.Net Foundation Design Reviews, 16 Apr 2020

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: .NET Design Review: GitHub Quick Reviews

Description

00:00:00 - Approved: Don't forget about JsonSerializerOptions.CreateForWeb() https://github.com/dotnet/runtime/issues/34626
00:07:37 - Approved: OSPlatform support for WASM/Blazor https://github.com/dotnet/runtime/issues/33328#issuecomment-614802442
00:47:16 - Needs Work: Guarding calls to platform-specific APIs https://github.com/dotnet/runtime/issues/33331#issuecomment-614826740
01:31:25 - Approved: New API: `InitOnlyAttribute` https://github.com/dotnet/runtime/issues/34978#issuecomment-614845405

A

A

Offensive the net, so today we are doing another review of our backlog. As always, you have some issues. First, that need to be done for the release, so we go over those and then afterwards we walk over itself up from all this to newest and then how far we get with that.

A

So this is a jason sealer options, one that was remaining from the feature: David I own, the call you want to walk to it.

B

Yes, sir, oh hey, we missed previous effort or making navistick the rutting us to add the robot by a speedo net. We should be called like pay for one something like that Oh and shall you wait actually that the only perk nice options but then now that we are also? We are also RC.

C

B

So put the mower.

B

B

Yeah I with the who are not an intent to have this on on systems education in space and especially on the agency related options with that mean he's kind of loose and doesn't kind of have any mean any something: it's not so much meaningful or the purpose of of their method. To insist, you.

A

Yeah so I think that does anybody end of it and understand the purpose of the API, because I mean it's not an API necessarily for us is just for centralizing. The settings.

A

You cool, then I guess nobody has any objections.

D

I mean the discussion that we had um that David was talking about was. This is very specific to a speed meant, and you know a crate for web. You know: does that imply if P net doesn't apply the current version of ESP net in order to change these settings in the future? um It's not very.

E

New it'd be where you can change it thinking you add a new setting here, yeah, so the that's the thing right. So the idea so I named it specifically.

A

Not like super specific to asp net when I proposed the name, because it's kind of like so the reason why I said for web is because it's effectively the default that most people use in the web right so like the it's, the it's the casing policy, for example, of camel case right, because most people that exchange stuff on the web there come.

A

Consumers are JavaScript based right, so they often want the the camel casing policy right versus the normally it's about the options that we have are just basically you know, in.net, you see lies in DC or let's just plot met objects. You don't really care about the naming policy, so we just set it to pascal right, and so I think that's why we went with for weapon then again like I don't think it's been able, make it break and change the either.

A

So even if they expose something new, I don't think this would be necessarily changing the default settings either right, so it would be as stable as whatever they have today.

F

Think the at the feature meeting we discussed that they should be calling this one and will acknowledge that they can't ever change their defaults again to be consistent with the package or the or the scenario just doesn't work.

D

I suppose those new features get added, you don't think and they can change the default if they need to.

F

In the version they get added, yeah right.

F

A

A

Did they did Ryan ever push back or anything? Sorry did you I, never push back on anything like I mean being better than the spec that they call the API and they seem to be okay with that. But then I haven't already need any anything since then, so I assume that's still the plan yeah all right, then um this goes into API approved and that no means we can remove the blocking label.

B

Sweetly and where it is right now right.

A

I would be I mean I've knowed. Anything else.

B

A

If you have concerns with the names, let me know like we can change it. Yeah.

D

A

I mean: are there any other ideas for names.

B

Well, I I, just I just copy what you put in there do. You know.

F

Right I mean just to you know, say things for the purpose of making sure that we said them and know that we're making an explicit decision.

F

We frequently don't like having public constructors and public factories on the same type, because people who find one expect the other one doesn't exist, but I don't think we have a better answer than this one. So it seems reasonable to me.

F

D

Yeah, the only possible way to do that would be to add some sort of bool or enum to the constructor. That says, you know Creed for web, but that doesn't seem any better to me. Yeah.

F

I mean this has a name, so it's very clear the purpose and you know we can slap documentation on it and and etc. I think people who are doing their own serialization. They may end up defining exactly a redundant copy of this if they're looking for it because they don't see the factory, but you know using the constructor gives you power using this, gives you a set of pre-built defaults, but they're all settable, so you can change it after the fact. But it's or can you is this gonna be a pre-sealed options.

D

You can I mean on date great David Gannett in, but we didn't talk about any any locking down. So if you call this, you can change the options or set different options until you do the first to realize, or do you realize just like today with the normal constructor so.

F

It's gonna return new. It's not gonna return the same option every time. Yes, I.

A

Mean that's why that's one of the things I discussed with Rhian at some point is like and know we need to be able to modify the instance right, because the most common thing you want to do is register some sort of converter, and so we took basically the ideas you create for web, and then you get your own instance to mess with. And yes, if somebody is an idiot- and they said the casing polish to do something else, then that's too bad sure.

F

But yeah it might hit something how I was deciding this returning a shared option, setting or but as long as it's returning new yeah, because it's it just gives you a better starting place. If you're doing the web scenarios that the the factory gives it a better name said I, don't think we made a mistake in the past with making the constructor be public.

A

Although I would say, like the Steve just said of the enum, it's not that bad, actually, you know, might be more discoverable right because when you say new Jason see lots of options, and then you see over lots of the construct and one takes an enum that basically has two values like you know: I don't know standard and whether right like that would not be insane right. Then you kind of get hinted that all there's potentially different default settings. So.

F

You want item Jason, serializer defaults. Yes,.

A

Something like that I mean it wouldn't be I think that's actually for my UNICEF I'm, probably better than the factory.

F

F

That seems legit does mean that we we do reduce the concept from factory plus constructor to just constructor and constructor overloading so yeah. We.

A

Just so what did you just said this? You said public enum Jason see there the default yeah.

F

And then I don't know what to call our current thing other than default, which is a bad name inside default, but it's not flagged. So it should be a singular aim on the enum type.

A

Let me just or.

F

We can decide that it's is singular in that it represents a collection of default things. So defaults defaults feels more natural to me. Somehow yeah. Then we would have you know I guess, yeah, whatever we call the zero value, which is what we're already doing and then the the next value would be web defaults, but again, defaults has redundant so web.

B

So the default, the zero value will have the same effect that just doing New Year's insulative yeah.

A

So something like this, you.

F

Yeah I'm not sold on the name of generic but ya,.

G

F

I had no better name I.

A

Don't have it like if somebody else is a better suggestion. Steve will be like that. It gives me a new propulsive.

D

I think I guess maybe it's more discoverable because he has you're typing intellisense will tell you the overloads for the constructor. So from that standpoint it's more discoverable um I, don't know about the naming. So much I mean generic I would just call default. Cuz I think that's what we do in a lot of other places, but maybe that's not what we want anymore going forward. I'm, not sure it's just a little.

H

Weird to have default, stop default yeah, you know, I, sorry, I missed I missed the beginning of this book. What was the criteria that led us to have the defaults? We currently have that might impact the naming.

A

I think when I discuss of awesome, initially I think one of the reasons was perf, because if you have an accept, if you have, if your kids inside, if your case sensitive and pass good case for some reason, that I think there was some benchmarks there used to be better. And so we went with that. Well because.

F

You, don't you just get to use the existing string table for the property name and an ordinal match is always faster than wait.

H

Wait, we picked our defaults based on per fry than on desired functionality. Have you looked in.

D

Turkey, you know, there's also Jason is case-sensitive by default. Honestly I think that was the number one reason and Perth was like. You know secondary those discussion on her remember, and so why would what do we want different from Jason spec by default.

F

I mean the case-sensitive comes from the speck. The naming policy is the easiest. Naming policy is used, the name you gave us.

C

Yeah and and to the points that Stephen Jeremy just raised to like we, we just allow things like trailing commas by default unless you walked into it like that.

C

Things that aren't spec compliant, but.

F

Since there's a an opinion, the naming policy is one of the things that changes across this. We can't just it's not Jason, serializer defaults. You know strict or or something because there's an opinion involved as well.

A

Yeah I think the profile specifically for therefore the case sensitivity. The other ones were either aspect. Compliance or just you know, security faults, I think that's one of the rationales we had let.

H

Me rephrase my question: what scenarios do we expect people to use the existing defaults for, if it's not talking to stuff on the net.

F

Talking to things that aren't the browser.

A

Yeah I mean, like my kids, for example: I just use it just to see all as settings and store them like I, don't I mean I just use Jason, because I can actually debug it and read it, but I don't exchange it with anybody else. So that's like a local yeah local to file, or you know, document to be when you're distorting the database and you don't actually care about what odds mean forward.

F

Or your front-end back-end and they're both net.

A

Yeah I mean that's the discussion I had with awesome right. If you look at the internet, for example like there's a bunch of stuff, for example, the the github API is have a naming convention. That kind of tells you that they use Ruby because they use the movie naming conventions, but they return in Jason right. So it's a it's not uncommon. For people to have you know, client-server just talk in a language that happens to match word implementation language they happen to use I mean.

H

I guess I'm wondering if we call the default like do T any team. You know have it B.

H

Or local, or something that implies like that, the scenarios for which it was intended standard I like to like that's what somebody citizens.

A

In the in the chat, I, don't know like I, don't like standard it kind of implies that this is better than the other ones. I mean that wouldn't be insane. It's just a little bit weird, because I think it only really makes sense with the naming policy right. Everything else is just I think the I spin it, for example, set different settings for you know the MEXT recursion depth that they protect, against, which I think is lower than what we have, which is nothing to do worked on it.

A

Yeah I mean I, yeah, I, can't think of anything better than generic. To be honest, because default seems weird well.

F

I actually think that net might be a good name, because you know if we added you know, you mentioned that it sort of leaks that github was using Ruby right, like we could conceivably later get enough demand for hey. Can you give me an accelerator for the Siri things that I can write the class and dotnet and it will see realize looking like Ruby? Did it etc?

F

So we could end up with dotnet and web and Ruby and Python and brain-fuck, and you know, which would I guess just be a series of pounds and exclamation marks or whatever you do in that language. These days, much.

H

Required for done at five I think.

F

But yeah so at least that's saying like this represents the defaults for or that these are the defaults that you get by default in dotnet, and then we have that gives us a room to expand for saying this is what the defaults, our equivalent defaults, are for some other platform. If we choose to do them in a future, you.

H

Or don't care, you know the first time we try and have an apostrophe in a name but I, don't know for generic I, don't care yeah.

C

If we wanted to use, generic I would probably prefer general over generic, because generic is a language term.

A

A

Against general, wouldn't be sent, wouldn't be bad if you ever have something that's specific to I, don't know that assembly or something like you could imagine. We have general that pep assembly right that would be not insane choices.

A

Fair enough, I mean the question also is: do we need an actual enum that we expect us to grow, because you could also just have a boolean right, but.

F

The bully, like we've, talked to we've, already talked about possible places. This could expand.

F

So I think that being and doing the enum now is better than finding out next version that that we, that a boolean was bad plus the the enum makes for much better descriptive code.

