.Net Foundation Design Reviews, 18 Dec 2018

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: .NET Design Review: GitHub Quick Reviews

Description

00:00:00 - Approved: Native Library Resolve Event https://github.com/dotnet/corefx/issues/32850
00:07:56 - Approved: Native Library Loader API https://github.com/dotnet/corefx/issues/32015#issuecomment-448324606
00:30:25 - Approved: MatchFailureException https://github.com/dotnet/corefx/issues/33284#issuecomment-448332714
00:54:47 - Approved: Add API for resolving component dependencies https://github.com/dotnet/corefx/issues/33165#issuecomment-448337807
01:11:16 - Approved: Setting/Getting HighDpiMode in WinForms Core Apps https://github.com/dotnet/winforms/issues/135#issuecomment-448340808
01:20:58 - Rejected: DataObject.SetAudio should be exposed with a new overload taking Span(byte) https://github.com/dotnet/winforms/issues/218#issuecomment-448343258

A

So all right, so today we will do a bunch of stuff, so I've opened a few tabs of things. I want to review first, so I, don't know what order is the best, so just open them in whatever order I saw fit so um people on the cooperate for the Toledo context, one or the definitive loved one which, which order? Would you prefer I.

B

Don't matter this one is the most contentious one, so you should probably.

A

Know better well, they I'm they all. Like the same like this one. Here, that's missing just one solving I manage the whole thing which I think.

A

C

That's just a note from de l'argent.

A

So you're saying something now.

D

I said this one: we have a look, it's the other one. We looked at a bunch of them in a product. I was just one more API that we said. We look at me in the next item, so this one we have to look at. You know I, see like.

A

We haven't I mean didn't we have another one already.

C

Isn't there like a topics the night, the foyer vulgar API yeah.

D

It's three two: zero one: five yeah yeah, let's say that's the overall one right and I think we want echo CBA does one morning K remaining I.

A

See I see it so this one is just the spin-off of this one right right, right.

A

B

You want I can provide context on this one, because this is kind of isolated think we can probably decide on it right away. So.

A

I think the last thing we talk about is the only.

A

B

Pretty sure we didn't talk about this one, yet this is different than the one I just was mentioning.

D

A

I think, like the only thing I remember, is like we I think we talked about this at some point at least I. Remember having a conversation. It was a different API, but it was the same pattern where somebody effectively proposed a call back with a thunk. So usually we don't do that.

A

Usually we created type for that, because it's easier to actually name parameters, for example, but otherwise it's just parameter one two, three right: we have no naming for those things, so in this case it might be, it might actually better to just have a delegate that actually has the name parameter spelled out.

A

So, for instance, is the first argument of path, or is it just the name right like you just don't know, but you just have to basically read the documentation ready to study all import resolver, which is not really, but you want to read this because it's not friendly and intellisense right. That would be my own.

E

A

That's the other thing is like it seems a bit odd to have a register, but in one register make a move.

F

D

You in this one you could pass in none when it is unregistered. All I can under this will be paid, doesn't matter in this discussion. There is like expression about whether it has to be even so callbacks and ultimately, we served on callback, because we don't want multiple events or this one, because there's no really use case for it and could make things complicated. There already were.

B

Apparently, like confusing this to the one which is shown on the screen with the other one, because the one which is shown on the screen there's a.

G

Good reason for.

B

This to be event because some load context already has events. It has a very similar event like this one for managed dependencies, and so this is basically just kind of completing the API by adding an unmanaged one as well.

B

We chose the funk because it's copying what we've done for two managed one, but I mean this is easy to change. We can do the look at here. No problem.

D

Use a different convention, but I pick this particular rotation because even just resolve resolving or manage event. So it's just exactly the same syntax. If we go down to the rational, I've actually have links to it. A that's already exists on.

A

Yes, so I think you alright I, think I'm just looking at this, so we already approved for that course 200 the API. That is resolving, that takes a funk. So it seems that in this case we should just also stick to the phone, because that's a bet I'm already having this time makes more sense. I mean.

E

It is weird a an event handler be having an event handler or even though, that callback that is returning something because we don't compose well.

A

Yeah, that's the other problem right like what do you do with that, but yeah? That's.

F

What we already did on that I guess.

A

It will be the first one that returns non wrong. I guess! Yes,.

B

That's exactly how its implemented for the one, which is already there, there's an event which is called resolving, which does the same thing for managed dependencies, and it works exactly that way that it does return on assembly and the first one. It returns not now stops the processing.

H

A

To me, I mean active yeah, be the only downside of people with all these patterns is you can never really extend it, but the nice thing with events where you have any bad arcs. You can just editor the property to the event arc. So then you know you can pretty much deal with that.

A

But yes, okay, that's a pattern. We have any concerns, then from anybody else. I'll just call it a day.

A

You need to go to all our again I think one of the notes, probably just killed itself.

E

A

They're both shipped or a.

E

Consistently inconsistent.

A

Yeah I think the I think that's. Why I remember the discussion, because I think we talked about this the last hour and I. Don't remember why we learned it on the action from pattern, one because they did all right.

