.Net Foundation Design Reviews, 28 Nov 2017

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: .NET Design Reviews: GitHub Triage

Description

00:00:00 - Approved: Add overload to Path.GetFullPath() to specify base path https://github.com/dotnet/corefx/issues/25535
00:09:18 - Approved: Need Span overloads for Path APIs https://github.com/dotnet/corefx/issues/25539
00:13:12 - Approved: Need Span based path join API https://github.com/dotnet/corefx/issues/25536#issuecomment-347624492
00:44:02 - Approved: Add PipeOptions.CurrentUserOnly option https://github.com/dotnet/corefx/issues/25427
00:49:16 - Approved: Add Scalar Intel hardware intrinsic functions https://github.com/dotnet/corefx/issues/23519

A

A

B

Jiminy energy, the artemisia gonna go over the path length.

C

This this is fine. We probably join. One probably want to do last, because the naming is a little awkward, but.

C

This is something I've brought up in the past and just haven't opened a specific issue for directly.

C

There's two there's two problems: there's one the very biggest specific thing is that, like we don't have a way to dis, disentangle yourself, away from the current working directory when you're, resolving bass and, of course, the you know, the current working directory is a process wide setting, and that creates a lot of havoc for build scenarios and anybody else it's trying to take impose.

C

You know it gets changed on a background running thread or whatever else, and you know it's suddenly all your paths end up in random places, and this is a pretty common thing that hits people. So this you know be able. We provide a base path for when you get a full path that would allow you to do but would be effectively setting the current working directory without actually doing it.

B

The only question I have is that should be made so that the new path space purpose now that effectively is the same as calling to get full path without a base path, meaning it's users, current working directory.

D

If you have an optional argument, just okay.

B

A

Still there yes, they're, not linked or.

E

At least some cases- okay house, yes sure me I, just.

B

C

B

C

You can you hear me, yes,.

B

C

Right sorry, I'm a little concerned about it because of the just the the opportunity for that too, to miss errors right like if they have an error case. You know the path may or may not come through the way they expect, depending on.

F

Should be very.

G

H

Specifies I'm not clear on the semantics we used Sherpa 8th clock based path. How does this differ from? Can cat.

C

Well, just in the same way that it basically behaves the same way.

C

You know the internal resolution of Paz happens on the on the Oasis and that if the path isn't fully qualified, then we will put in whatever the base path is as being the working directory effectively.

C

So I call it out in the implementation notes, but basically we're checking his path fully qualified and if path isn't fully qualified, then we use the base path to fully qualify. It there's one weird corner case and that's with a driver, a little Paz, oh and then doesn't I call that out is the last last one rather than make that overly complicated if it doesn't match so.

H

The semantics of this are essentially if path is fully qualified. We ignore base path. If path is not fully qualified, we can cap them and inculcate full path. Yes,.

C

It's not precisely King cab because of the weirdness on Windows, but in in the case of UNIX. It certainly is directly I can get in most cases on line. Does it is akin CAD drive relative stuff makes it weird, but that's that's a corner case for most people and.

H

This is common enough. That I mean what I just described it in my mind, it's basically a single. You know if, if fully qualified called get path with one argument, otherwise call it with a different argument outside.

C

Of Windows- and there are the windows- behavior is much more complicated and it's harder for people to get that correct. So because you have you have Drive routed and you also have current Drive routed, which is just difficult for people to get.

H

The steps right for that and.

C

H

Second, question is the other issues you've opened are all about. Adding new API is the path that takes spans. What he's taking strings I.

C

Don't I mean I? Could we could have them to expand? I didn't see the scenario as being that common, where you would want to be resolving off of a span, I'm happy to do both or dues, and instead it makes them a little bit more complicated.

C

You know, I guess you know they're, particularly from the standpoint of like allocations right, because you know we have the optimization forgetful path that, like you know, we don't generate another string if it ends up matching the same.

C

So if you take, if you take it and turn it into a span and then we I can't.

H

Be can the person yeah.

C

So that's why that's why I leaned this way, because I couldn't see the real like scenarios for actually using a or needing to have a span for resolving the final thing? Well,.

H

In the scenario it wouldn't be for base path, necessarily it would be for path, presumably you'd say like path pocketful path with in, for example, if you were able to get a span that represented the the file name portion of a path, you might be sure on that substring in that that gets the dead. Still it's.