A

Yeah I agree. Alright, then I mean let me update the thing. I just approve.

H

Sorry one question: Jason serializer defaults or Jason serializer options. Two faults: are there going to be other things that where they have things forward, we just sort of assume that the only thing defaults could have applied to our options and therefore these go hand in hand.

A

Yeah I would think over the letter because the you know defaults kind of our configuration settings right and the only thing we have four traits, no jason sewell I don't want.

F

Yeah and then Steve as I I think I did the second-guessing myself with singular versus plural right before you joined. How do you feel about the plural name on that.

H

Tom defaults.

I

H

Think the singular sounds really weird right.

F

And to myself it's because it represents more than one setting like they're, getting the collection of settings and right. So it is a singular term for a plural concept. Yeah I mean we don't need to go all Gollum and make the make up flags version of this mean default, says.

F

F

All right: do you feel better about that? One David.

B

F

All right, one liner in half an hour, we're having a good trajectory today, Oh.

A

In my defense he approves the delivery and then we had ideas right all right, I posted a comment. So all good. So we are happy now but I still like D forces, I, think that will be a great API I'm, very supportive of it. That.

F

One needs to be flags, flags, flags.

I

G

A

I mean, then, you have to make a component and.

C

Then, for what will be processes.

A

Yeah, so I could talk to that because I filed it so for Blaser we need a way to identify the place, a platform, and you probably all aware we had I, don't know 100 gazillion discussions about what that actually means like what is blazer alright. Is it? Is there such a thing as a blazer platform, with a blazer operating system? Is that you know?

A

Is it web assembly, and so what we ended up with was to say that which we do wisdom itself as as an architecture similar to x86, because it's effectively cogent technology and then, logically, when you, when you're executing a blazer, app or a web based app you, you ever hosted, execute you your video pep assembly and we treat that host as the operating system, and so today we only have browser.

A

But in principle you could imagine other environments that are not the browser that executes that assembly right, and so, if you ever need that we can, we can add that it's also worth noting that we don't really corner ourselves too much with whatever choice you make here, because we don't really have a way to ask for what the current West platform is. The way that it works is you say: runtime information is a West platform and you're passing the one you want to check. So, if later on, we have, for example, more specific OS platforms.

A

Let's say chrome vs., Mozilla vs. edge. We could expose them as well and then we will just say yep. If you ask us to be a browser, we say yes, but if you also put the VM I would also say yes right. So this is not necessarily like a cornering decision here, but I think that's kind of the the minimum we need in order to make that assembly available, so that people can make runtime decisions based on whether they can call certain API is because you already know that certain API is won't work.

A

It webassembly, and so that would be a very easy way to check that oh you're running inside of a browser, so you know don't try to create threads or not try to write to the file system.

J

Is OS platform the right thing here.

A

J

As opposed to, like any other any other thing like OS OS stands for something right: yeah.

A

I mean you could do that, but I think the question is what I mean you know: what value are we generating?

A

Because, fundamentally, the reason why you want to do analyze check is because you want to call an API that might not work right and so the general mechanism that we have exposed, so that is runtime information. Dot is OS platform right so that so by exposing that and in that fashion, seems more. You know self consistent with the other west checks. You would have to write like.

F

If we call it a a a platform or operating environment, this would fit so I think it like it works I wonder if we should expand wasm to web assembly, though.

C

I think we showed yeah. We tend not to use acronyms like that. I get it.

F

Passes the websearch test of, if you search for wasum it does, the top hits are to expand it to webassembly, but it just kind of looks weird and I. Don't think people are gonna type it enough that we care about saving this sixish characters.

A

Yeah able to me, I don't have, I think, yeah I just copied a pacer, but somebody suggested I mean in fairness, though, like architecture is already like a graphic of things. We really have x86 arm, and so it's not like the most readable names already, but I also have no objection to the web assembly right.

F

But those are more charade, marked aims and and such verses this one would be like I mean you know, pick an assembly language language is for x86, and you know: wasm is just one assembly language that we're using as another intermediate. So it's I think that we should yeah I feel like we should expand it. It's not yeah, it's not the trademark.

F

C

There is al Azim that comes up in the future. Would would we want that to be called lazon, or would we want it to be called webassembly.

A

So I said again: wait. What is it to come from so.

C

Say that say that there's some capability that gets added to it in the future, but they have to come up with a new language design, so they call it was into instead of wasm, because it's not fully backward-compatible at that point, do we end up with architecture having a bunch of different fields in it that says: webassembly webassembly to webassembly, 3, blah blah blah? Or at that point, would we say: Oh like this is getting a bit verbose. Maybe it should have just been wise in the whole time.

C

I, don't know what that's likely to happen and.

F

He would depend on this.

K

That's pretty likely to happen is actually underway, so there will be possibly or wasn't 2.0 some point. I.

F

Think that so looking at the things that are in architecture right now, which is arm arm 64x64 and x86 like those sort of go along with our level of you, know the sort of rid except. We also include the OSS in the rid, and so the question would be is why azzam two fundamentally incompatible with masam? Or is it just? We've they've extended it, because if it's extensions, that's versioning and not not an incompatible architecture. Right I mean I'm aware that x64 is a superset of x86, but.

K

You know we will probably use this term as a break as well. So if you kind of standardize in architecture possibly, then you might force people to actually type you know, possibly in pedigrees if they need to kind of specify who rates, which might not be what we actually want.

C

You think that red will map exactly the the letters that you type in a red will map exactly to the letters that are present in this field. Well,.

K

Good like strange, if you use you know, but somebody at one place a Muslim another point I. This is basically for the same stuff. It describes the architecture and the wreaths are about the architecture right. So they kind of map pretty clearly today to whatever be I using across the paper.

J

Today and get run time, identifier is get the current or architecture to string it to lower okay yeah. That's.

C

Reasonable I'm.

F

Gonna say: there's a difference of the enums are in developers faces and rids are mostly hidden behind nougat packaging infrastructure. But if we're gonna have a read generation problem, then I think.

A

In fairness, though, like I, think, the vast majority of users will never care about the architectural. The only reason why you might care is because you are loading assemblies yourself, somehow or libraries you solve, in which case, if you do architecture checks, you're very likely, also having to understand the directory structure of your package in order to make sense of it, and then you basically come from the book both naming conventions, no matter what you do.

A

C

C

You I had a question on the on the OS platform stuff real fast. You had mentioned that there's an is OS platform API in order to do a an OS version check.

C

Is it reasonable that, if somebody would ask, is OS platform, Windows is iOS platform Mac and expect that to return the correct answer, even if they're inside of a browser process but.

A

What would they do with that answer? You can treat the thing as Windows, because you can't P go for anything, no.

C

But you can, you can do things that are specific to the operating systems. For example, if I know that you're coming from an iOS device, then I might pop up a notification saying hey download my app from the App Store and I might target that, based on what OS platform I believe you're coming from I.

A

Would say that that would be a different question than but I mean in principle. We could I mean so there's nothing fundamentally broken with the design. We could only have it's reasonable for us to return. True, for if you ask about the real Windows right like we can beta return true for browser and for Windows, but the problem of that approach is how much more code. Do you break by doing that, because I think it's a reasonable assumption that most code takes when is Windows returns.

A

True, is that you can be invoked or anything that is in their given version of Windows right, not.

C

Necessarily you you do often still have to do a capability check separate from the OS.

F

Depending I mean, if it's something that's been there, since you know in t1, then you're pretty well good to go. Yeah yeah I mean like that may mean that we want either OS platform. Dot is host platform which would be what's running the browser and then the standard console things would just return the same answer, but or that it would be a something webassembly specific somewhere which maybe still works in terms of OS platform as input, but I, don't I, don't think that we would yeah I share you most concerned. That is OS platform.

F

We essentially wrote it for these classes will or won't work. These P invokes are aren't insane and in making it support. Both concepts now feels like it's just setting people up for failure. Okay, that's.

C

Totally fine just wanted to raise the topic.

I

I will just add to that that beard like there are similar emulation kind of environment sites like when you are running vine on Linux in a VM on windows, server rights like each level of the emulation you paste.

I

You know, there's no yeah.

F

Determining what host means is part of why I was thinking and something specific to a azzam scenario. To answer that question, because at least it can give you a clear and concise answer of what host means, curry yeah. So for my.

C

Own edification, then this API would just return the thing that is closest to you in this particular case, the browser virtual environment, right.

A

I think Johnson. Allergy is quite interesting because if you, if you really want to do the inception kind of thing, but you want to know what's outside I think then it's okay, if you have to call different API, is right, because, basically you have to decide how many layers that were you aware of right.

A

So, for example, in this case I think it's reasonable to say you know, if you say it's a platform browser, then you can say: okay now, I can ask some other API from you know that blazer folks provide you, for example, to check the browser string right and say the the browser user agent string. So if you want to know that you're running on Windows, you can still figure this out by looking at that.

A

But you know: I have to progressively know more about your outside environment right, because otherwise the whole point of emulation is to kind of lie to you right to give you the illusion that you're just running you know in a particular environment right, so that that's why it's hard to like give people. You know the the red pill. If you will right like that, you have to kind of opt into the how many red code that you're aware of I.

F

Mean I can't see the browser caring slightly. What host environment is because it would be things like suggested path, presentation and like where settings like there. There are scenarios where you're trying to tell them how to use their own come now to use their computer outside the browser. It feels very nice, so yeah I would say that that's a different API I.

C

Don't think it's really.

G

A

Think the most common thing is probably download so I'd, like you just say like if your windows I'd point you to the Windows XE rather than the you know, the the Mac OSX II right like, for example, what dotnet does right if you go to do to any T like we tell you or don't want, don't need core for the operating system here right.

C

So maybe maybe we just at our level, don't worry about them and we just say like we'll, we'll deal with it. If the question comes up from customers.

C

D

J

Look at environment, oh s, version at all, there's a platform ID in there.

A

I'm unit, so we had this discussion recently on the on a side. Conversation yeah I, want to I want to talk about this one as well, but I would like to make that a different topic of conversation, because it's a it's a bigger issue. So there's a few things but there's one then it's already around an environment. A lot of people are using it I'd looked at usage recently and it's actually used a fact on to use the technical term, and so the question is: what are we doing with that?

A

I think I think the version number you know yon already corrected me that we could that we were took the correct version number. So we no longer return bogus information from there I think the platform ID is still somewhat broken, though, because that already has value said example: I think it has Mac OS or something like this, which we, which doesn't return my co s on on Mac OS.

A

So there certainly is some issues with that API, but maybe there's some some part of it that we should just accept as it's current, for example, the version number we could say it's current, but then we say for platform ID we just obsolete that one and say: please don't use that one, because it's no longer the right information, but there certainly is overlap of environment OS version and for better or worse, we've created the other one, and so we have to reduce it overlap. Somehow.

J

It so what are we doing for platform ID on wasm or we're just saying it's obsolete and it's it's undefined. What it's going to return.

A

um I'm not sure I understand what you mean by that when.