A

All right one down and then this one is effectively also done now right.

A

B

Is a completely different thing like it's? It's about 95 Aires, but it's a completely different API.

A

Sorry I confused this officer business, the same toy but yes, I. Think.

I

A

Was proposed on the other time, I think that's what we learned it here. Yeah.

D

Yeah, so you just change the name, it was BOM name. You can see two changes to this yesterday: label resolver. Just look at that. So.

A

What were the open issues again? Every show we this was more or less that the shape we discussed last time and I don't see any and your response to me on this one here well like I.

B

The proposal is to add one other API and I. Don't think we decided where we would edit, which cause the you see the last one there. That's the one which we discussion is about now.

D

It happens at a different place. It happens at different granularity like this is like a pearl assembly yeah.

B

D

It happens early and the other one happens late. There's no need it's yeah.

I

A

This one is effectively global right. The other one is effectively relative to some other gem. There.

D

And this one is personally, the other one was for assembly load context.

H

B

So you do the open questions here are: where do we put it like, which cause.

B

A

It, the only thing that is weird, is having a better stuff of no unregister, but I. Guess that's just the way it is I.

D

Mean we could, under this place, not a big deal or just call register with the null pointer.

A

Okay, so this is for assembly right, so you can so if you so, basically you but afterward is that for assembly is that their customers, you can basis a if you pass in Nollan, is.

E

Registering it was all were setting the resolver or adding it to a chain.

D

Sitting there saying that only one can be sending.

I

I think that's there.

E

Because register I expected means you're, adding it to a chain which so I would expect an unregistered. But if it's there's only ever one, that's that's the verb. There is that that.

A

Makes sense, yeah I believe this is their.

B

Do we want to make it to delegate in this case, since there's no like existing sync to be aligned with.

D

The combination sort of sits exactly like low-level like every trial on every pattern, but if we want to have a named delegate, that is fine with me. I.

B

Personally would like the target because, as Emma said, intellisense gets way better with it.

A

And you already do both namespaces, where it's already grabbed back so you're not polluting in his face anyway. So I would yeah. We want to say, make it there make it a callback and I can actually delegate it's.

E

Also, just yeah.

A

You kind of have to look elsewhere to know. What's what's that, so let me just try to introduce here.

E

Makes sense for you this stuff, so.

A

The reason we learn it on Marshall is because it's a low-level API and we didn't feel like adding another time for that I think originally, it was a separate type. Originally we had like here I think it was called low library or whatever, but and then it actually was a stateful type, I think they were originally, like instance, members, and they you would dispose them, and we basically all said we don't want to do that, because that what makes it appear safer than it actually is. We.

E

Don't want the disposable context, but, like everything, everything else on Marshall is about an object like every single member on Marshall is about an object, and none of these are about that object at all.

A

You say that, but um that's actually true and there's a plenty of.

J

Like helper methods, the closest.

E

Thing that I see that's not about an object. You get tight, but it's get tight, but it's you know destroying pointers or structures. Reading from an of read enough and an object is another tally. It's all about how you take a thing and you rewrite it while you're doing both is what Marshalls for.

A

Utility functions for 4-pin. Well, then I would argue this. These Memphis may clearly make sense right, like.

E

Scanning the list everything here looks like it's about um how to really an object, and this is about how to manipulate your process. Space and.

K

In other other places where this could go.

F

I'm asking the question and that objecting strongly didn't.

K

You know wasn't the original proposal to have an ADA partial class that all of these would sit on, though we would only know if you ap are equal.

C

Where was it again, progra breaking up.

D

Was a native library class it.

K

One of these method- yes, but I- thought that one of the intermediate after that one of the intermediate proposals, I thought- was to have a native Marshall class. That would contain me invoke specific functionality like this.

K

Maybe it just wouldn't be interesting not to have a class with so few members.

A

Yeah, we didn't notice, doesn't seem to capture that part, but I think the I mean it is a fair question right. What would you call it though Native Marshall seems weird because I mean.

J

A

I mean you could say, library marshal, maybe but.

K

Marshal smokey another word, especially since.

A

The only one I could imagine what maybe it's just native library, native library, load, native libraries, don't try load native library, dot, free native library. Get export ban will be horrible right.

A

Because my other concern is that, if you, if you put these types on Marshall, they are like spread out because they don't have a common prefix, so you basically have to like they don't show up like here where you have logical equivalence, it's a bit harder to discover what the you are supposed to do with it.

K

A

Again that, given that oh I think it would still go into Indian Interop services, where I yeah I mean it would be nice if the natives will be cleaner, but that's not a problem. We have right now. It's a grab-bag.

K

Service isn't the worst database in the world.

A

Yeah, depending on your I mean system is worse. They give it well considering the scenario yeah, that's what I'm saying I, don't think I mean if it would be super clean and it's a bit more concerned guy to type to it, but you've never really like plenty of time in there I think making that a separate type might actually invest.

E

A class called native library and I.

A

Think we decided specifically against him once.