C

The previous point and I'm trying to add it.

H

C

You're talking about and.

H

I take a slice of it. I couldn't return that original thing anyway, sure.

C

Sure yeah, but Lee, you know like taking your strings and to disbanding them. You're gonna have an unexpected allocation that the normal case forgetful path when people are calling it as it gives you back the same path you pass him.

C

It gets called a zillion times all right, particularly in our own internal code. Yeah.

H

I'm not arguing against that I'm saying yeah I'm arguing why there might be a case for a spam version as well sure.

C

Sure and I get that one I'm just like how do we communicate that that people don't one run into though I just didn't, think your ear case is that common, but I'm, fine with it, if you'd, if you think it's.

H

Not comment- and you know this stuff more than I- do it's not common, you don't need it, but.

C

I'm, happy to add it later to based on is.

H

That questioning the need for this API at all, but you know it's.

C

A pretty common problem and it's a super common thing to happen in the particularly in in build scenarios, I mean.

B

Economic example is you parse, a project file that has file names in it and you want them to be relative to the project directory, not the current directory right and then you do this Hector II, so yeah I'll, if there any concern that doesn't fool the API as is I'm fine with them not accepting. Now it's a bit on the weird side but go down. There was like one comment from Jana Bordeaux of those things: oh I.

A

Would match the existing one there's more important.

C

That was that was my comment as well: I thought we were trying to match to them and.

G

Especially yeah logically.

C

I preferred the other way around, but for consistency. Well,.

F

The difference is with combine is that over there you don't play the games of like if the path one is or the path 2 is absolute, that we kind of ignore the path 1, which actually is the case over here. We ignore the base off in certain.

I

Cases it actually does it as well. Oh yes, that's another one of my my issues. All.

C

But it's, but in that case it would actually.

C

I mean, as it is right now.

F

A

Sure you end up flying, so we don't here. So by the way, can anybody write a USB driver, because that would make my life so much better.

B

Yeah, that seems a little then alright, then that's just an excellent.

C

So, let's do that one last, just this one's pretty straightforward.

C

So it's a pretty common thing like for, like manipulating like source and destination paths, and things like that to be getting the file name or the directory name and then combining the two.

C

So getting those getting those spans out, you know we internally, we have, you know get index of and that's essentially what these things become. So.

H

The first five listed here all the ones that became with get those are all basically substring operations or slices in this case, and each one just has its own logic about what the start and offsetting counts are. Yes,.

F

How come on our dos my thoughts? Are they showing up at visual studio? I must build or.

C

Somewhere, like that, they Emma's build they're all over the place, and this is actually specifically, you know. Discussions I was having with David the David King stuff, they're they're actually emitting up writing most of this sort of similar stuff themselves, because that's that's where their allocations go. Nuts, because they're doing this getting slices basically based off of the directory and.

B

F

That's one thing like you know how many hops actually use that one off path and msbuild is a good case for that on a visual studio, so I mean.

C

This is Stephanie. This is definitely a very hot path for this. To do anything.

B

The house of symbols is a Rozlyn.

J

You seem quite a bit too, but it caches a lot so.

C

Logic is his path rooted. Isn't it isn't it great yeah, I added as pepco, fully qualified, because people think that's what his path really means and it doesn't we will get into it here. Think.

B

Of it as starting with the drive letter, it's not correct.

D

Oratory actually.

C

Really you have to get into it, because my next one actually involves this path routed so like.

B

Any kind they all over those existing things, so there should not be any ambiguities.

H

Is passed fully qualified, these are overloads and it's just replacing like search and replace string with read only span char, yep.

C

It's Beth fully qualified is pretty much the same thing as well. Oh that.

H

Exists with a string argument today, yes, is.

K

There much to gain this you're, not manipulating the actual and.

L

Put here for those guys to pull the bull ones yeah well, certainly the has extension.

C

Well, you're not manipulating.

M

Anything you're.

C

Right, but like is path for reading data like it's not as useful as the other ones, but like if you're reading, some data stream in and you don't want to actually resolve the path you know you want to be looking for it does that thing actually have the extension before I go and create an actual string to pass it around to other things, then it would be useful, like is that incoming data stream? That thing coming in actually a CS file, for example, or something like that.

K

We have any other thing else on path like maybe combined. That should also benefit from Spain. That's a little exclude. Okay,.