J

I call when I'm running on resumes and calling, but that OS version platform what value am I gonna get back on them. Well,.

L

That's something I think I'll throw in 32s yeah.

A

I think so, there's two sizes there's the version number itself, which is also not clear to me what it would return because of probably the the return. The elective, like the actual version of the browser's, probably also not very well defined, but I think like if you ignore browser for a moment by the question, is what is OS version return on on the neck, so Windows today, right so like there is already there's some things where we return correct information. There's some things very down. If we just have to think about that.

A

So, but let's make that a separate discussion. That's let's ignore, as version for now, because it's a different thing.

J

You okay, I mean it's named weird that it's OS version, but it's also got the platform ID in. It is not just the version right right.

A

But like the reason why I would ignore it is because it's prior hard right for better words. We have both environmental US version, which is already shipped to the table. We can't change it west platform and in top services has also shipped in this table, and so we have both concepts, no matter what we do right and so like one of the problems we have to do is you know, reconcile this overlap.

F

Right which may mean that in five we need to add a platform ID unknown. Just so there's a bucket here that we dump everything else into in the future.

F

Because it it didn't, it used the zero value to have semantic meaning, instead of represent unknown, which is part of its biggest or part of its problems. You.

F

F

Yeah I think what's proposed is good, but I agree with Eric that we we do in this release need to understand what we're going to be returning for the other API and maybe figure out a permanent we're done with this solution to platform. Id.

M

Is calling it just browser enough, like? Is the web assembly standard like solid enough browsers? You know they have a history of varying behaviour.

M

What would you call it in the center? Well, do we need, for instance, Firefox and chromium, and things like that in here or is, is browser truly enough? Like can someone make meaningful choices off of this, that they don't have to do feature detection elsewhere to do to figure out.

A

So my position on that is that I don't know, but we start with the most conservative approach, which is we just give you the way to check that you won in the browser? If we need the other browsers, we can add them later there because of the way the the OS checks are being done. You can't ask what you're running in you can only ask. Am I this thing so we can return is true for both browser and chromium. If you have to, for example, or.

F

It's like what what host OS is this which browser engine is. This is something that's a browser, OS platform, specific API somewhere else.

A

F

Those are both sort of user-agent scraping at that point.

J

And we said when we're running on wisdom like not in a browser, this will return. False right is is OS platform. Browser will return false if you're running on the server in wasm and that support it. Yeah.

N

M

um So there are waves of environments that aren't a browser.

M

What, if someone's running on a desktop like what? If we have blazer on the desktop or something through through one of the JavaScript platforms, there.

A

So you think it looks on the collection, for example, I mean in that case, I would I would think we would probably had the other thing whatever it is. That say it's electron everything we would have browser and LX one will probably return true for both of them.

A

Everybody want to know because I think, fundamentally, we will ship this one first, so most of the code that will try to make work instead of you know, webassembly would probably just check for its browser and then later on, code might get more enlightened to say all turns out, I'm overly aggressively exposing functionality that is actually available and when I'm running against an electron, then you just add the other.

A

If step that says, we know, if you know browser then do the other thing unless I'm also an electron which is in which case it will work right. So that's you know you just make incrementally more stuff work. This way.

F

But the architecture in that environment would still report wisdom, yes, which is why it's a different vector? Yes,.

A

I think probably they're both useful right I mean whether ms is required in order for you to package up native dependencies in being able to load them right.

A

For the same reason, you need to reuse x86 today, so I think that's a that's kind of an orthogonal choice, but then the OS platform is I. Think most before you know degrading your experiences in a library being one set of laser in a meaningful way.

A

All right so then I think we just concluded that we don't change the API at all. As proposed freeze, we keep wisdom as the name for the architecture. One.

J

More stupid question, sorry, is the writ gonna be browser, wisdom.

A

If I don't know what the writ plan is America, you know.

J

Because, in general our rate is like the OS name, architecture right and the other OSS are like the next Windows OSX, etc.

J

So I'm almost wondering if lake.

J

Same pattern, where OS platform goes first in the raid, and then architecture goes last. If we're gonna name, it browser wise, ohm, we're gonna name it something else, then maybe this name.

J

If the answer is on, nobody knows, that's fine, I guess. Let me know what we put at the beginning of that rig today. The truth.

A

I mean I mean I. Think browser wasn't would not be insane right because that's kind of what we do today right.

J

I mean if this is what we're calling OS platform as browser I, think it totally fits that that that is what we would want to call, though right.

D

J

A

Yeah I think Mara doesn't want to call but like yeah, we can I could just add a note and says.

A

Let me actually tag American men. You.

A

A

A

All right so then,.

A

All right remove blocking, remove a parody from review and API approved.

A

K

A

All right so- and that was the easy one- no, no brace yourself another controversial one. This also the remaining hour. Probably they are not recording trajectory. So, oh god don't touch. My iPad, apparently I think I'll have a minute of that.

A

So basically you have the situation really want to provide people with an analyzer that can tell them when they use ap eyes that are not available to them right and that can be either you're the wrong platform, because the API is Marcus window specific, but you're, actually on iOS or because you're on the wrong version of iOS right. So the idea for building for iOS and Android is generally speaking, that you compile against.

A

You know the latest version of the SDK that you have available to you at the time, you're politically application, and then you may decide that you want to run on all the versions of iOS, in which case you add a set it to your project file that says, I want to run all the way down to, let's say: iOS, steady, but UDS together, you have as far as 12, in which case every API, those introduced in 11 or later will be flagged in wiser and say you know you.

A

This API might not be available to you. So please go out the call, and so the way they analyzed it will work is that it will basic check whether your API, your type, you were you know containing symbol. If you will is itself markers where specific, in which case you effectively just shift the burden to the caller of yourself or you have a you know, code in your method, body that basically causes API and says you know, am I running on iOS, 11 or higher right, and so this is basic model.

A

By saying one time, information thought is always platform or later we already have its platform, but then the West platform or later will also allow you to pass in the version numbers. You can quickly check whether you can, you know be not you your minimum is that version number interesting single tee with the SDKs is that you know, as you probably all Apple makes what we will consider breaking changes and that they remove a api's then will generally not make incompatible changes, meaning they will not change the shape of an API.

A

So that's what you can, generally speaking, always compile against the single SDK and just guard your method calls correctly and so indicates where API is removed.

A

You would also have to do the inverse check, but you say: well: I only call this API, if I'm on, let's say iOS 10 or earlier, because the API was removed in iOS, 11 right and so the way we do, that is, you would say, is well as platform and earlier or I think I already renamed that based on spec feedback, I think it's called West is well as platform earlier then I think it's what I expect right now, but you know the idea there.

D

Is you basically pass him the the version number.

A

At which the API was removed, and then the API checks that you are in an earlier version than that, and so this way you always pause the diversion number that is actually you know the that corresponds to the annotation of the API and you basic get the correct outputs of that.

A

That make sense to people.

A

F

Anchor details.

A

Yeah like, if you have yeah.

D

Let's first talk about the concert, because let me just quickly update it.

A

Based on what I have written in the spec, so we actually talk about the same here. This is the problem in your copy paste content.

M

You so if we, if we have this attribute on various api's today, and we find a way to make the API cross plat in the future, is it safe to remove this.

L

Yeah, so that's a general question that we should talk about.

A

So also coming up in the spec, so the way the spec was written was very focused on iOS and Android and to sort extent Windows as well, where you really don't talk about general-purpose api's, you want to talk about a very specific API. Let's say you know some API in the iOS SDK or some API and the Windows SDK, like registry WMI, things that are inherently we're, specific and they're, inherently not portable.

A

They only exists in one OS, and so the only thing your model over time is effectively the version under what the version I was also given right that you know individual version of the platform this API was introduced in this stays the same, and so the only thing that can change over time is that you started obsoleting it in a later version, or we move to the later version right and that's what this feature is really meant to address.

A

There is this other concern that we have today in the platform what we have some api's that are inherently portable, but they don't support it everywhere. That's a good example: here's or favorite you know console window with right, which apparently we can't I forgot, can't get or can't set on on anything but Windows. And so the question is: how would we model that? And my personal opinion is that it's very hard to model that over time, because the premise of the feature is annotations in the reference assemblies.

A

So that means, if you, if you target something portable, we can only annotate what we know at the given point in time right and then you can, you can either you know, have a model of a inclusively say: there's API supported on these platforms, so you can say this API. It's not supported on this platform right and there are cases where one is better than the other, and but it's most single, you know single best answer I, guess so.

A

I think my general expectation is that I wouldn't use this mechanism, for you know portable API, so let's say reflection. Image program we would say reflection amid, is a feature that it's not available on all platforms: the set of platforms that the feature might be available to might shift over time. So we don't have a static knowledge about where that is, and so we just asked it to call the capability API.

A

But you basically asked at one time: is this feature available or not, and I have another spec for that that basically talks about how we would mark capability api's in in the same way. So the title has a coudn't principle check your code and check whether you are checking the capability, but if you go through the motions, I'm, not sure it's always worth it, because it could become fairly noisy and the question whether we are better off than that.

A

But so, if you think of this feature that I'm that you know I've on the screen right now, it's really more about what is the experience of an iOS or Android developer if they want to make the application work on an older version of the platform rather than I'm a donate class level, developer and I want to make the API you know as well portable as I can, because there will be some awesome fidelity by doing that.

A

A

You is the silence sign of agreement, disagreement or confused.

M

By that for ETL Debs, maybe we make a choice that only ever use this specifically on very platform platform, specific things, but I also wonder if users would would fall into that recommendation as well. If there was a way we could influence them, even I, almost wonder like we have some api's on sockets that are Windows specific today, but we have a peak era up. That is updating to add support for other platforms, so.

A

Yeah I mean this is a discussion that we had a discussion with Steve and I. Think I forgot things particularly ago. I. Think the example was there's a pep talk of the API, not sorry, there's a named pipes, API right. If you pass in the name for the pipe, it only works on Windows if you don't pass in a name for the pipe that works everywhere right. These are examples where you.

D

Know attribution for the API itself doesn't help you much because.

A

It also depends on what values, you're passing in and so I think in order for us to tackle that we can either do one of analyzers, where we just check for particular patterns in the code, which I think it would work for widely use. The API is what we find some sort of mechanism where we can. You know, extend our attributes so there's a personal ability where we just put in enough attributes in the in the API surface that we can have generic analyzers, they can understand what what what works and what doesn't right.

A

But the problem here is really the last mile right. If you, if you shoot for a hundred percent, it will be very hard because there's a lot of nuance for you know. Particular things don't work the same across all operating systems right but I mean the example that we have here is that if you're building an iOS, 12 app anyone run on iOS 10. You know. Unless we do this feature, you have zero way of knowing that you call an API that was introduced later.

A

You just have to read the docs, which is not very satisfactory and that one we can relatively easily check.

M

Yeah, it's I've got to run to another meeting, Thanks.

J

Sorry go ahead: tanner I.