E

Really I, don't remember why we would I thought that we discussed not having the free library well.

K

The original and the original proposal that I was working on a year or so ago there was an actual handle, but the handle was not a safe handle and it was not a disposable like once. It was brought into the process. Your dominance whatever, but.

E

K

Well, so that we could have instance, methods on top of them.

E

For which kind of it are you supposed to pass in to get library? Yet what.

K

Remind me does get library export perform, name, mangling.

E

No apparently get address so.

K

D

One man, it is going I.

E

Think it's system, runtime services and they make a new static class called native library yeah all of it, and then that would also I would then also move the yeah. What do you know just good? We.

K

Can always make that class and safe handle later on? If we wanted to there's nothing stopping us from doing that, I mean you can go, you can go from a static class to a non-static class, but they'll be the breaking transient right, but but.

E

The blowed returning in it when your then we would have it like. Okay, what we want you to do is call load and then call the constructor. It's like. Okay,.

I

This capture bucket will be on so whatever the names here I would.

E

Rename for young.

D

I

That's what I think bring my bleep on so this one here right. It's the.

A

I

D

B

Name right because it gets to the exact thing which is in the deal import. Yes, it's definitely.

D

B

A paths could be, it can be a path, but anyway it won't be a part.

A

The order from here right, which was named assembly and then path with the other way around right man.

B

Happiness because.

A

I

One swap these two things.

A

I

The path malleable or not, but we know that.

K

I

Mobile, yes, that's an enum.

E

J

The comic do you get past now ever I, don't think so. Right.

D

The beam will does not have a choice, but I didn't I mean you could always get it from the assembly.

D

K

When you first look at the pain, what it is not, then you look at the assembly? Yes, but you can. You could already do that. Look up by the time the stuff gets involved. I.

D

Mean you can look simply and then get it well.

B

You can look up the assembly, you cannot look up the table into attribute and it's.

K

B

K

The call site someone has or has not specified a zero import search path by the time. This makes its way through the system, to the point where you're, where them to the point where the event itself is being involved, like presumably the DLL import search path, has already been resolved, as if someone hasn't given me, anything just pull it off the assembly. Yes, yes,.

B

F

D

Could be knowledge is not there in both places right, so he still the Taliban in that.

I

A

The problem is, though, that there is the thing you give the user right, so I would say that probably we should do the work to determine if, if, if somebody didn't say, we.

B

Will do the work? The question is: what happens if there's no attribute anywhere.

D

E

Of do that extra magic thing: okay,.

D

The p-value is legacy behavior and it has a very different meaning yeah. So that's why it a libel in the first place and.

K

Mike was the reason it was novel in the first place. The reason it was knowable in the first place is because it might not be specified and if it's not specified you wanna fall back to be subway, because you don't want to override. What's on assembly, it's not specified.

D

It's actually not always required, and it has some security implications and that's why we will default to legacy behavior. Okay, well,.

E

I mean this is, if we have this this, you know delegate type and we have this resolver and we say that it should be knowable. We have to tell someone what to do if it's known right and if the thing is oh, do what do what you would do if they were given the value legacy, behavior that we should they get non-billable and we should just legacy behavior. But after but Nolan and legacy are different. Then we need to describe that somewhere.

E

Right and that's the that my common turn option. I had the do the weird extra magic thing like we could get rid of the question bar I think we should get rid of the question mark no matter what, and if we need to add a new value of do magic, then we should do matter. You can add it I think.

D

It's not simple to add a flag to that session because, first of all, second of all, it actually sort of sits on top of another enum that is passed to load library. So there, if you look at the values, they're, not linear, because they actually overlap with the load, library, flags and they're carefully crafted so that they will use only certain bits, in fact that, as immediately has a value to because that has special meaning to low library and it's never used, both of them cannot occur in the same way. So there are no.

D

There are no more bits in that. You know. I can tell so that's why I have this identify, whether this attribute or not,.

K

So just to clarify, if I don't have a GLM for search, half assume I have no gll import search. Crappy attribute on the assembly assume I have a P invoke that has dll import search, half legacy behavior. Does that behave differently than a P invoke that has no dll import search path attribute at all, assuming that the assembly does not specify one.

D

I think, if it's not present, then it makes a call without this load without research park like and if it is present, it is this boat with all results like that's. The only difference.

B

So, in short, we don't know.

E

Zero, it seems I mean I, would guess that it really means the you didn't, have an opinion, so I'm gonna do the thing that we used to do, which means that we don't need adorable, because that's what zero means I mean.

D

I said, basically, if the legacy behaviors enum is present, then the Lord Larry X function is called with load without results bugs flag. Otherwise this is just called without the search box. Well, that's the difference.

B

That do we know, what's the difference in terms of behavior right scrapping how its implemented differences.

B

Sure, but what does it do right? Do we know that right now.

D

Laura Thoren sulfur flag means basically he ends the.

D

It looks at the current directory before before the whatever the default search for us. Sorry set the search paths. There's some APH says the third parts, and this one actually used those search paths before the.