H

C

A separate issue: what's.

H

That I have your gate guys when it doesn't like there is no filename, for example. Does it return null? Is a return, an empty string? What does it do today.

C

Some of them return, no I, don't think. Generally, they return all I haven't like gone through them exhaustively. Some of them might return empty, but null is the typical thing and.

H

So these were all returned empty, empty.

C

D

Okay, you couldn't turn down as well. If you wanted to right.

H

B

What I mean is if we don't spend that effectively, has a null pointer right? No, isn't that the thing that we're that's that's an empty spam.

N

Okay, but if I do a d-plus spend, is it not different from an empty one? No, no. The same I mean.

H

You can have it, you can have a span that has a length zero with any value in its ref. It's still a zero length span. Sure.

C

I'm pretty sure they all return null but I'm, not a hundred percent sure without going saying.

B

Is that the Koala should be able to do PN chip I call their neck is default right, which will be different right.

C

If it's actually a no there's, no valid like empty thing anyway, so I mean any of these things. If they return empty. There is no such thing as an extension. That's empty or I can treat that Sunday or whatever you're.

B

Asking the general case: if you really have to differentiate between empty and no it's just awkward for the caller, they have to call it to default, rather, the chicken clarity as.

H

Well, I, don't think we should be encoding semantics based on the value of the ref field inside the span and whether it's and whether it's zero or not I tend to agree with that.

C

B

Points to the screaming poles.

O

B

Translation from an API reference.

O

A reference type today we really need to do pH if we do have the guidance that we don't follow.

P

This as one should treat null in the industry with the same thing. Yes,.

O

P

This just says we are not treated with the same thing: yeah I guess it's.

B

I think it was funny yeah I think it director, which is not officially.

K

Your file name just returns. This is the actual path you passed it if it didn't find any directory. Scepters separators, your extension returns null. If it doesn't, it's not.

C

Yeah, there's not a null vs. empty semantics. For any of these, so I mean it's. It's either going to return something or no. You.

B

Like even if you have to change this is about level of work like that, it's fine all right so then spend path, phase join API.

C

This is, you know, based off of like meaning what we actually intended, or most people expect combined to be, but the combined API is actually very subtly confusing and it's a very common thing for.

Q

It's a big room.

C

From what specifically has the problem, the behavior of it looking at his path rooted for the last it returns? The last is path rooted in the list of arguments so which most people don't expect, and every other framework just puts the paths together and doesn't care about like whether there's a proceeding separator or not other than it doesn't need to add one right so like this. This example here path combined and Sifu and bar on every other framework, at least that I've looked at you know those will get feedback, seafood bar and ours.

C

It just gives you back bar and I.

H

C

H

Dust that example gives you back in bar. Yes,.

C

It should do the same thing on UNIX too, because I believe we implemented it the same way. We check his path, routed.

C

Like there's actually multiple ways to route, of course on Windows, but that particular example is really.

C

My most people don't expect that until they actually run across the bug and then they get frustrated and go too.

C

And that's that behavior and I talked about it more further down because it was brought up by a few comments and, like my last comment on this I, go into more extensive detail on like why the weird behavior for this thing then I linked out a few issues on on the external issues that actually go into that, and so I only bring that back up here. So why is it directory separator considered routing?

C

This is the holiest path, routed thing right, the routing is it's considered routed because it's routed to a to a drive right. It just so happens. It's too routed to whatever the current drive is see. : would also be read in C, colon backslash is routed.

C

It's just the current semantics of what is routed, but both of those things are actually relative on Windows and most people think like his path route. It is to be whether or not it's relative or not, and it's not that's, why I added the the other API in our prior release for is fully qualified, because we didn't have a great name for it.

B

We'd only spent so far it's pretty straightforward, yeah.

C

And then the the questions that were brought up in the comments was like: why don't we actually use combine or some variant of combine and I'm? Not keen on that, because the behavior is so radically different. I actually wanted to appreciate combine because nobody expects it to behave the way it does then encourage people to use join instead and plus the other api's that we're adding here for resolving.

C

So if people were actually trying to like mimic the what we just added the path get full path, they really should be doing that and then using join for intermediate steps. Jordan seemed like the best name to me. Given the way string works.

F

It's a brand new API, it's not adding over all to join, join, doesn't exist. Joy.