O

Was going to say, has anyone talked to the windows SDK team to see if this will work for their stuff, because they've got a Microsoft Windows SDK contracts which exposes the Winton version, specific api's for use in general purpose, dotnet apps, for example, they've got the 17 134, the 17 7-6, 3, etc, packages for use and I imagine this would be a perfect example of where they want to use the attribute so that people can tie it into this check, et cetera, yeah.

A

So just catch people up, so my understanding is the way when Artie works. Moving forward is that we basically so today the way it works? Is we import directly? We do study every now.

A

Well, not window stop William D, but start out with MD files right then we have some sort of projections that happening at the minute, either side and at the runtime side, and in the years that there will be a tool that effectively it generates c-sharp c-sharp net specific view over the 180 surface area, and they already have version attributes in the vendée files and so I would expect us to work with them and to make sure that either we literally just produce our version information levy, spec here or the REO analyzer just understands, whatever version number they end up having, but it will be very similar right I mean all the API is already annotated with the version number they are introduced in and that would be.

A

You know something that we would light up here as well. So logically, it fits whether they actually ap apron. You know api's themselves for 100% I think is something we have to figure out.

J

Reading the windows doc, it makes it sound like you, shouldn't, be checking the version, and instead you should be checking for capabilities. Like is this API here.

J

If this is just for iOS and Android, would it be better if these api's were iOS mandroids and not every in the world? So.

A

The way I see it is that I don't buy capability API it's for that level, because the problem is unless somebody designed capability API that you can check what api's are you looking for so, for example, take our favorite example right like Ram, it is not supported everywhere, but we have like what 12 types of 14 types and like what total, probably hundreds of members that are directly applicable to recommend. So if you were to write this code against Windows, what symbols are you looking for?

A

You have to pick some symbols, but nobody tells you which ones you need to look for, and it's famous through the JavaScript world like if you check for the presence of an API, you never check for the presence of all the API. Is you always end up picking some sort of proxy that you say if this type is present or if this method is present, then I assume all the other. Things are also present, and that is very fragile over time versus the nice thing.

A

With this version, checks is that the onus is on the SDK author and the SDK author is responsible for telling you for every API, which version this number of this API was introduced in which know means that the correlation is very trivial, but the correlation is over. The version number.

A

The problem that Windows always had is the people get version checks wrong all the time, and there was basically no tooling right but, like I, think the idea here is that what we have the tooling, what the analyzer will tell you if you do the guard checks incorrectly or whether you check for the wrong version number right but like like aesthetically speaking, we have that information available. Then we can. You can surface that by addresses. If you try to do simple, checking like what zoom was.

A

Are you checking and basically there's no coding pad or like mechanisms? You can generally apply to check for arbitrary things about checking every single API. You are calling right which doesn't work in general.

J

Was answered, though, thank Y? If, if this is for iOS and Android, why aren't these iOS Android no.

A

I, don't think it is I. All that said well said the feature was designed specifically because that is a requirement that comes from iOS and Android, but based on more experience of linearity. It equally applies there. We can say, as I said, they are the same mechanisms right. They have the they already annotate, all the API surface, with version numbers, and there was at least community parties. It did exactly the same thing that we are doing here.

J

So Windows doc, are you not to do that? The shock say this is how you should check for it's a thing right.

A

But if you leave it the docs, they also check don't check for the best version ever right. I think this is the. This is a general problem that we have of the platform. I think that people have this knee-jerk reaction, that version checks are bad, but I have yet to see a counter proposal that has a better usability order and- and you know, fewer issues right so I, don't think this picture is iOS or Android specific I think it equally applies to the Windows SDK all up.

F

I mean just so, we can be talking about it at the layering level. What does it mean on Linux.

A

Well, I mean the question like what is the Linux specific API right, because realistically we talk about Linux this. Basically either you talk about POSIX, which you know ideally is everywhere or you talk about, like you know, the arbitrary API changes that happen between Linux distros right, so I. Think if you want to talk about a Linux and I, think you basically have to talk about a particular disk. Alright, so you could say you know we can.

A

You know you can effectively model red head over time, if you, if you, if you wanted right, modeling Linux over time, is a bit weird, because there really isn't a definition of what the next actually is. There.

F

Is Linux is the kernel, but usually at your pee, invoking you're, not you're, not asking. Is this syscall going to work right.

A

And I think like yeah. If you look at Linux as the s, just the kernel then sure, but that also wouldn't be bolded very useful to people.

A

But yeah I mean, like I, mean Android is a form of the next right, so in that sense I mean it equally applies there. Assuming you have a distro that that has some sort of you know binary backwards, compatible thinking right and SDK like any that somebody needs to give you the ad in the information which which version of the platform this API was actually introduced in right. So that's something that just needs to exist and that doesn't exist for all of Linux, but the dead thing that exists for some distros well.

F

I mean like it: we're gonna have to have a behavior somewhere of what any of these things do on a boon total a boon to 2004 right like are we gonna start a distro specific OS platforms? Do you do you use the fact that it's just a strongly type string and ask a boon to I mean, or maybe it's just that we really don't expect anybody to need to do it on Linux and well, we'll need to have a behavior to find right.

A

I mean I mean the behavior is kind of hard to define I guess in general, but you could just say you basically forward the information or not forward. But if you basically extract information from you name and then have some say, mapping right and then your special case. You know the the actual things we know about, but like OSX FreeBSD and why do I like known known things we have and then the rest is just you had a long tail.

A

F

Okay, so given, but let's let's take as a given that that we want to do this, and then we can talk about the specifics and maybe that'll change the given right.

B

F

Now that you've updated the OS platform, what I was gonna point out earlier is Apple this year: split, iOS and iOS and iPad OS. In addition to the already split out TV OS watch OS right, so we probably need iPad OS and now we need to understand the version relationships between all of those which is gonna make that complicated. But that's this libraries problem and now you've added Mac OS. Are we equivalents in that with OS X, or are they going to be something semantically different? Oh.

A

That's a good point, I think I, just probably just a copy-paste error. I would probably don't do that I mean so I by the way, as we talked about anything from talking to the German folks, is that iPad OS is a marketing term. It's not actual a destroy in itself, it's effectively iOS. So basically they said. Don't wanna lie to us yet, so that's why I base it didn't bother. I mean if it becomes its own distro or its own thing. Then yeah.

A

It will be a different thing, but there's also no compatibility relationship between them right. So basically, if you say, is platform iOS when you're running on an iPad OS than the answer would be no right, so the in that sense like we don't have to do the the thing that we do when you get that we have to map things in particular version numbers to each other and claim they're compatible and they're, not I.

F

Mean that's actually I. Think wrong from from the base is OS platform, because right now, if you're on iPad OS- and we say we're not modeling iPad OS because they haven't yet split the API service, we're gonna return. True for iOS, we can't return false when we decide iPad OS is a distinct operating system because there's different API of it. You.

A

See we're already.

F

I

A

Think I think that is kind of a function of what Apple does right like if Apple does the same thing in their API, send we just we just expose the platform and the platform made a breaking change, powerful, minor break and change. That's just a way that works there right, but I would say that on the on the windows side, I'm, sorry on the on the wessex idea, we probably should not have OS X and Mecca was that's probably a bad idea.

O

Why wouldn't we with that.

A

O

Was proposed elsewhere in the issue that we only have one overload and make minor build and revision optional did that get considered at all so.

A

I yeah I did consider it. My personal opinion is that wouldn't and the reason I wouldn't is this because you know version checks are one of those things that people get wrong all the time. So I try to make the message as as brain-dead as simple to call as possible. The thing optional is you basically have a cop's notes, aren't really ordered whether you can you can set them in any order. You want as long as you use, name syntax and that's I'm, not sure that makes sense for these kind of things.

A

Right I mean you. If you want to specify those, you should have to specify a minor the affording it's a zero seems kind of odd. You.

J

Wouldn't want to default it to zero right, because then you couldn't say: is it 1.0 done three or is it like 1.1 done three.

O

Well, on the other hand, Windows is explicitly 10.0. Dots are thing, it doesn't do 10.1. Today, all the checks you on Windows are gonna, be 10.0 dot. Some magic number.

A

Sure, but what is wrong that I mean that's just the wording scheme that they have.

J

Saying it smart ferbos.

O

Better right, but I was just saying that you saying that you wouldn't want to set minor to zero or defaulted to zero. I. Don't think that's valid, because there's lots of platforms where that is the case. You want a major and potentially a bill, but the minor version can just be zero. Yeah.

A

But you owe me you're, not saving any characters, but if you say ten kumar build pull on version number, that's more characters than what is it like major. You know coma so yeah, if you say ten, comma, zero, coma subversion, but that's shorter than ten Kummer, those coulombs in version number right. Okay, you know you're, not saving any any characters conceptually by doing it, but it also will work for you know. Let's say one is fixed right and then you just pass in comma, 1, comma or whatever. It is right.

A

Whether then you know build cool on something else. Right.

O

Because I imagine if we implement this, we're just going to have major forward to be minor bill revision.

J

No I, wouldn't I would check get the major, but there equals great like it. Is it major ten? Even if it's ten point one? Yes, even you can put it. That's.

O

Why it's platform or later.

J

Right and if you forwarded.

O

To the minor one, what did you pass in for the minor? It would have to be zero, the same thing as the original as the one without it would have to assume.

B

O

Have to assume just major 10 is less than major ten minor one. Maybe.

A

I'm not seeing about like what what is the benefit of making them optional.

O

It just reduces the number of overloads we have and it allows users to skip parts if they still want to etc, but.

A

This said, the skipping part isn't you're, not saving any characters right, because that gives you skip a part you just have to put in whatever number it is right. Then you need to know what that number. That is, but there's some implementation detail right, because either 0 1, 2, 3 or whatever it is right.

A

But that's what I'm kind of saying like to me when I said number of orbitals is not a concern, because I don't consider overloads as concept count right, they're all collapsed and intellisense, but, like the point is version, checks are hard right and I'm concerned that people do crazy. Things like 10.com or revision equals some video because I'm not sure that they understand what they just did like. Are they assuming that the thing in the middle is a wild card? Are they assuming it's? Well, whatever the u.s.

A

fixed component is or are they understanding that they just pass two zeros for minor and build.

A

That's why I think the I mean saving overloads I'm generally, not too concerned, but I think in this case I, given their version checks, have this history of people them up all the time. I would honestly optimize for, let's be the boss, but like it's also, you know blindingly clearly obvious what you know what the behavior of the API is.

O

Fair enough I mean.

A

The other thing that was come up, though, is, should we have an oval that takes a version. The reason I didn't do. That was well, it's more balls to call, but also because these are api's.

A

You want to get analyzer to to look at right, and so as soon as you're positive version number, and now it's harder fertilizer to extract things, because if you, if you construct the version, other people are also more likely to put this in a local variable or put it even an aesthetic variable and all we have to you know crawl, that information out from somewhere. We just take three ins or four heads.

A

You know put the way more like if you just put them as literal bolts in the code and then we it's very easy for us to find them.