D

Before the whatever defaults expose the low-level uses, there could be moderate, I, don't know exactly what this is. My understanding so.

B

What you're saying is that, if there's no attribute anywhere, you will get a behavior which is not described by any of the in apps right.

D

Yes, that's my understanding here.

B

Maybe we can double check that and if that's true, then we would make it malleable. Okay,.

A

So, what's the summary so we roll make, so we will investigate whether we can add a minima that that's not the case and make it malleable. That's.

E

Whether whether the state, where we would pass null or the state or us0, are actually different, and if they are, then we need to question marking is not we don't yep.

B

J

E

I will be able right.

L

Alright, so let me just paste all of this in and then everybody agrees.

E

In try a little bit means fall back to whatever it is in business, but in there actually dissolved it. Do we need to convey this.

L

J

L

A

Then I bounce it back to API needs work because you guys need to figure out what to do there once you're ready with that, then.

A

We're done with that, okay, all right, so we think I hate to do that to you, but maybe, if there's contention we should just do the exception. First go to the compiler guys right down here. It seems City simple to just agree on that thing to me is also on the CRO. Yes, so for some some reason we understand that this is the exception that you guys will throw when the developer writes a pattern matching, but he didn't handle all the cases correctly.

L

So you're right, oh.

A

L

M

Okay, this that's almost correct. For example, this particular example in the in the summary we'll throw because the particular input that was passed was not handled if the particular input is handled. Of course, it won't throw simply because it doesn't handle some other case. It's only if the particular input is not him is not actually handled by one of the cases, then the exception would be thrown right. This code would get a warning a compile-time by the way saying this switch doesn't handle all the cases.

E

Required to use the language feature or know.

M

If the platform that you compile against does not have this exception, it will throw in valid operation exception, which is the base type. Yes, so.

A

That's a similar concern that IRS that have said it seems like it's. What's breaking change you upgraded in youth or different exception, but the idea is that the down level behavior is basically a subset of the new behavior in the sense that we've still performance so before the operation exception. If.

M

You depend on a throwing and valid operation exception and you catch it. For example, you'll still catch match failure exception because it's a subtype. It's only if you, if you test for the precise type of let's thrown, would be a breaking.

E

E

Feature if a library that is compiling against the latest set of reference assemblies catches, match failure exception, but they depend on a library that used a lower targeting version that library to read valid operation exception and broke the contract. In the feature. Sorry.

G

If you're, using a low level library that low level library will have a contract of what accept transition, yeah.

M

I I don't think that I don't think this is a kind of exception that you're likely to find in contracts. For example, I. Don't think it makes sense for a library to bend depend on a library throwing argument exception, it just doesn't make sense or no reference exception. These are you know these are. These are basically failure. It's a kind of failure of the program that you want diagnosed at runtime.

E

Non catch like that, it yeah that it's not a contract except remember what.

A

You raised a good I mean it's like you, basically have the the backwards compatibility or the forward compatibility issue. My anywhere newer code, consumers in all the library and it's still unfortunate, but I- think they I mean in practice. I agree with Neela that you don't you shouldn't depend on this, but in practice you would you depend on this right. I think that's people that.

F

G

F

G

Promise, let's say you're depending on their low-level library, that did not have the the new type. So it's.

G

Okay, so you're saying you're, your catch is character of inbound operation exception and that works yeah. Well, it's because the low-level library used to throw that type, but then the low-level library upgrades and you're still catching it. Maybe can you explain.

E

The other way it's.

G

E

So you so so you maybe you were both targeting. We got on that standard pillow and now you're you're, taking your library which is calling into another library and you've upgraded to you know, that's an OG one. You see this exception, you see that's actually what's being thrown, then you hit it in your case. You're like oh. This is a more descriptive exception.

E

I'm changing my catch instead of adding a new catch to catch this one, because that's what the feature does now, except this other library, because it targeted an older net standard with the compiler rejected, was the invalid operation exception. Another that's right.

G

How do you know that that old library would throw one exception or another? Well, the the case that you end up with is when you tested it, you were getting a newer version.

E

Of the library.

G

E

Remained implementation.

G

Of that library is using a switch expression and you're relying on that implementation, as opposed to relying on a contract where they say. Okay, here are the exceptions we could throw for some scenarios is, it seems very, like you know, the internals of that library is what's.

A

More, like people write code, the code blows up with an exception. They hadn't handler for it and then that's the tested that they should be right. I mean I, can practice and Ornette people are not really.

G

Explain the scenario where, if the person who does that does not run into problems in my well.

A

Let's say a ship ally, big the target, sonic or three all today right I write and then you guys omit the call for this new exception right then I have a bunch of customer, pushing back on my library and say well. Can you please support on it for two all and I say.

G

The library that downgrades, which.

A

Is valid right because I mean previously I was consumable vos you consume from three: oh, my own features that doesn't require a three or feature that just become part. My source code I know my call. That's broken right now that scenario yeah, but it thinks the creditors who.

E

Care, the other case would be if the middle library that you're, depending on, if it's multi targeted and you you've, tested it only using the three o runtime right. You always saw a match failure exception, that's the catchy wrote, but they will you send it to a customer.

