader.

P

Prefer for this, if there's out of band width required runtime support, then it's not really out of band. It's really limited in its use and customers don't have a workaround, whereas at least with others, even if people aren't doing it in practice today, and that may just be because there's not a lot of people targeting multi targeting desktop versus net core or some other reason. Having said that, this.

D

One, so it depends on the intrinsic feature, which is a runtime teacher. It.

Q

Cannot be both previous version, so we would have the multi target and then have like a peer saw for fallback and then it's like way slower and we have to maintain two versions. It's like well.

P

Are we going to be maintaining two versions for the software fall box, anyways, potentially.

Q

If you want to run it outside of that core that well well,.

P

Well, I mean even on net or because, if you happen to be on hardware, that doesn't support the intrinsic. For whatever reason it's.

Q

Possible that then we.

P

Q

Want to support software.

P

Plus, at the very least, a software implementation provides a consistent. This should be correct, no matter what changes are.

I

You listening on.

P

So I think the software implementation is going to end up existing or any functions that are implemented.

A

Okay, so some resolution here, what do we do? I would approve the three yeah the.

Q

First, three and cubic.

D

Root only Jeremy said that I mean I, didn't like it, because he's.

L

Calculated I, don't like I, don't know any time that I have.

B

Ever wanted to compute the cube root of something square roots. I do all the time because you're thinking to be that III I guess that's busy. The pops I mean the me in my computer. Graphics class did not get along.

M

We think it's I don't like, but I guess we're.

M

P

Just like that, both because that's what appears on graphing calculators and the I Triple E spec, as well as everywhere else, a we legit job telescope.

J

Does the same right? You know that I'll check that okay, oh.

D

J

Like what would happen.

D

J

Did we have inches.

P

It's and we kind of have a bit more flexibility, so hypotenuse with intrinsic becomes essentially not needed. Just like you just like a FMA instruction methods not really needed, because then users can just call the fuse multiply. Add hardware close: it.

B

Okay and then the so exp to is that to to the argument or e to the argument.

P

Yeah, it's essentially POW to in that number, which.

B

You can simulate with left.

P

B

Yeah yeah, if they need something with a fractional exponent and they can call POW and if they went into an integer, they can do much it so that the benefit.

P

The real benefit of x2 is if you're, actually working with non integers, then it is significantly faster than Colin Powell like order of 10 to 100 times, I ended.

B

There but I think math is cluttered and I I would say we probably want to have an answer of what we would do with things like Earth and gamma. If we were to add them anywhere. Where would we add them before we start.

M

Being like well, let's.

B

Just join everything, let's just like urgency,.

D

Like we need them for our scene, you know you just have to help them now. I, like the idea of starting at mount extension project and seeing how far we can run with it in a separate DLL, and then then we will at least learn like what things are possible. What are not once we have, you know more requests for even more exotic math functions. What to do with it like it will answer a lot. Yeah I.

Q

Think it's good yeah and I've also wanted to do like other experimental things that probably wouldn't immediately go anywhere. So.

R

I explode blog gone.

Q

B

Yeah, so the top three or four symmetry and apparently cube root is.

H

P

Calculators not changed since, like 1960 I still so like the same book. So did it.

B

P

Root button: yes, the the TI graphing calculators do I.

Q

Root and then you would type in yeah.

B

So it had had a to the like in through into the in power, but it didn't have a cube root button I think that's a my head, but.

C

Picture the TI 86, okay,.

B

This would be that fanciest calculator, I ever owned and.

A

I'm gonna stop recording and because my computer needs.

C

P

A

Gonna reconnect again, all bit later there, okay Steven.

H

Okay, so this is approved yeah.

B

There's no cube root button on this.

D

There's square root, which is shift square; okay, now look at TI from 2017 J. So it's not.

P

Recording now it's.

A

The recording sorry I'm.

H

Gonna finish my review, so this is the you.

J

Know it's something like.

C

J

Dollars what the heck yeah.

C

So ready or no.

N

That's low-res, okay, good, okay,.

H

L

Thought not action hiding.

B