A

So any more concerns of the of the actual checking api's, because I think the actually the ones will probably be even harder.

J

Like why isn't that? Just oh s, platform is always platform blah and.

A

That's what it is, but like it so the implementation of these. So if you think limitation standpoint platforms would only have to implement the first four. The last four are basically almost ethically implementable, but the problem is how good a user do that right, because naively, if they're, just in the gate, is platform or later that's the wrong check, because now you're chewing on every platform. That is not that particular platform right. So you always have to check, is West platform, West platform double-end not is West platform or later right.

A

So you pass and West platform twice.

J

So it's it's strictly a convenience thing. It.

A

Is a convenience again same argument applies, so I didn't want to make it kind of hard to write incorrect code, and the only reason we would call to the API is effectively to not koala, remove the API or to want to call the deprecated API and call the Eternity.

J

Do we know how often that happens? You know if you're building for version 1, 2, & 3 and it got deprecated in version 2 like wouldn't you just want to do whatever is supported in 3 per all of them, like the only case would be. If whatever is done in 3, doesn't work, it 1, 2 right.

A

Yes and that's something: I try to get the inner honor section, not that easy. With the move, the of the Apple side, I talked to the iOS folks, it happens pretty much in every release, but, like that doesn't mean it's. A lot of API is right and I also don't have a good intuition on whether this is a popular ap is or not. My understanding is the way apple usually works.

A

Is they they I, don't know whether they call it obsolete I, think they call it deprecation I think that they deprecated a ton of api's and they remove some of them right.

A

Not all of them, though, and so my expectation is that the way most people will probably get away with that is by only using its OS platform or later, and then the ideas and, let's say in version 1 they're shipped in Apix, an inversion to the deprecated API x2 introduced API, why you never have to do it as platform oily, then, in this case you can just say it's platform or later to and then always fall the replacement API right. That would that would work.

A

That's a bit harder when you have to support a range when you have to support that, say: 4 or 5 version of the US, and you have to support both cases where the API wasn't obsoleted yet so the only one that exists is the old one and you also have to support operating systems. But the only API that exists is the new one because they removed the old one. But in this case it's actually nicer.

A

If you can, if you can do both API calls because also means they don't have to be in the in the same location in code right, because if it's in the same location in code, well, you.

D

Can write if check in such.

A

A way that it works right, whether if the API is a different behavior and you have to call wondering start again when doing a shutdown or something like that, then that's that's much harder to do right and given that it's just you know one one API, logically before overloads, it seemed more straightforward to have that, but I I can't tell you I have my intuition as 90% of codes are in is a ruby against its OS platform or later and its OS platform earlier then, is rare. That will be my expectation.

A

How rare I don't know I.

F

Mean effectively it's just the irreflexive operators, so it lets people use less effectively less than if that's the way they think yeah.

A

Yeah, so the version attribution, so that's one thing that I'm not sure I like the outcome, so I would be very happy of constructive feedback here. So let's look at the derived ones so that we have edit in OS platform version, which basically tells you that the API was edited particular version. We have removed in a particular version and we have obviously the dump in a particular version, so all of them have to say which OS platform, all of them have to say what version number right.

A

Now they are all mandatory which might not be nice for metadata conciseness. You may want to have constructor over those that allow you to only pass in the ones you care about, and then they only apply to the things that make sense, but you can only apply to an assembly because you came back an entire assembly s as what's specific, for example, you could put this on the registry, so registry was introduced and I, don't know, windows 95 or something right.

A

So you would just put one attribute on the whole thing and then there's other cases where you would put them on individual types or methods right.

A

The OS platform, unfortunately, for metadata reasons, has to be a string, so I did not had string constants my expectation as to why you would construct this. You would say name of West's platform, blah blah blah, so the so the strings would match these field names here and then yeah.

A

The just exposed to major minor built-in provision, as as API is but again like our analyzer would never do that because they just look at the metadata construction side, but they never that's the end share this attribute, but it felt like exposing that it's just a normal pattern that we have and the only reason so people.

O

Are going to get the name wrong? Os platform is currently ordinal, not ordinal ignore case, and the constants are all uppercase.

O

So people are going to get that string wrong all the time.

A

Let's see a saying, let me say name of Android, then that would not match, because the actual string 400 will be all lowercase. No.

O

The actual string is all uppercase if we go based on the existing convention and we're doing an ordinal, not an ordinal, ignore case comparison. Yeah.

A

Then I would argue be ourselves badly with that. That's not great.

O

Would it be, would it be worth taking the breaking change to make it in ordinal ignore case comparison at this point? I, don't think people are realistically going to want to overload on upper case windows and lower case windows, for example, I.

A

Agree with that I mean you could, in principle break people, because somebody happens to rely on that returning fault somehow, but I think that's far-fetched I would agree with that. I think we should probably make it coordinate in your case.

A

Especially because, looking at the naming conventions here, iOS with lowercase I to the US of lowercase TV, like seems recipe for disaster,.

A

A

Does anybody remember we did ordinal.

J

I'm, assuming it would probably be 4-perf but I, have no idea for real.

A

Yeah, but that you're calling this in entitle, you.

O

I'd imagine there are places where you might want to, but I'd hope that for Android iOS beauty should take care of that I mean yeah. I was.

P

O

I hope maybe we could eventually optimize it or do something if that is an actual scenario. Most of.

F

Them don't have equal lengths, so they'll all early abort anyway,.

J

It goes back five years, so it looks like we just wrote it that way.

O

Like I think, the simplest case is you're using one of the WinRT multimedia api's like clearing a performance counter, and so you want to check if that exists, and you show you're doing that in a tight loop.

O

Similar could exist on iOS or Android, especially if you get into light unity and stuff, but I'd hope that peyote covers that and then maybe the JIT can optimize it eventually yeah our rid specific support and NuGet packages system. Great, so I, don't think aisle trimming will be necessarily easy for that. Right now,.

A

A

Yeah yeah, so the other thing I wanna point out is I, have a inheritance hierarchy here for the US platform version attributes because it seemed in the docs it's nice. If we can group them and in principle you could imagine that somebody- the tools that benefit from being able to tell that it's nice platform version attribute, but that's really the primary reason. I don't have a there's, not really a need for polymorphism here, but it seemed nice to do that because we've done this in the past, a group attributes so.

F

You have allowed multiple, it's false, yeah.

A

F

A

F

Eighteen and Mac OS 900, like.

F

Unless you're supposed to like I, don't know a way that you can actually model what Apple does right.

G

F

We're looking at their their split OS hierarchies so.

A

That's so originally I had allow multiple, true and originally the design also allowed us to do things like console window worth of it would just say this API is only supported on XY and Z and then because of my thinking on portable versus high specific I removed it and said screw it like. This is only for API is that exists on one platform and then the idea. What you just said is also true.

A

The apple of folks basically push back on that and say we need to be able to say what you just said to us versus iOS versus you know Mac OS, so we probably have to make a lot multiple chew, but I would expect it to be reserved for things like that right effectively, there's in the West family, and that happens to be one SDK for the West family, but I would not expect us to use that for Android versus you know, iOS versus windows, on the same API. If that makes sense well.

F

You started with something like console window width right and if we, if we got to a point where we could make that work on Mac and Windows, but not you know the X windowing system generic like would we not if we were painting it with with this stuff of this works on Windows, and we would need to also paint it with this networks on Mac. Well,.

A

That's I'm, saying I wouldn't like I mean we originally when I, when I, when I designed this feature that, but that was my perspective, but my perspective was I'm. Writing a portable class library and I want to make sure that maybe baby eyes attack over across the whole operating systems right, but just basic what the platform companion eliezer originally did right. But then the problem is that, because these things are moving targets and window width is killed by a specific API right.

A

It's an implementation of what you consider a portable, API right and so I think that the mechanism that we have for that using these attributes will not work very well because yeah they would change over time right and so then, how do we do that in any same capacity? Right philosophy would not have who's. You know at one of these attributes through what the VCL, if you would, if you were to do that, what I would do is I would use these attributes for clearly window specific features.

A

So, for example, all the exit stuff I would mark because they're, if they're fixed in time their windows only they will never be working anywhere else. We know the version numbers so they're, effectively immutable attributions, console window width doesn't seem like it's in the same category of api's and that's why I also said. Like you know, it's not a lot multiple, because you know you should only have only ever the need for one ways but unfortunately iOS so it makes it a bit more complicated. I. Imagine.

O

People are going to end up using in their own api's for things like that anyways, regardless of whether you want them to use it for something that is specific versus, not I. Imagine they might version their API, such as in this version. It only works on Windows, but in v2 it works on Windows and UNIX, but not Mac.

O

Right but I mean like, like the Vulcan, for example, Vulcan is a cross-platform, and originally it wasn't going to work on Mac until someone went created, a shim that wraps metal and so that works on Windows and it works on Linux as is, but if on wind, but if on Mac there's certain things which only work on Mac and you have to do a couple extra steps to get it to work. But eventually it was made to work.

A

Right I think that's a I mean I'm less concerned about that, because I mean the analyzer will be well defined. So it's not so much that the feature wouldn't work in this case. It's just that the author of the API kind of forces, their consumers to a bunch of breaking changes over time right. So you know, if you want to do that, I mean I, can't stop them anyways, but they can also make arbitrary other IPI breaking changes.

A

The question is just whether we would use it at the platform layer right I've been willing to do this to our customers, who quark and console window width over time right- and my current thinking is the answer to that is no, but maybe the feature works so well that we use that for that as well, because I mean the other reason why I think attributes are bad was because they were still thinking through the botnets standard.

A

You know lens were the basically the API services frozen as soon as you ship it, and we don't really have a versioning story for the reference assemblies itself. So it's the only thing that would change their attributes. We don't really have a way to model that either, but maybe the dotnet 5 world, where we just update once a year. Maybe that's also no longer concern, or maybe that becomes more viable I just want to make it clear that you know. From my point of view, this is not maybe the mechanism to do that.

A

Or any other things or names, absolutely there's a bit more, it's kind of sad because obsolete, basically conflicts of the other opposite. The tribute that we have, but unfortunately, obviously attribute, cannot be version right, so we don't know which version this was in. So this would add that the.

D

Only thing I added here is the URL.

A

So similar to the opposite, the attribute there's a URL that you can point people to that say. You know read this document. Maybe we should add this to remove tin as well, so that there are some information forward. Keep what you're supposed to do instead could.

O

That, maybe just be a general description property. That's on all of them. That way, you can even link to like the relevant Docs for a particular API set for added yeah.

A

I mean so description, so the reason I don't like description is because you don't know it's a URL, so you can't in the UI actually whether there's a link versus if it's a URL. You know it's the thing, so you can actually make the diagnostic it's seeing in the air all this, for example an actual hyperlinks and people clicker they actually would go somewhere.

A

I mean you could, in principle, detect that it's a URL right, but parts of the description or something but people very often with description fields will just have you know, basically a body of text and then within the text somewhere. There's a hyperlink right. So that's and it does it. You can't localize it right, so I mean obsolete. Has a message. I think my proposal would be. If you use obsolete moving forward, we don't ever use the description.