E

They have a nougat binding that pulls it to the other con target or the previous version of the library and it now it's because that middle library, because the compiler had to change what it broke, because it's depending on an exception that doesn't exist like I, think it it creates for a very weird upgrade like this is the the good old you need something to have existed for ten years before you can depend on the compiler problem.

A

Set a breakpoint and see what's going on.

E

But the I would.

M

I would I would catch it in temporary code, while I'm debugging my application, yeah.

E

You can catch. The invalid operation exception then use the debugger view to see that it's a mash failure exception, which is uh glass inside.

A

If you find the type you use it, if you don't find the type you omit the time.

G

A

I never plan to have.

G

That type in libraries so.

E

Then, as far as being a thing.

F

E

System that might fail your exception. That feels like a name that people are gonna start well.

G

Before time by the name, can we stick on the corner? Should we need, should we have the type at all, or should the compiler just omit it.

M

Yeah I like getting the implementation right is sometimes subtle. They implement get object. Data with a security, critical attribute and I mean it's. You know, then we're done. The compiler depends on streaming contacts and serialization info yeah.

J

M

These are all things that we don't want the compiler to have to know about, and suddenly the compiler has to know about all this stuff I would.

A

Say like omitting time to make sense to me for, like marker attributes or something that I effectively policy free right like what, for things like this video generate a ton of code that you want to be able to fix at some point or like evolve. Iii, don't think it's super clean.

E

It doesn't sound like an API feature. You.

A

Are not wrong like the same time like how would you build that? Otherwise, basically, the only way to for the compiler to give you the developer, experience that we warned we would have to if.

E

You don't catch it, it prints match failure exception. What max million or something well? Who cares? It didn't come from? No one, its global colon colon match another exception. It.

F

Just generated it into your assembly.

E

Is a true when.

A

I say debugging I mean both like literally they're, five in your box, as well as you have an application deployed on the server, and the only thing you have is a is a catch-all ended up logs right, so at which point the text of the exception meshes is interesting right, so yeah.

E

There it's like having a static, CS file that they just really they have to write it as a resource and just inject it in the stream. Every time is.

G

There to put this type in in the ECL, but not make it public does that. Would that help us?

G

The compiler could still contest honestly.

A

At this point, I would I will honestly say, like they don't level scenario or the scenario where you have a breaking change is so remote that I'm also willing to say you know you have to change your code like I. Think that's like it's not on airfare Markway I mean the thing that scares me. The most important framework is an implied updates, an area right maybe, but this is not an implies after scenario.

A

Failure even for that member, because you have to recompile at which point my care level is already from like here to here and then the next thing is like, even in that scenario like how likely are being affected but is not very likely and I, think it's much cleaner, don't have a public API that we can document that it's in dog, that is in reference to summaries, that has a behavior and that's a compiler I depending on now the the alternative is also we could just say the compiler doesn't allow you to use pattern matching when you don't have that exception type, but that seems also overly restrictive itchin.

A

What's another fair, no so I'd rather say yep. You have that and there's this additional caveat that if you dependent exception type, you're you're, you're busted, but that seems to me fair I, don't know so.

E

Yeah, if we assume that we're adding it, if the belief is no one should be catching this in practice, then we don't care what namespace it's in you're, never gonna write it and by having an assistant my trailer exception. That sounds, like a general exception that people are supposed to work off, of which defeats the notion that this feature is going. It is a you for them.

A

That's bad, wouldn't I would guess because I mean realistically, the only thing developer cares about is the result of two string, mind which is really the only so like. Would you ever expect code to access unmatched lately, probably not, and if anybody ever once said I mean they can discover the exception time.

A

J

This program and then.

A

The message is: look it up. Okay, yeah.

J

E

System that time that compiler services that match tell us again.

E

G

You're changing the namespace yeah.

M

Changing the namespace is fine good.

E

Because that makes it more clear like this is a niche thing. You shouldn't be throwing this in your own coat. It's it's a. This is part of a compiler feature, and then they can get people who are depending on it. It's like really you're catching the exception thrown out of compiler services, yeah.

M

We should we should also have probably put a null reference exception in compiler services to shouldn't. We.

G

M

I'm serious but I'm serious I mean obviously we can't change it at this point, but whatever logic it causes us to put this in compiler services should also cause us to put no reference exception in compiler services. It's so you don't throw it. You don't catch it, but it is something that that you expect to look at and understand as part of your program as part of the language is part of the programming model. It's not something that's sort of hidden away.

M

That only does compiler compiler services is for things where the compiler is talking to itself it's for things where the compiler generates and the compiler consumes, and no one should ever see it. This is something that people see it's something that people should I mean, there's a it's a kind of failure in your source code that arises in your source code and you are going to see the failure and it's meaningful to people, not just the compiler talking to itself.