C

Does not exist, the question that was brought up at least by um Justin, was why don't we actually use combined so.

B

If we say be needed user, you name this kids I hate your life. We wouldn't have to overload the tank stream because they would correct.

C

I, don't see the need to add a string one in this case, because you know we're always allocating on the out right. Well, not always. Actually there is. If the spam is empty, then we would get back the original, so it's possible. We can have this written on as well. I just didn't start there I'm.

H

Wondering if that's actually true, if it's span is empty shouldn't, we just ensure the path ends: I guess: if the path to was empty and path, one already entered for the directory separator.

C

That's a good question. My current logic is not you know it's not like it's. We don't need to add a directory separator because there is no. There is no path right and that's kind of matches. The current behavior, it's sort of the other things ending in a trailing separator is not a good thing in most cases in Windows, that's just generally wrong.

C

We have a different semantics around it, but, like you, can't actually pass a path to any of the windows API with a with a trailing separator, except for one.

C

And then that's the other route of the drive C colon backslash is were actually required. Everything else. It doesn't mean anything and you actually get an error.

C

So I'd be inclined not to do it because of that.

C

And I'm fine with adding string ones too, but they have an immediate need for those ones.

B

Will be go as far as they combine is deprecated. That's.

C

What I'm encouraged down at the bottom and comments, hey you'll, see like look at the look at the stack overflow links like this widest path, combined, not properly concatenate finally, aims to start with patcher ever occur, and you know it's also on the pretty it's reasonably high up on the list of the what's the worst gotcha and c-sharp, or done that.

B

C

The problem is, is that it's people could conceivably actually be, depending on that behavior.

B

He would not change it for existing apps. You would just do it when you make them pop like yogurt. Another member could say that's a new behavior. You shouldn't depend on this behavior anyway. If you really care that who you want stuck.

C

Sure that's possible I mean we could do that for the existing run and still depreciated I'm I'm concerned more about the the porting scenarios and stuff, like you know, working it on Quora I, just kind of rather not do that I'd rather not change the behavior either.

C

Some reporting perspective of combine that is so.

B

What I don't like is, if you have basically an API that does the correct thing in 90% of the cases as a 10% corner case, that's annoying and they majored using new API and tried to get you a point off of a very commonly used API to new API. Just progressed the 10% seasonal move, not only concern I mean I would say that I would I would probably prefer people to find a way and crooked if necessary, and then just have combined having the correct.

B

You know, code on the boat, behavior and then just use that name by the introducing I I'm.

C

I'm, okay, with that I was just constantly concern, was just that whole courting right scenario: I mean because it is, it is a pretty it's pretty likely to break things, but I'm fine with that as well. I just looks uncomfortable making that a starting point. The.

D

Only one that's ready here or.

K

Powershell calls it Joyner, they have Commandant's, go join path. Yeah using combined under the hood I know they do. What I'm saying the notion isn't there for.

B

The call again, given that combining source they can search in my like a huge mind, share like it seems weird to like pull off of it again. I mean like in order to do this properly we would have to do is anyway, the I would definitely old API. So every time you combine it as quickly after you call the other API, and it seems we can't I, don't.

K

Think we can change the behavior of join successfully going.

F

All the big my.

K

Sorry, combined I think we can successfully. You don't think we can change combined, so why? Why do you say that? Because we could work it right, mm-hmm, that's gonna, be hope, I! Think it's going to be able to bring a breaking change to the court. Why you say that.

K

Anybody who's depended on it today, accidentally, which is I'm certain. There are lots of people who are will be broken, but whatever we program in every compiled is I mean, like the closest.

P

Thing if we want to avoid introducing a completely in verb, would be to overload the existing combining to take some sort of you know. You know what sort of like which level of weirdness would you like, and you hold your current level of weirdness and then over time. Every time we find another stupid thing to do we'll add another email. Look. Do this different set of stupid things we.

O

Won't you go back because that's that's actually harder for Paul than just change.

J

The name you could always check and see how badly this would break msbuild and BS, and if because this is one of the top api's, so if it breaks either of those significantly enough than working, it probably will be possible. I.

K

Know I've debunked through some things, I got surprised that we were actually dependent on the behavior accidentally.

P

Again, so dirty that yeah.

P

K

Debating on whether or not we can accessory court combined, I.