A

We just set the diagnostic ID and set the URL and that's it and everything else is then localized where you go so I. Think I, don't know. Description doesn't seem like it's the best way to do it, but I think URL probably would make sense at least on removed. I'm, not sure it makes sense on edit in I mean maybe, but.

A

Yeah I mean maybe we can just add obsolete director to the base and then say: there's always an optional URL that may or may not be present.

O

Yeah I think I think on added. It would be useful for things like here's, a summary of the API and how to use it would be a realistic scenario. Could you get used for and then with deprecated and removed? It would be here's the docs on how on why it was removed in the alternatives to use yeah.

A

I'm, just thinking of like now keep in mind that, in in an actual s, SDK, this attribute is literally on every API. So if you don't have a string on every API that is significant metadata upload right, because also that that link, you know, will be different for every API. If you actually link to the to the actual thing where, at the actual section, in the release notes, for example, that you can't have string and turning for that, for example, think of the Windows SDK more like 100,000, api's I'm, not sure we would do that.

A

O

It just be nullable that way, people who don't want to provide it can make it all.

A

Optional, but it would be a property, get set ID. So you you you. If you don't pass it, you don't pass it right.

A

All I'm saying is that I think what you just said for introduced in a given that introduced in is literally the most used thing. I, just don't think people will use it much.

A

But maybe it's reasonable to say what they're really pointing to as the release. So maybe you can have one string per version number which actually interned right.

A

All right so I think that I have enough, based on this review, to actually update the spec I will mark the api's still need to work, because the thing needs more more vetting on the actual spec, but I wanted to make sure that you know we at least have some sort of understanding of where we are with this.

F

Because I was curious, I was just digging back through the OS platforms, history and it I I find nothing that suggested it needed to be ordinal. It was simply first written that way and that's what we shipped right.

A

A

Let's just talk about this one here, the compiler guys aren't here, but maybe we can just make some forward progress on that. So is everybody aware of what in it only properties are or should I give you an introduction to that.

O

A

Sorry, 10 I would. What did you say, Julian.

O

A

Doing is here then Julian. Please talk about it.

N

A

It would be used because I don't think everybody has that context.

N

And made some clarification and so part of what's confusing? Is the scope changed so presently? The only thing that's in scope is in it's only accessors in it only setters I should say things are not in scope or init only methods or in it only feels so there's a syntax for init only setter. We debated you know easily in itself or just in it and then in terms of what we want to omit after discussion.

N

So the banking fields can just be read-only and then the sets accessor would be tagged with in anatomy attributes and we'll have a mod rack with the same attributes on the parameter.

N

So the the thing that's I think not analyzed is whether the so that tribute is very simple. There is no, it doesn't store anything, it doesn't have any parameters.

N

One of the questions is: what are the attributes targets so strictly for this scenario, the only attribute target that we need is a method and there's a question of. Do we allow fields proactively or will we add it when, if we decide that we need it later on.

A

I muted the way understanding is that you need only properties, so they are a different accessor in c-sharp land, but in metadata land there would just be regular, setters rights. In that sense they are the superset of what set is, but it's basically just it like a type system kind of enforced limitation of when you are allowed to call it, but it's basically still at the metadata level. A regular is.

N

Just a setter, so you can only have a get and as an init we cannot have a set. An init. For example. An init is a set right.

A

And then actually import metadata. If asked for the setter, it will actually give me back the upset access all right. It's not a separate access or either right. That's.

N

Correct and then the only change is, will allow you to set this in a couple. New locations, I think there's an example of at the very top first line, basically, if you're in an object's initializer, so between the curly braces there. We consider that still part of the initialization phase, although it's not part of the construction phase. So this is not a place where you're normally allowed to set read-only things, but we would allow you to call the in it sets.

A

That there is a display like today, basically the the initializer syntax is just syntactic sugar. To avoid saying you know, C dot right and now you would actually have a semantic difference between saying you C and then C dot in it only property that wouldn't work. What if I say you see in an open grades and that would work right. That's.

N

N

Calling out is this would in the current proposal. This would only work for in it's only properties, so there you know, if you have, there would not be an equivalent for fields where you're allowed to set it in during the object creation.

N

A

Right so my other question is so you are mod, wrecking the parameter pass to the setter.

A

I assume that means that both c-sharp and VB, unless they do work and support in it only they would basically correctly report an error that says sorry. I can't call the setter, because there's a what wreck I don't understand.

N

F

Correct, hopefully, for end-to-end experience for working with the intellisense team that if they have a compiler that wouldn't understand the mod rec, it just doesn't show this.

O

F

O

Think C++ CLI respect, mob racks and I. Don't think reflection, respects moderates.

C

Reflection does not respect mod rocks on our serialization libraries. Also don't respect mod racks and I know that we've had some discussions on how this behind effects system text Jason I'm wondering if we should actually build into libraries like that knowledge of mod rec in general, and then say if I see something that I don't understand, and it has a mod rec fail.

A

Interesting so.

N

There was a lot of things that were mentioned, and so I lost track. A little bit for reflection does not reflect I'm, not sure that reflection. We respect, we don't need either so I mean the purpose of this mod rec is to try and prevent people from just using this as a regular cetera. So, based on that, we can write.

O

Well, the property map table supports more than just get set. Is there any reason we can't just name this init name to make it clear that it's not a standard setter.

N

It is a so the thing is we want to make sure you cannot have a get a set and an in it.

N

So in that sense, where we are saying this is a a setter, not a brand new thing and adding it as a brand new thing, we think would be a lot more impactful ecosystem.

C

And also just to comment on something that was said earlier. Reflection does sometimes forbid setting read-only fields.

N

F

I, don't know how it only auto properties are actually emitted in the compile. If they're directly setting the field, then there's no setter for the reflection to bind to it would have to go, find the generated field.

A

Yeah I think there's a general question right I mean my understanding is: isn't that also tied to the validator thing? Julian, isn't the expectation that you have to call validate after you said all the inner, the only things right.

N

So there's a separate feature under the same umbrella, so this is all under the umbrella records that were kind of breaking down into a bunch of separate features. There is a proposal for having a validation at the end of initialization, so yeah there will be.

N

Well, the main scenario is when you do with so with is a kind of new construction. It would be a Crone and you'd be able to set some of the properties and, at the very end, when all of those are sets, you'd get a chance to validate and fix your fields be close or I know, as the initialization I guess,.

A

The reason I'm asking is because it sounds like if that's the case, we probably have to rethink our reflection, interacts with that right, because you know a naive, Lee I would think. Like we just block setting the the you know, it ended only property but reflection when you say property info dot set. But then how would you ever finish it like them? Because there's no such thing as a.

A

What is this thing called yeah? Basically, they couldn't often initialize or everybody. You can either introduce that to reflection and say, construct an object and passed a bunch of things that you want to be in it and then that one will actually do the right thing and will actually set the property and also call the validate method if it's there, but that doesn't seem like something we can easily do with right. Now.

N

So first they're not as solid in terms of being in c-sharp nine and they will have holes so languages I think we're not planning to try and lock down validators super hard. So there will be ways that you can initialize your thing and like if your compiler does not recognize things or through reflection, you could get in there bad sites.

A

You right- and that seems like something we should have to think about like we should probably have a conversation about independently of this. So my other question is: why do you have both the ma track, as well as the only attribute applied.

N

That's in the contexts I wish I wish enter we're here, I might like them under anything. If you want to draw me since someone is writing the back for this.

N

All right, so the context is that.

N

So initially we we also allowed in it only on fields and those who are getting the attribute on the field and then in the context where we also allow in it Tommy methods. We would put the attribute on the method. So putting the attribute on the setter here seems like a more consistent thing to do.

N

We could technically, we I think we could use just the Modric as the signal to your recognizer and then we would have an attribute that has no targets or we will have a type as maybe just not an attributes. Yeah.

A

I mean that's how all the other mod regs work today right, so you would just have a system run time. Compiler services, just the market types that it's just in it only right.

N

But since we expect that we're going to need to set an attribute are so I the other scenarios I mentioned I'm, not sure that we will be able to do just with the mod right. So.

N

That's why we went with a with an attribute, so that harder to say is based on extensions of the current scenario and.

C

See, oh, for example, with the validator method, be probably an attribute, rather than a mod rack.

N

We don't have a speck for a rather value, they.

C

See, but is it is it may be a reasonable guess that that might be one possible implementation.

N

So it might be an operator, it might be a constructor with an attribute there's a few ways. This would be represented. Okay,.

C

So it's it's something that we would still need a special case from consumers of reflection. Then, once we get that stuff, okay, no.

A

I mean you would always be able to call a validator all right. It would just me it's just that when you set the properties, you required to call the validator afterwards, which well.

N

There is one we're not sure about the the syntax. One of the proposed syntax was to use the same syntax as a constructor, but with no parameter list, and if that's the case, then at least in c-sharp there would be no way for you to address that and in brokers I see. Oh okay,.

Q

So I got asked to talk about metadata. Why are we talking about validators.

N

Do we need the anatomy attributes on the set accessor? Could we just do with the mode rake the.

R

Parameter on the on the parameter on the parameter to what this accessor, why would that? Why it's so.

Q

You basically have to put it on tight that it's going to either be on the parameter and 20b. You can't put a mod rec on an access, so you have to put it on a tight, so it can only go on a.

R

Tribute itself so.

Q

The attribute, though, is just a question of: do we said: does the compiler need it 100%? No, don't care like we can. We can absolutely like we ever. We have no need for that after you whatsoever, the question becomes, how do serializers look at that? I'll do Syria. Do we like if we want? We because part of serialization is like unwinding these and recreating them, and if you look, there are some serializers. Today there was a. There was a talk. There was an email that I got see, see that friend responded to were essentially serializers.

Q

Her, like hey we're not going to touch read only feels bad bad juju. We don't wanna. Do that completely reasonable stance. So the question is in the future we want like heart of doing in it. Only design is to make sure it can work with serializers like and because this shows up as it gets that access for her what's fantastic.

Q

However, if you get into more complex serialization, there's a question of: do we ever envision a case where serializer might say like construction is done, I don't want to actually set these values on these properties and further, if you think about ahead of time serialization, if you think about people who are writing sterilizers today, where they're like they're gonna, use source generators to build them ahead of time, do they have enough information to generate c-sharp code that will compile, because in order to use kind of this feature in c-sharp, you pretty much have to use an object initializer and you can't do like a simple property set now from the c-sharp simple model, we will absolutely it'll be in our simple model.

Q

It'll be very easy to tell, and that works great for I guess it's more of a question to know how will people know it's an it? Only firm reflection can we see moderate and is it valuable for the pasilla I think that's where the attribute comes important.

R

Do we want to you're saying the attribute is a convenience to tell that this.

N

Is not just a regular set yeah.