E

You know like there's like system data invalid data exception and because I feel like system data is a default template of some version of ES and you get people who start writing. You know throw new and then intellisense the you know they're. Like start writing the word. Data live like fishing for exceptions that sound right now, like oh yeah, like invalid data exception, it's like well. No, that exception has a very specific contract that you violated and it in system match failure. Exception sounds like oh, it's exactly the same as like key, not found exception.

E

So, like sure, I'll throw this it's newer. It must be better, so I'm just concerned that the combination of a general-purpose sound like match failure. I know that's specific to a compiler feature, but it's not specific to a developer. Writing a throw statement so either it needs a very more complicated name or to move to a namespace. That's not a default namespace, which.

A

Which it is right there.

G

Might be a way out that might resolve both the previous from and this prom, what, if we mark this type as obsolete right out of the gate, so you cannot why? What so you can why it's not designed to we do that. Why don't we mark no.

M

Reference exception obsolete you can catch it. Then you can catch this too, but we're we're not.

G

M

G

State is no reference exactly making it obsolete. We say: that's not recommended, not that's, not the recommended usage, the recommended usage as you catch the base, the base type. If you wanted to catch it, I, don't believe that.

E

Targeting versions didn't have the exception: I'm.

M

Sorry, I don't believe that the main reason you would catch it is while you're debugging your code right, but it shouldn't it shouldn't be up. So it shouldn't give me a any kind of diagnostic when I insert code to catch it for debugging purposes. It's just like null reference exception. There's nothing wrong with catching the reference exception. I would be very suspicious of production code. That does that, but I've written catch a null reference exception in my code, so I can figure out. What's going on. Let me do that.

K

In production code, here, no I need more than a show. I mean.

A

The other question I mean, like I, think this was also raised by them. That Mitch Vader sounds very like maybe too generic and like basically when it begs the question where that would work for verax as well. Can we include the name pattern somewhere like I? Don't know pattern match or an unmatched pattern, or something like that like makes it a bit more specific for what it is.

A

Like I think I suggested, like expression.

M

Failure exception that.

A

Works for me, switch expression, yeah I,.

E

Get my concern is just that if it's in the namespace system- and it has a generally applicable sounding name yeah, so as long as we solve one of those two, my concern goes down.

A

You should never have this arrow at least once right.

M

Wasn't clear to me whether we're gonna move the namespace as.

A

Well, yeah, we did, we settled on that. I would I mean it. It basically is not something we expect people to you mean system is kind of already like a mess, so we should not add to that, and if we don't expect people to use this type, frequently then I wouldn't add it to the system and compiler services is usually used for API is that there are public. There has to be. Probably we really don't expect anybody to like ever do something with it yeah that question it should be.

E

Stealing is not a breaking change right, Christophe.

J

Not here so well, you.

A

Cannot say no to get it I think attribute to generally try to see it and I think even exceptions. I think you want to think about the hierarchy. You don't want to deal with other people in everything, Europe.

G

Which is a little bit between the language and the exception? There's there's a scenario where you you do a switch on a single value and there's also switch where you're allowed to pattern match basically multiple values as a balk, Neal I know. We discussed that and at least for now we're saying we're we're rejecting the multi-value scenario where we were not going to attempt to capture the values. The problem, the specification.

M

Does not I, don't think the specification should say anything about the situations in which the value is or is not captured as an unmatched value. In fact, I think we think we may want to change the criterion. For example, we might want to say that it's never.

M

At the level that it is, is fine the documentation.

G

So much as the in in practice are we going to capture multi values or not? It.

M

Depends it depends on whether the multiple values are needed for the switch expression.

G

A comma B right.

M

If you have a case bar X, then yes, it will be captured because case var X requires the thing to be constructed. So in the failure case, let's say is when false in the failure case it will be, it will be captured. Sorry.

G

What is captured, it would be the a and B right. Those are the values that you did not or you handled or you did not handle right. No.

M

That's a tuple.

G

Expression, it's one value. Sorry, you can't isn't there a oh, you think the switches don't ever only apply to a single value. Yes got it. Okay,.

A

So the only thing I would say is like what you just said like we should feel free to detect not by the value in in debug mode, I. Think I'm fine with having an but I, wouldn't use people an assembly level way to override this, because we know there's plenty of people. That's like overflow that basically we want to have full fidelity for logging, a production code and I think there may not like the fact that you that give no way of doing that without like making the whole thing. Devon yeah.

M

Yeah but there's also a risk that you're gonna capture user data. That's.

A

G

I think if you have a server and you can easier.

A

Yeah, if you release your exception to the public in the web, server like you already like and think.

G

A

G

We hit this inside Rosa day and we get a Watson report. That's a and our first question is going to be what what value did we not handle? Yes, yeah exactly so in unless there's like a big saving, the privacy aspect is I, think general to any exceptions. It's not new here, no.

A

I think I think we actually have this discussion like I think two years ago, in its core, affects any information on the exception, including the message is something you should never disclose unless you're so.

G

So I would argue both in the bug and release mode. We should yes.