R

Think Oh welcome. You think we ever can't I, don't think we can successfully I. Think was this thing I, don't think we can Wes Berg.

C

Was you that was my initial? That was my initial gut feel as well? Is that I didn't think it was possible, particularly from the example, so the commercial.

P

Leave almost guarantee the bus, stop yeah I think we just run msbuild and PowerShell and whatever, and if we find that we break them, that would tell us we can't work it. If we don't find it, we break that we have data. So it's hard to hunt those down all. It was heard that we shouldn't quark it, but we can't prove that we can yeah. That's probably true.

P

C

One of everything I wouldn't have high confidence outside of the finding the readings you.

C

Know that said, people will come back and say why don't you just fix the first one you've done daily.

R

They do I can deal with the repercussions. That's why yeah just stating it, because if you look at those those threads I-I've been annoyed by the debugging, all many timers, why didn't we pick to the core I've been annoyed by this many times as well, so I know both.

K

R

Also wide adult student methods come back, why it's bigger than it used to be? If they can either goes on. I know: I'm core is one yeah.

G

That's combined to ooh.

F

Yes, but it takes free parameter, swagger, psyche.

K

Very shocked if you could work it successfully, you're.

B

Gonna play great.

K

More much like a.

B

Race or a lookalike.

K

B

Still know them right, no I know but, like you need to super culminate here, I ride and like all fleeting the API and give you a new API, doesn't seem like they're, not.

K

Saying we I I I, don't think that's been necessarily obsolete. It I just don't think we can change that behavior well,.

B

But that seems like that's the weirdest thing we have to API to do the same thing if you bet.

K

In theory, basically make the combine over let's take span, have the bug fix at the.

R

Issue fixed it there's.

C

No really no using path combined unless that's the the existing ones where people are trying to actually leverage that behavior aren't really being doing it correctly, because this bad food is just fundamentally a batting the eye. That's what this thing is based on yeah.

S

So so why don't we introduce a combined options, Ino and then it has the old behavior than you want that.

M

Would basically be working is oh well, working is hit yeah.

F

Well, based on what your target and what you run on and that kind of like magic right, which is dangerous, we.

C

Have one or we have one path combined that actually takes a params.

P

C

Additionally and I talked about it done with the comments like I, can it just seems really weird to not deprecated the first one I don't see how to do that without actually creating another name.

B

Yeah I mean every basically know thirty percent of that status.

B

It's unsafe to be this reminds.

D

B

Sorting it's like common API, the corner case is there, but how many people actually broken by that and then it.

J

Just move on I think the other concern is with ops. If we decide to obsolete it and add a squiggle, how how big is the effort to fix that in msbuild or Visual, Studio or Rosslyn or anywhere else, where the code base millions of lines and they have Warner's error turned on.

P

Well, that's, he would also lead. Would you.

C

D

In fact, array.

G

D

Is used less than that course.

B

I guess combined.

B

I'd say I'm work fondling, using in your name, I mean that that seems like just pushing the candle.

B

Like I think we need to find a boat where we can do the right thing in an experimental way without breaking the world I, don't know what that would be. We.

K

Can make it an opt-in as an we can add another API on path that says: have the old, combined behavior it'll be true by default, and then you can set it to false. If you want yeah that one I would honestly say is like you're making a capacitor which, instead of a court, which is a way to get human argument or I, mean a normal regular capacity, a sickly. The.

G

K

Detected, that's.

G

K

If you have libraries and things scented light, you.

G

Know yeah, but an arrant when you're in yep.

K

Then you can, you can have a bit better control over, you know well, and what would be the default be it have to be. The thought would be exact behavior as it is today. Well, why can't be combined it with working and say for the old stuff? It's effective.

B

That's exactly how we do.

K

Working brother, my practice, we've been central, a default. We set about the bad switches from The Fault value based on very nice. Exactly record is.

B

The thing in order to get off the bad behavior that you need some point in some version has people has to change its own. Otherwise, you never get appeal. It.

K

Doesn't matter for people who a lot of people don't.

B

Ever even hit it because it doesn't work in you just set the quirk. That's that's why you have a compact. Such the people will never opt-in to a competitor for better behavior. The competitor is always for making things more compatible. It's not with making things less problematic.