Q

Because from the compilers perspective we just look at the Maverick is enough: we can. We can live with that, that's not we could. We could easily just pull that out of the data and I'm buying easily I mean I'm, saying that Julian has to do the work so that easy for me, I tells me, Julian's name- is currently right over his mouth. So I can't see if he's cursing or smiling okay, he moved his head back. So how can you smiling, but it doesn't really matter so.

A

That makes sense, I think then the question becomes I think I met a question earlier about reflection like it seems to me. Reflection has to be aware of it, I think for say, Lazarus and that's I, think what we talked about validators is that you know there's nothing fundamentally wrong with setting in it only property. If you don't have a validator, it only really kicks in then you know you can only set it when you actually validate all right, so you expect it to set all the things and then call the validator right.

A

So in that sense is effectively a theorization thing needs to be aware of it, and if you don't do any work, probably today, given that reflection, I think you can watch someone wax, then a deep reflection code that works today about two networks.

R

C

You can get mod wrecks and bot ops. The reason the reason I shared that by the way is that there was an earlier question of whether our serializers in general should fail these mod wrecks that it doesn't understand. I.

Q

Would say, but this is one it should understand. Emo said something and I want to make it got it. Incorrect, I'm gonna make sure he didn't mean it the way he said it validators and it only properties have nothing to do with each other. They will in concert work together later, but and it only properties exist completely independently of validators sure.

A

I guess the question is so and.

Q

We absolutely want absolutely once you realize they're setting in and only properties. Yes,.

A

So then the question becomes: okay, that's let's talk about how the section works today, because my expectation is that probably it will just work the way it is meaning, there's a motor act at the surface, but it's probably going to be ignored. So if you just call the center of the value and the sooner they'll, probably just work, it will probably just set the thing right, which probably and validator that you expected to call no harm is being done that that's our.

A

What of the thing, maybe there's other features that you also have to honor it. But the thing is: if we do nothing, you made me feel, as I said, doing the right thing already and they would not do the right thing if there is a validator that they expected to call right. So then that is the case. Then I think we should have a reflection feature for that, where maybe the regular property in for that set value will blow when there is a boat wreck.

A

That says it's the only thing and then maybe you have to call some other API that you have to batch them all up. So that the projection can call the validator afterwards right, there's something to the extent yeah. Let.

C

Me push back on one little aspect in that scenario: if you haven't you serializer, that's hydrating an object out of like out of thin air. Then it would make sense for that D serializer to call an it properties. If you house it, you serializer, that's hatching an existing object in place. You might argue that it's improper for that. These serializers are calling it properties, because the object has already.

F

Been constructed, one yeah I mean the weird part to me about it. Using the set slot is public reflection. If we don't change anything in reflection, it will find it and it will look completely Kabul and yes, it's a little weird that you're using reflection, but it it's weird that now a line of reflection without needing to do private reflection is doing something that the c-sharp language says is invalid and will will probably bypass the logic of somebody who's. Using this feature, like you, you thought that this couldn't change after the object was created. It.

Q

Was read-only? That's but honestly, that's what reflection does today? Reflection have zero respect to read-only public private whatever, but.

F

Generally, if you want to call private things, you.

Q

F

The but they're they're weird, but the reflection, they're gonna, show up as exactly public setters do today. It's not aware of any of this metadata, and the question is as part of this: do we need to make it aware.

Q

And I would say that that's we can make that decision, but the context I've always thought about this in is very isn't. Essentially, the following is publicly public reflection. Public reflection can change, read a public read-only fields. Game over, like, like reflection, just doesn't care to bits about read-only and instance, members, and so this is no worse than what we've been doing since one oh, and if we wanted to say, if you wanted to look at this and say okay, you shouldn't use one O's an excuse for continuing to be bad people.

Q

Then we could talk. Then we could discuss that, but we really need to make sure we understood like how does impact serialization and make you dressier Eliezer. So Bobby can like function with this.

G

Yeah I mean I, agree, I mean just just because something was in in p1 doesn't mean we need to keep doing that right.

A

What is the expectation and I think with only fields relatively where right, like in the sense that people who actually write you know code? That is see us over the wire having public? We, the only fields, it's relatively rare instance, few students right so there's, probably not a lot of code that is broken by that right versus this one will be one of those features that we put on our blog. We talk a lot about it at conferences.

A

Everybody will have their own syntactic cuteness to do that and now the question is how much realization code ends up being broken, because the reflection doesn't actually call it validator at all, because it's it's not aware of it and suddenly you're, bypassing all the cute logic that somebody has written right and I. Think that's the.

D

A

Than being broken, everyone.

A

To Jarrod's point, my concern is more about other features that you build on top of that right, because only the problem is the semantics of in it only which is you can only call it after the construction and before any validation.

A

Logic, right, I think that's kind of hard thing about the feature right: whether the type actually has a validator, it's kind of irrelevant in that sense, but like practically speaking, it matters right because if you don't have a validator, then having an in it only property just for the people to put it in the in in the curly braces. But if a detail eyes are just calls it nothing. Bad really happens right.

A

It only really is broken when you actually rely on the validator in your time to reject sorry, you know input which is also executed by the DC Eliezer right and that's where people probably get burned and I think that's why we should probably think about how it surfaces in reflection and make sure that the default behavior that I mean I guess the question is people that write the fraction code today with Eliezer Soros or somebody else?

A

How likely are they're broken by whatever meta data, encoding, you're choosing and and is that, is that the right outcome I mean you said you would break less of the ecosystem when you, when you reuse the set concept sure, but is that actually what you want? Because if you expect people to call the set method, have to do special things then maybe using the sets. Flawed is not the right thing to do. No.

Q

But here's the thing we want the ecosystem, the work as it does today, like the reason for choosing set, is you're saying: should we do it and I'm saying we just? We explicitly want this to happen. We want it to be easy for serializers to use this, because the majority case of people who are using reflection to do these properties are serializers and we want records.

Q

You want all these types to just work with serializers without having to find every serializer ever written, give them increase their concept count and get them to do things a different way. Generators are gonna, be broken. Dealing like people I understand people get kind of like caught up in this, but it's like the bar we are starting from is serializers. They can break everything today and it's buying forever.

Q

If you took what we are doing today like in terms of our fields like people still use reflection, even if we did the most extreme thing where we didn't use that accessors, we generate an unspeakable memory. We did all this. Reflecting could still beat what we were doing like. We can't stop reflection from.

A

It's not about being bulletproof, I. Think I, think. The question that I would have to use is that one world is better and is better wear by default. In it, only properties are not called, are not supported by the analyzers about the cylinders that have to do work to support them properly, or is it better when they accidentally use them and then just they don't use it correctly.

D

A

Q

That's the question that you.

A

Have to answer for yourself and.

Q

I think the last the latter world is fine, because it's I would say it's more about educating the the cases were serializers like today. Do it incorrectly? Are me exceedingly rare, so you realize that are just like store data create data like they're, not trying to do. They are not giving the data back and then manipulating it afterwards, like they see the construction is one basic operation, a unit of atomicity and then handing it back, and that's: okay, like it's not particularly valid, like a problem.

Q

Now when we get to validator like that's, when we're gonna want to go and talk to essentially serializers and say hey, this is something you might want to respect when the fan goes about, validators is validators between their properties. Validators exists, validators are a problem that we have with object initializers today that people want to solve forever, and so the decision for, like I, don't think it's valuable to say like. Oh, when you do validation regret. This is like no. We won't because validators have all the problems without a net properties.

Q

Validators are going to or essentially that's the time we have to go to serialize, Lucas and say: hey you have to help like we can. This is not something we can do for you like. You need to participate in the crazy here, but I.

A

Think the question is like, if you so, basically it will be called after the construction and after all in it all, the properties are being set right.

A

The friends right so, if, if there's no such thing as a validator right, then, is from a distillation step, but no observable difference in a public setter and the public in it right.

D

A

It only comes to become an observable difference when there is a validator right and that's why I'm saying the question will is like it's not you.

Q

Have to divorce validators from Anatole, you really have to. They have nothing to do with each other, nothing. They are completely orthogonal concepts because valid like here's. The thing if a like validators were supposed to run after everything after object. Construction is finished on get set properties. So if, if the serializer DC realises runs a validator and then calls a setter and then only are not event, only that's a problem, and this is why they like they're orthogonal concepts so.

A

Let me get this red thin.

A

So are you saying that if you have a validator and you don't have any in it only properties, if I construct the object using the connection which other syntax, I would just set all the normal properties and whatever thing happens, and anybody later is called because, okay, okay, that makes sense to me so that that no means that yes, there would be no observable difference in that context as well, where you know, if you have no init only properties and see that it's not aware of validators at all and will not call it, then that's also a problem and there's nothing you can do, except if you would now mark the constructor of the types with validators that the constructor itself is basically not callable of the whatever heck we have right, which was also not you honor by reflection right.

A

So this goes back to like reflection will probably like. Unless they do work, they can always buy pasta, validator and there's nothing. We can do about that. Probably okay I mean unless you change the flexion and say if the constructor is marked in such a way, then we blow up unless you call some other thing, but I don't know what that would be.

A

Okay, so that makes sense to me so in that sense, okay, so I, so my personal opinion is I would probably prefer it's not being an attribute, because that's just more consistent of what the other mod recs are they're, just plain types and they're just one Rex. And then, if you really want to service this information from reflection, I'm, not sure the attribute is helping much because at the end of the day we probably have to do. You know explore some API on reflection in order to make that lucky.

Q

I feel I was I, was more iffy when I didn't think you'd be a mod Rex reflection, but then leave I showed the sample on how to get it, and if I go okay, you know there is a. There is a escape hatch. It's not great one long-term. If we wanted to expose it, though we could like take a different designers. You just.

Q

Q

Or property info whatever by.

C

The way I'm looking at the I'm looking at the in modifier on parameters and it looks like that's a mod- opt instead of a mod wreck and I. Seen that correctly do it on a virtual methanol, flip.

Q

Okay, got it, it's not it's on a non-virtual, it's not a mod on a virtual it to Ma crack Oh God it okay, and that's because you can do some really battle.

Q

Is that really comes down to the fact that we, if you look at how you can pass array elements, if you pass an array element by in and it's an unsealed tight, we don't do the type check, we just say: load element, rest and the reason, and so, if you're, essentially, if you violate that, you can basically cause type safety issues in the CLR, because it'll allow you to assign.

Q

Essentially, when you get to covariant arrays, you can end up assigning the wrong elements into them and violating like array, covariant rules, and so we needed a mod rack to basically say like no. You should never ever use this member. Unless you understand what in is so, you can not assign to it sure.

C

And then I think I misspoke earlier when I, when it's on a non-virtual method, it looks like it's not a mod Hoff. It's actually an attribute on the method itself, with some kind of parameter ordering data. Okay got it yeah.

Q

And I don't think we use I think there Tanner could probably spike chapter inverse of how we didn't use the same thing as in and out, and we did something different because of some has historical, crazy reason.

O

Where is this sorry? I was I was reading, something else. Hanger was not paying attention. He came out.