A

Capture values, I mean I, think I think you should probably like I, think I think my only people would be don't try to the separate release from the bug, but I mean there will be cases where you can't capture the value. So it's kind of a best-effort thing so, like you know, as you said like, if the, if the, if it's a rest track, you can right so like this. This case is where it's not possible, but what will the right.

K

M

Prepared to decide what are the precise conditions on which er it will or will not captured on matched value right I mean we have something in the implementation right now. Yes, what the implementation right now does is if the input value can be convertible to object, and it was not a tuple expression right that was not needed as a whole than its captured. Otherwise it isn't I'm perfectly fine adjusting those, especially on the basis of actual user feedback.

M

We doesn't depend on release or debug right now, but I'm also open to having it depend on release and debug I. Just don't I, don't think that I, don't we we've sat down and discussed this in three different forums and come to you know have different.

M

We leanings at different times and I and I, just don't I, don't I, don't want to do that again. Yeah. My I want to focus on getting the API I. Think.

A

That's fair, like I, think I think my only problem would be if, if you have different rules for debug and release, it gets hard right if it's the same, if the same kind of code compiled for the back and release has the same behavior, what no matter, what tricky that is, I think people can. It is reason about that.

E

Multi match was: do you need a different property for the multi match? Your answers? No yeah.

M

Right multi match is not it's not a thing in the language, the switch expression, the experts, the left hand that the the input expression is to the left of the switch. It's not part of the switch syntax, it's just whatever expression. It is there's no special multiswitch. Oh it's just whatever expression happens to be there right.

M

Yes, the switch statement is a different thing. Altogether. Switch statements can't fail.

A

Too late all right, so then let's do this and.

M

Prove all right, awesome.

A

Every once in a while, yes so back to VTech, so thank you very much.

B

Okay, although I must admit I only thirty minutes, then I will absolutely have to leave.

A

B

So this is a new thing. Basically, what this is trying to solve is fact that in.net core there's no API Wedgewood. Let the Yak invoke the same resolution mechanism for assemblies as we do during startup, which is the all the glorious logic around tips to JSON, parsing and resolution.

B

All of that- and basically this makes it really hard to do any like kind of plug in loads where the plug-in has meaningful dependencies, and so what the proposal is is to add a new new mechanism, new new API, which would let you point the code to the plug-in or the deal or the component which you want to load and the thing would look next to it or it has a depth of JSON file, and if it does, then it would parse it and it would apply the same rules as as if it was loaded or almost as if it was loaded during startup and the way it exposes the information rather than providing a full-blown object model which, which is what we do in the Microsoft extension dependency model DLL, which we already shipped as you get.

B

Doesn't actually load anything.

A

So and the reason this is a class that you instantiate is because you're holding on to the result of the apostates and I suppose yeah.

B

Internally, it stores the results of the parsing and resolution, and then it just answers questions. So the actual work happens during the constructor. The result methods are effectively just lookups.

C

E

You want to defer the work. Look because I get ru is the string of the path. They're gonna mean the each constructor possibly is hitting a different Jason. Are they all using the same Jason? Yes,.

B

E

You pointed to you.

B

Pointed to plug-in right, constructor parameter and it's the load that plug-in.

D

B

It's not just plugins right; it could be like dependency injection all of those kind of things possible. They didn't want it to call it plug-in, even though people typically think about it as plugins.

E

Should there be an accelerator for getting the one that corresponds to the entry point so.

B

That's a tricky thing, so in theory the one which is used to start yet is kind of inherently part of the default context, and so you can sort of get to it by asking questions until default for context. But.

A

B

Doesn't have the ability to give you paths at least not right now, so so what we were thinking about is that maybe there would be a way to get an instance of this cross as a property on the default context, and that would basically we did it.

E

The one corresponding to but yeah my question is looking at this yeah. If the thing that someone would want to do is say well, I want the thing that would happen if I added this word in main and how hard is that for them to do.

B

So this class only talks about files yeah when you, when you call assembly upload on the default load context, there's way more things which can happen not just files. You can maybe potential load from memory and there's bunch of events which are triggered during the resolution process, which can customize the behavior. This class doesn't handle any lit, and it's intentional. This is basically building block. This is in this is actually the it's kind of theater way around.

B

This is intended to be used by when you implement one of the events in either default, load, context or custom world context to actually get like the desired behavior. Okay,.

B

So actually there is nice, I was thinking. This is contentious, is first the class name which I mean I, don't know whether it did something which we like, and the second one is that the other idea is that this is API which can which currently makes sense online.net core. We can't really like on mono. You can't really implement it today or literally it's not the same way. It would have to be very different accommodation and I. Don't know whether the same constructor would even work.

H

On mono, then, so that's because you don't have this adjacent thing.

H

B

H

B

Can you can also think about it on like just a framework where we also don't have it, but there's like binding redirection, configs and the fusion logic, which in theory could implement something like this right and so the other I? The other thinking which I had and I again don't know is. Maybe we should have a interface or abstract base class with just the two methods, never do to resolve methods and then a concrete implementation which is based on the.net core rules, which is that's adjacent.