B

I can sing in order, for this would be viable like DBS to say at some point and related should be. The first version is dizzy, float switches to nope. We give you the will be pleaded to correct behavior by default. If it doesn't work for you well, then you settings the old behavior back, but that's the only way to force people don't have.

C

Yeah, that's what we do for everything else, yeah so I think I mean only because.

B

I didn't want to talk either we just need to current behaviors is, and this ban base version says: take the exact same behavior, what we fix it, but I think introducing a new API and then have people like basically stop pulling. The current one seems I know that seems.

K

Like a very bad one, so we asked our data friend to see if there's any way of gathering this data, if they're at work people might be depending on the behavior I know, it's got I, don't know any clean way of doing this.

B

Yeah, that's a good question. I mean. Let me just take a note on that, because that's kind of question, if you should be able to answer I, just don't know how.

C

What was the question we.

K

Should ask our data friend Scott watchwords Macmillan company if they don't have a way of gathering I, agree weird how's this? We should ask for data on whether or not we think that a lot of people are, depending on the old behavior, essentially.

C

I think it's there's definitely people that have that.

K

They have a bunch of profile data, for example, I think they have profile data, for yes, even might be able to give us frozen again.

C

Well, oh well, you know: I'd stayed on this particular one that I think more people were broken by the current behavior. That would be broken the other way.

C

This is since up usually being the big gotcha. It's people.

K

Hit it and they like what's going on and they fix it, but yeah.

C

Yeah, but like it's coming it can it's not a thing that happens immediately right. You have to get your data set up in a way that, like suddenly you have this weird behavior and like it can it's, because that thing is relative. You know it may actually be correct if the current working directory is set to what you like would expect that thing to been in and the current working directory gets changed and then, like once in a blue moon you'd, have an error.

B

This issue, this span basically should have the behavior that combined has and then the second issue, which is can be fixed to combine behavior, because then we can I, think approve one and it's all the other ones separately. I.

C

Don't want to in unless we fix the combined behavior I, don't want to encourage people to use it by adding span /.

B

Like a but to me, that's a completely orthogonal issue. Right I mean one of them. Is you want to be able to use combined with spans all that combined I? Don't I, don't think we want.

I

To use combined at all, unless it's fixed now ignore.

B

The behavior that combined has we pretend for a second. We can fix combined because the ability is 30 persuade can fix, go on short.

E

Today there.

B

Exactly really.

I

B

Fixing the only thing I want to do is they want to like stop allocating right yeah so that that's that's issue one and taking issues we don't like the behavior of combine right there. They're they're, not necessarily related I, would not tie them together, because that means we cannot give you non allocating api's, because we can't agree on what the behavior of combine should be to me. They're like separate issue.

F

Well, the argument might be that people who really would use the know locating api's or they are afraid of this API today, right their own.

F

So we don't want to kind of the no allocating api's make sense only for people who really care, like an honest bill Rosalee in these type of people, I think the argument is that data aware that this is dangerous, API, that they'll rather don't use it at all. They have their own copies. So that span, you know overall, try not use. You know, help them at all.

P

Still work things up.

C

O

Why would we add a method, as this is a brand new method, don't call it? Why did we take the.

J

Time to write, if we decide to do that, if we decide to do join instead of add it instead of fixing combine, then we wouldn't want to add, combine overloads for span and then deprecated them immediate yep. No.

B

I'm saying I wouldn't do that period. I would I would I would say to me there's a true separate things right. One of them is I want to be able to have a combine like API that operates on span, and the second issue is I. Don't like the behavior combined has right. So one of the face you can fix, the combined behavior is to say: I did use a new set of APIs. All right, let's say join right in other ways. We fix it in other ways.

B

Well, we had this behavior for 15 years stuff like this to be able, we have and I think that they are like like. If we, if we decide introduce new names that aren't agreed, it makes no sense in the same version. To add, combined EP is that hitl operate on spam, and in that case you would just have to join once, but I think that that is I, don't like to me.

B

I would would make them as independent design choices rather than you know otherwise, like we cannot approve this issue here, like that's, that's effectively what it was down to the right as we tackle two things together, where one of them is a nobody, the other one is more of it. Oh, what do we do with that?.

F

Do we there's opinion that they are kind of inter mine, they're, like you know, related to each other? They don't want to separate and make separate decisions. No.

D

I heard that I.

F

D

I don't buy it.

B

Because I think that you know one of the.