Q

Don't worry tanner, it's not that important. It was the.

F

Q

Around then, more.

F

About the language feature than the attribute Jared that, while you're trying to make serializers work by using the set slot, presumably code generators like most designers, are actually gonna be broken by this. Because they're gonna see a set, they're gonna write the set call, and then the code doesn't compile generators.

Q

Will have to understand, there's no way around that and that's why, in the Roslyn API is very, you have a whole little things rods, an API to make it visible, so code generators can pick it up and use it. But if you're generating straight line c-sharp code, then yeah you're gonna have to understand the new code patterns. If you're gonna want to use it.

A

One more question in it only is the only thing that applies to a property correct at.

Q

The moment yes.

A

Well, you say at the moment so I'm actually.

N

Suggesting either this.

A

N

If we decide to allow it on fields or methods in the future, that's where having it as an attribute will have been useful.

Q

Why I understand her feels? Because we can't really mod wreck the field, but for general members, I wouldn't say the logic would apply.

N

For methods, where would you put the my break, you.

Q

Can always put it on the return type of the method.

N

Q

You could yeah every method, everything all methods have return type, so you can always put them all back on the return type. Okay. Well, this is my seat honestly. If we want it, if you want to think that far forward Julian, we could always just say that put it on the void part of the center and don't put it on the parameter.

Q

I'm, pretty sure you can attach mod Rex even to PT Boyden. So.

F

That setters return to the input, type senators.

Q

Don't return anything well.

C

It thought attached to the type it's attached to the parameter and the return value is a parameter correct. Yes,.

K

C

You can get the parameter in for the return value, because.

F

It is a property setter, a property set expression returns the input value cast to the property type yeah. That's that's! That's! That's! C-Sharp!.

Q

Not IL and I have this adder is void. Oh good I all makes more sense. Some confuse our lazy.

N

Ass rules, so that is the question to them. Podrick go on void then, and should it be an attribute or just a type so.

A

What so, what would Jerry you just said? He can't apply the two fields, but fields wouldn't the same logic right. If you, if your market feel as volatile the volatile is applied to the return value as well, so you could always apply to the return captain.

C

Yeah I'm showing that fields can have ma dirrection ma tops as well right.

A

In which kids, that would just say that's its clothes, only just make it not an attribute, and then it's just more direct in their usual spot. Whatever that means.

R

Okay, that works I have some thoughts about where we put it. Julian I'll, send you a quick little but you're busy, you're, okay,.

N

With saying it's not an attribute and it'll be, we can reuse that for other scenarios in the future. Well,.

Q

F

Because I know that with nullable you were there's the exploratory thing on saying that this method runs before the knowable states. Validated is was the purpose of an it only for method to say that it it's still an initialization phase concept.

Q

Or yeah there's talk like we basically said Melanie in, like we can put that our pocket and if people say they want general member give it because it seems like reserving the attribute for that is or but we would do the same thing that we would. We would just mark them as in it and that birds will you treat him exactly like properties.

F

I mean if it has parameters, it would now be a complicated indexed property. Now.

Q

That's why I was gonna, see that's what I was gonna tell Julian offline. If we think about the more general feature, we should just always throw it on the return type. Okay,.

A

Okay, then, the only other thing I says: let's move market, that's its name, it is in it only because it's also consists of all the other modifiers we always talk. Move is, and then it's just.

D

Public see what class no.

A

No, nothing- and that seems the easiest- a none for everything.

D

C

It's a behavior that, if you're targeting a framework that doesn't have that built-in type, you would just admit it into the DLLs it's being compiled. No, no! You have to provide it really. Okay, because would lead to hurt the gullible in that regard, because.

Q

With null and the reason, why is because null- and this is so basically the divide, as always, isn't a mod rec or not, and the reason? Why is because, if it's with null ability, the the attributes don't have to be public, we could just make them internal, generate them in the compiler and then the compiler can just make like by name copies and every little we can basically just say like get or in the compiler.

Q

The problem with the problem with mod wrecks, though, is they have identity, and so it only works if, because they have to be as visible as all of the constructs are using, because we have to copy the signature eidetic into all the assemblies that essentially do overrides.

Q

The signature has to be eidetic, has the exact height match and the compiler team is. We have not really crossed the boundary yet where we will silently emit public types into your assembly that they spoke, and so we we have not crossed that boundary. Yet we did design the feature such that we are saying that we will look for this type by name and if this type, if there's two versions of this type, we will tiebreak on what is in core them.

Q

And that means that people who want to go back to like next an or two they can just define the type and you know it'll work. And then, when you go to when you cross, inter up with nets on or three the compiler can work out. The magic.

A

But the type should be probably internal right.

Q

And it can't know it has to be as accessible as the methods yeah.

C

Q

C

Someone's querying for the Maude Rex and bought offs, they need to be able to check the type against unknown thing that makes sense, but.

A

Dan Thomas. No, you have aliasing issues right, because if you have four libraries that all defined in it, the only ones, no, they all have a public type that is available right well,.

Q

That's what the that's where I just said, the tiebreaker rule is. We will take Karla that that exists and if not normal, c-sharp tiebreaking rules come into play and you have the exact same problem that you use to have the extension after you.

A

Yeah I guess the only difference here is that customers never have to refer to the types that don't see the editing issues. That's fair.

Q

Yeah, but we used to have this with the extension when he did this right, which the extension attribute. Do we pick one.

C

Of the things that was brought up previously is: do we need a friendlier API on reflection, like property info dot? Is an it'll may do? Should we open up another issue for that just attract that, or is that not an interesting thing?

C

A

Should I mean this is general thing that I started last too late, so I we should also explore solubility and other things. I mean there's some pushback on the unfunded from the runtime team to not make it too short to sous-chef specific, which I, quite frankly, don't buy, because realistically the vast majority of our customers, as you shot people and other languages, eventually tend to support.

A

C sharp, like features if they use for so I, think we're better off exposing these things, because if everybody does the crazy, then you have more bugs in the system, because not everybody gets the crazy right right.

A

So we I think we have to think about how we do this in reflection in general, but this picture seems to be something I can see becoming popular and then, if there's an expectation with you know, other related features like you know: validators, not a stuff that we expect reflection to somehow make it easy for people to do that then, before we want to think about what the future will look like the perfection.

A

But the nice thing is we can just people can create today because it's exposed.

O

Just a quick meta point on it being a mod rack: I was just getting through the spec and apparently mod recs aren't allowed on public members that are considered CLS compliant as to compiler, taken that into consideration for existing mod recs it submitted and don't they're about CLS compliant I thought they compiler still emitted warnings. If you had the attribute at the at a higher level and then emitted a member that was not yeah.

Q

And if you do that with new features, you get Lee again, okay, we basically honestly like we. We should make it. Someone pointed out two years ago at some point that there were certain things we were like I, don't think that we give like proper seal, ask employment.

N

Availability, for example, relies on attributes are not CLS component, a if I remember correctly, yeah.

Q

Cls compliant kind of went out like at some point. We realized we hadn't been tracking it for awhile and it really just it's not been a part of the features we've done since then, and I think we should. We should I know we've had you know tentative discussions on. Should we have a bigger discussion on. Do we even track it inside of like done at runtime, because it is affecting your build perf so see less compliance is a non-trivial jack and the compilers good yeah, but it's good point to bring up.

A

Have you looked at other options to encode this in metadata, like, for example, move the some of the flags that we have.

Q

I think when I look through it, I didn't see anything that looked particularly appealing.

Q

Generally, at this point, I'm were somewhat wary of using existing Flags because it wouldn't cause many interpretations like if you look at when we did in. If you would, if you had said that c-sharp 100 had n, we would have definitely encoded it differently, because there was a much more natural attribute to use right. However, given that that attribute have been the eco system for 24, you know 18 years and we could easily find places where people have done it before.

Q

Like me, babe Reid might have said, like yeah, that's the most obvious and natural way to encode it and we're not doing it. That's where I was saying. Tanner could probably cite chapter and verse of where the more easy way of encoding in should have been, but due to conflicts like P and VOC and stuff, it was there's pretty obvious. We couldn't do it. You know.

O

That was the wit without it's just a rough with the out attribute in Nome adopt, but within we had to put the amount off on it to differentiate the back and patent story. But we do actually omit the regular in attribute, as well as the mod up to make sure that P in VOC works correctly without breaking back and back yeah.

A

It wasn't necessarily asking.

G

For that issue,.

A

The purpose existing mechanisms more like Eddie, a new flag, quite property attributes right now, I think- is like four flags, but it's a you short right, so we have still Flags left on the property that we could use for that.

A

If people understand what I said or.

F

Presumably, that requires a runtime change and gods already dropped off the call.

A

Yeah, that's the thing, I don't know dick I, don't know like whether we just blacklist things or whether we just don't care well.

O

I know with the calling conventions we we only have a limited number of bits as well, and it was determined for future calling conventions. We're going to use a we're going to use one of the bits to mean you have to now go. Look at this additional attribute, extension yeah, so I, don't think using the attribute is necessarily bad. It all be all the bit would really do is say there might be.

O

It would be a slightly cheaper check, but not necessarily free and there's probably more useful things that those two bits could be used for well,.

A

Keep in mind, we have only one two, three four five property attributes right now but see you short rights. We have a ton more slots that we could use and platform. Is you know, twenty years old and we haven't really done anything like that. So what.

N

About fields and methods, I think.

A

It's probably similar I, don't think we have that many slots taken, but I could be wrong.

A

Sixth, with local field attributes.

A

Yes is modeled American, one, two, three, four, five, six, seven, eight nine 10 11, 12, 13, 14, 15, 16, 17, 18 and a bunch of them are probably flags or masks, so yeah fields might be at the limit. Already. A method attributes is fully yeah same here. It's probably we might be already at the limit of fields and methods. Yes,.

A

Now never mind them.

A

Alright, so then, and we have an hour overtime and we have talked with you- the backlog, you always feel the time somehow.

E

Thanks much all right, thank you guys.

O

Mo before you drop off yes, I sent an email. We need to schedule more arm, intrinsic API reviews. We've got I, think seven issues open and I'm going to be opening I think two or three more today. So we're gonna have a I, think 10 total we have to review and then we're done. Okay, so.

A

The next load is May May system make.

Q

Sure you twist the emos arm to talk about arm issues.

O

G

O

I think that's probably fine, but we might need to schedule a couple or or maybe even additional slot to make sure we get all ten done since it's a non-trivial surface area and we've used up the entire time. The past few times, yeah.

A

I mean it's the thing: I, don't think we have additional slots available to us like I already booked for extra slots, so nobody owes twice a week and I want to use the next week to do the back, for because I haven't done this in a long time. So massively urgent I would prefer my fifth okay.

O

As long as we, the feature is still supposed to ship for not net five. So as long as we still meet that I think we should be good. Yeah I mean Nick Mays. Nowhere.

A

Near the shipping date, so that's fine, so tell me roughly I think it would set an email right. Okay, I will take it up and then I will say to the meeting. Okay, all right guys enjoy.