E

A

I mean I, think yeah. So it's something that mind the abstraction I just think like in order to design an abstraction. You really want to limitations like. If you only have one implementation, then it seemed like you know you might be unhappy with the abstraction you end up building, because once you actually do it on mono, you try to implement it. You might find that the abstraction shape you have. It's, not the one you wanted.

B

The approach is that we can approve it the way this and because we'll be doing, go to work and prototyping around this throughout night course reader, though by schedule, and so if we hit something query to learns more abstraction, we can revisit this right. I think.

A

Is as far as the name goes, let me give them that you put in the loader, namespace I, think it's probably fine to use that name, I mean component or you know, resolve our or you know there is a medically these things are all like, so overused internet that it's not really clear.

A

What any of these are, but I mean one other design point that you might want to consider is instead of having this thing, where the the s policy is inclusive, like just have a type that basically allows you to parse the JSON, far away just basic st., she ate this guy, pointing to a stream or a path of the JSON problem, and you answer the question. The context of that.

E

We don't give me the JSON path, it's like tell me the assembly and we know we don't have a so.

B

We're already in this situation, because today it's completely valid to not have to touch the JSON file even for the app during startup, and there is a behavior priority searching to current directory kind of thing, and so this will implement the exact same thing. So if the assembly doesn't have a deposition found next to it, it will do the local search basically.

K

So did you want to mention the requirement of host policy present.

B

At the ball and yeah so right now, what we plan to implement it with is by calling into the host policy deal, which is a native component.

B

What that means is that, if you happen to use this on a run time, which is hosted through custom hosts, which don't use toast policy, then this will not work and it will throw in the constructor.

B

There is a way to make it work everywhere, but we would have to introduce new interfaces on the hosting side so that the custom host can implement this functionality.

C

B

Custom hosts are a scenario which we publicly support and is recommended I'm, not personally aware of any like good example where it's used, but well yes, you can, you can say Quran. Quran is the reason it's it's kind of using.

B

The reason it makes sense to make it host specific is that the tab is adjacent thing that during startup is actually specific to the host policy hosting, for example, Quran doesn't have any of that and it uses different logic to resolve dependencies, but, and so it would make sense for it to implement this functionality differently, but.

A

Basically, what it just means: it's like it's, not a problem of this abstraction, but it's just basically, if you will a problem in the extension points that are hosting API said right, I mean you could just say in order to have a host that works. The host has to effectively answer these two questions meaningfully. Yes, and then they do whatever mechanism they see if it, if they don't pass a JSON father, they just you know, use the magic 8-ball to do that, then that's fine, as long as they give answers back.

B

In a way, it's a point in time think where what I'm saying is currently, we don't plan to.

C

B

C

Is reasonable the same way that one all right? You know you went, you know abstraction until you know what you I mean.

E

I, well, it's good to know that you currently have that limitation. I, don't think it's unless you think it would change the structure of the manage side here. I that now sounds.

B

E

Good mom, like that, would be a level thing and I would personally I'd wait until someone reported it. Yes,.

A

I mean if you really don't have many hosts that don't have that I would probably just expose a capability API somewhere. Like a study that says you know, can you actually a huge component resolvers or not the only other consideration that I ever said of calling it component dependency level? You might want to call it a semi dependency resolver, because, given that you Betty started with an assembly.

H

Okay: here's an ex fusion, oh I, love it actually look at this.

K

A

Mean the only reason I suggest assembly, as opposed to component, that it seems a little bit less overloaded and you basically resolve dependencies of an assembly.

C

Why would you name an assembly.

C

B

It does resolve dependencies of saracen, it's a bit like, so that there isn't so there's a nuance to it, which is that you can have a component which consists of multiple assemblies, and in this case this API requires you to point to the main one, the one which has two types adjacent next to it right, whether if we call it assembly, it might kind of give a feeling that you can point it to any one of those and it will kind of work which it won't. But.

J

B

Don't see that as a big wouldn't.

C

B

Fine, if you decide to call it assembly, dependency, result yeah. We.

C

Basically, creating a new term component, which we don't have right now right in this context, right right.

A

That's why it's always nicer I, said I suggested like anything like that, because components like so ill-defined, it's like what is it yeah I mean I.

M

Mean you could use the context or something.

A

So what Marlo is that context? Yeah.

K

Just because it tends to work with assembly load context, I was wondering was like a context: dependency resolver might make sense, I don't know.

B

Yours, your resolving dependencies of a given assembly right- and that's always true in this case, so that's why I simply dependent originally makes sense right, whether it's a managed or unmanaged dependency doesn't matter.

D

B

Yeah I'm fine with it I mean with the career that you cannot write like just point it to a random assembly. It will probably not work right, but that's, okay. This is meant to be a low-level API and the appropriate will need to retox to like use it for it.

A

Let me just address the only caveat.

A

To the assembly that.

A

Was it called one time but J? What is this Jason? Why called it's.

B

The same name as the assembly and the extension is dot eps, dr. Jason I see so in everywhere we just say: deficit like dr. Deb, stop Jason file without specifically.

J

A

Well, it seems to be roughly where we are right now.

A

C