C

Behavior thing is the most critical thing here, though, well.

B

You know the way you opt into the new behaviors by converting the state to the span right I mean you, so we could say the combined behavior to take some span has a different behavior that they combined that operates in space right now. Just because you know well, that's the sort of evil.

B

For me it is but like I, don't know like a conservative look again, the old one doesn't seem any less evil to me, because one of them is only for people that want to call the new API they're, the one it's affecting literally everybody who calls the API today.

B

Right so, like you already sub sitting into the people that care about our locations, which is you know, I will be a much smaller set of people. Then everybody close combined.

C

Could we start with a you know, fixing combined and then letting it seeing if it falls over like.

B

That I would like to do that over not even trying the questions of whether we can get an update to make this and.

I

The questions at all I mean I, can you know I can make the change to fit to quote unquote, fix path combined.

I

That's without causing havoc.

I

You're gonna say that last.

C

Part again, y-you mean in core right. You can try making the change. You know like I, can't even get it through like CI, that's a pretty big indicator but like if it it's pretty easy to unroll. It.

C

And you know giving it giving it a few weeks of actually being in production might be very useful.

F

Well, there's not really many people picking it up. You know what you're gonna get is coverage from our tests. Maybe you know I misspelled Roslin couple.

C

Or that those those are the key ones right that we care about. How often does how often does the PowerShell pick this up.

C

Does anybody know no.

F

Idea, prolly not too often, we might guess.

F

Probably only released versions, maybe RC.

L

And we done any quirks in Korea, not quarks. We've got compacts, which is do we're those up.

P

L

Of our jumping on me, I'm, not sure it's dumb ed. Is it a nap context thing as well yeah? They will probably be using app context. Okay,.

B

I want this I bump this back to you Remy, and that you would say you wanna do.

C

An investigation.

B

Over that you first.

C

So I'll make the change to add, fix the behavior and see if that actually gets the least are that part of it the are main gates and everything? Well all, but may not look it all. Look York working it as well. huh Yeah.

F

All but mate with the partners to make sure they actually do pick it up like they don't have like artificial delay or something we.

M

Have fun walking.

F

Things for Rosalyn to look at does the pipe options.

P

F

Wanted that currently, roughly this block from some reason, so sorry Turner, 30 miles well.

F

And via frozen video I can see the video building.

D

C

On YouTube, oh.

G

B

I have no concerns there. Anybody had any concerns there.

G

So there's no condition to begin with.

G

Steve we just planned a passing.

H

Unfortunately, the moment yes and we can change that or we can strip this off before we do or whatever it's.

C

Not uncommon, we do that in other places, so.

H

H

So we should try and pick a value, that's currently meaningless to Windows, just in case.

H

Yep no I agree. We should strip it off, but just in case you know every little bit of help. Protection helps yeah.

P

I mean I'm just wondering if once we're once we're adding our own to now, lock it.

P

C

We that currently, we throw on most of these things, I think it may be the same case here as well. So that's something that we already have defined in other file. Io cases we've actually like.

C

It or we check for things that aren't pass-throughs for things that are that aren't defined windows because people needed to use them. It's not a correct. It's not a great solution, but that's that's the pattern we've been using if.

P

We're already unknown values. That's that's really all that my my bar is on this because it right now we're if somebody can pass. Let's assume that you found a Windows API that doesn't also check its inputs and you pass negative one and we'll just set all the bits and windows, and we does will do something and that will have done something and now we're changing the behavior.

P

So so, if we're already rejecting unknown values and then because we know that we have a one-to-one mapping with Windows, then we're passing it that's different than where I assumed we were just blindly passing it. So as long as we're already guarding it.

H

I'm checking my honor I. Don't remember, I'm, trying to find it as well.

C

Where is it actually exposed.

B

To a shape, otherwise that would just well I mean it's so does implementation of a concern. How are your policy I mean? Lately you could have switched it to whatever you.

P

Or if we aren't currently being defensive against unknown values, someone could be using this platic. It could be being interpreted by something in Windows and now we will never let them specify that flagging it. So we need to if we're introducing it as a mechanism. This may not be the right.

O

Way so you're saying somebody coach today cause a new battery came through this thing and then you've, basically a flattened City to disclose. If we don't currently already get stuffing replies.

J

For Windows ever added a flag with that value in the future, and we want to consume it if.

H

It was here it's what we do. We do validation on the arguments. Okay, then it's good right.

B

We did something like by accident like.

P

I know that in in net FX with certificates, we had like three different api's that took a Flags value. Two of them checks that you didn't use on unknown flags and we had to court whether.

J

It was a quick simple as that, so this is essentially the same as the other proposal for adding encrypt Hardware intrinsics. Except this, the previous proposal essentially only covered the pact form of the intrinsics. This would cover the scalar form of the intrinsics.

J

What does the pact look like two term ones, basically, the same, except without scaler on the end of the name, was.

B

The things are the same of they're, basically different, of course, of a different nature. Yeah.

J

But there's several: there are several cases where you might have where you might want to only might be modifying the lowest bits, in which case you want to use the scalar form. There's also cases where you might be loading a scalar value and you or operating on it and you don't want to have to continuously be converting between vector and float or vector and double you just want to keep it in vector, 128, entire time sure good one cold water head yeah, so there's like at their square root, there's low, there's broadcast.

J

You can change those it to be basically.

B

Add packed square with packed. What do we think that ad, you know defaults to packed essentially because it makes the most.

J

B

J

Packed form will be the most huge case and the scalar forms will be for more specialized cases. So can we add.

S

An optional gun here as parameter which defaults through the pectoral, so that we only have half API signatures, so the bottom is set in zero. Yes, I think in practice you would have to do.

B

Something like we would have to make it an enum essentially and then there's another compiler thing that says this argument has to be positive as a ritual, because we have d-10 trinsic right, so we have to decide when we look at the at the call side which opcode stupid. So you cannot declare variable or like positive thing in whatever right, and so then we may not. Only concern is essentially like overloads with different made use.

B

I also, don't know how I'm going to call it is I mean like that that I don't know like if it have jelly speaking in terms of slightly more annoying to use it all.

O

Side then, just a different name that the signature yeah, but.

B

You know again, this is something like you know, get actually code. It probably looks already verbose as already because you have like a bunch of like you know how to do a plus B you have to do like I, seem to add past things in and then the more things you have to pass enable with more unreadable the already you put code kit but like it's hard to say without thee, the same I would just say like yeah. If we're happy with the default being packed and I, think that looks fine.

B

Otherwise it would suggest be renamed the other ones to be generally explicit about which forms we're operating under right and.

J

I think I think packed makes sense by default, because that's what most people use SSE instructions for and then scalar being explicit, is good, because those are generally used for things like normal floating point math or what a scaler actually literally just this for loop victory over the other yeah it does it well. No, no, so the packed form operates on all each entry in the indie vector, whereas scalar operates on a specific entry in the vector for for SS e and sse2.

J

That was only the lowest bits, but one of the later SSE iterations at the to specify oh I, want to operate on specifically the fourth value in the vector, for example, there's so.

B

And scanner- I guess that means you take that whatever the definition is the lower 16 bits or whatever, and then you already operate, and that is that what it does yeah.

J

And there's I'd have to look at the actual hardware guide, to remember how you specify how the overloads are shaped to specify which one you want to do. It might be an immediate bite.

F

To do bundle that dose as well or well.

J

Yeah there would be there there's literally hundreds of these, so I didn't specify yeah.

J

It seems of it is not yeah I think we discussed a bit about whether or not adding all these overloads would make intrinsic lookup more expensive, and we decided that with a binary based search, it should be reasonably fast enough to not be an issue.

B

The drones made our concern of like how many intrinsic scan behave. I mean that is all weighted to this issue. This is true, for you know the whole area essentially in which is you know the date, a little.

B

Whatever but I would politely nor this, and otherwise who constant why Carol Sumner middle says here I'm fine with this infinity, but we need to ensure that the vendor throughput us be expanded physically.

B

But that's general problem with you know all the passages of detecting critics. It's true for the entire area that we basically approve of expanding the sets so I would say this is to me an elated to this. One looks like so unless anybody has any other completing opinions. I'll just prove that well.

D

It's going going, oh great.

B

D

On actually, they can do everything.

B

D

Airdrop builds new all right, then I think that's it for today.

J

Evaluate yeah, so looking at the hardware manual there's a third xmm registers that it can take, which contains a bit mask of the values you want. The copy.

O

G

J

A tomato you would expose.

G

It as an additional.

J