.Net Foundation Design Reviews, 18 Jun 2020

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: GitHub Quick Reviews

Description

00:00:00 - Approved: SslStream API improvements for enhanced use cases https://github.com/dotnet/runtime/issues/37933#issuecomment-646228075
01:12:54 - Needs Work: CompareAttribute.Validate method does not create a ValidationResult with MemberNames https://github.com/dotnet/runtime/issues/29214
01:21:41 - Needs Work: Add support for validating complex or collection properties using System.ComponentModel.DataAnnotations.Validator https://github.com/dotnet/runtime/issues/31400#issuecomment-646249648
01:57:38 - Approved: Add easy way to create a certificate from a multi-PEM or cert-PEM + key-PEM https://github.com/dotnet/runtime/issues/31944#issuecomment-646253846

A

See it's about hair we're here for a non-linguistic another set of API review where today we have a five item back or five item specific agenda and then the general backlog, so first up to nine or two one, four compare attribute validate, does not create validation, result with member names.

A

That sounds like a bug, but somehow it's an API proposal. Anyone have context on this.

B

So so this is Arthur I'm, the area owner nominally for this, but we need print, have to actually make any progress, I've already discussed with him and mo offline on doing these things, but he was the one bringing these to the API review. So if he's not here, I, don't think we can proceed with these.

A

Alright, well put that one on hold.

B

A

B

The next one is also print out.

A

Alright, Tomas, three, seven, nine three: three: that's the cell stream API improvements for enhanced use cases; okay,.

C

So this is something which we've been primarily asked by Kestrel in New York. There are some shortcomings in the corona, so so stream, API and behavior. They would like us to fix. So that's something we discussed with Jeremy and is David Fowler, and this is the outcome. So there are a couple of problems we trying to solve here.

C

One an SSL stream is called to negotiate SSL it. My new sink calls or saying HTTP calls behind the scenes to fetch artifacts like intermediate certificates and the color has no control over it whatsoever. There were cases where the whole service, for example, was taken down because they were network connectivity problem to the CAS and were not and again I all happens behind the scenes, so you might think erasing and it's stuff gets messy.

C

So what we've been asked to do, or are we were planning to do- is basically trade, an API where not only the server can give SSO stream the certificate you want to present as the server certificate, but the complete chain of certificates, or to avoid the need to actually do the fetches behind the scenes, at least in cases where revocation its disabled and all CSP is disabled, so you give it set of certificates, and the SSRC would then present them on Amara to the peers.

C

There was some discussion back and forth about what it should look like, whether it should be just a collection or x.509 chain itself or whatnot, and the proposal is to have new container type bridge for first time ever there will be an object.

C

You can share between SSL streams and right now it would contain only the certificate and possibly the intermediate certificates or whatever additional certificates you might need, but in the future we would be able to type some other state to this object like OCSP responder, maybe as an SSL session cache or some other aspects, and we would have a place where we can share state, possibly between different SSL sessions.

C

So the proposal would be to create the SSL certificate context, have a constructor which takes the server certificate itself, take some additional certificates optionally and put it to the server authentication options and use it. If it's set to basically do the SSL handshake.

C

Any questions on this part or concerns about naming or the general strategy.

C

Seems scenario seems pretty straightforward yeah. So that's that's part of this proposal. The other part which we've been trying to solve here is we've been asked to. Actually they were to asked which are lump to the single new delegate and he asks where right now, if.

C

Let me access the base, so on the server side you can either specify certificate to use, or there could be a coal bank or elegant that the client can ask for a different server names to single server instance, and that's how it's done, basically, how you share multiple sites on single IP port and address.

C

So there's a delegate that, if you are in a situation, then you can set the server certificate dynamically to a callback. However, this callback, it's synchronous, so it's very problematic for people to actually fetch the certificates from key world or from a database. There are some people who try to do. Logging in the callback for auditing purposes has been also problematic, so that was asked to actually make this callback asynchronous. So if this action, which needs parking you can park, it do your I/o and come back with the answer.

C

The other part of the ask was to basically be able to impact more things than just as server certificate. So, for example, in case of reverse proxies. If you are in the multi multi tenant situation, maybe some sites you want to allow SSL or TLS 100 or 1 1 for legacy reasons, maybe some sites, don't you might have different cipher policies. You might have some other things you might set for sort of a server or site or HTTP server, and right now this is not really possible.

C

So after again, some discussions back and forth what the proposal is that, rather than having of the decade for outer decade, server s which takes an option bag and consolation token, you would have a value task, delegate, a stern escarpment delegate, which would then give you the options based on the server name. So you would basically postpone the decision the similar way, what we do for certificates.

C

Basically, first, the server client would send the name they wish to connect to or talk to be, would pass it to. Data like a delegate could construct the option bags with the certificate with or a certificate chain.

C

It's the SSO options, para dead site and then because this happens early enough in the end shake we could honor this setting and basically do that and either you have everything in a memory or ready to go, and you would just return completed value tasks, and in that case you would pay no penalty for the asynchronous a or again you can talk database or do your iOS and whatnot, and when ready, you complete the task and be basically good a date for the task to finish before we proceed with the authentication.

C

So that's the general spirit of of the delegate and the options we discussed is right. Now, Tallulah get would take the center which in a pass he would have just object and right now with Treaty, SSL stream itself would take the host name or the name. The client asks for and will take the and it would get. The consolation, though, could be just passed through the original call.

C

It would just that be passed to, and the discussion was whether, first of all, we need a host name because, as a proposal, the prop or paper to make the homestand available on a stream itself, so we don't necessarily need to pass it in as a parameter, which is what we did in a past. Any call back and possibly there was also a question whether we would need or who'll be used to have more information from the client, hello.

C

So right now the server name is only what extension client can ask for, but maybe we could or should create a wrapper object or container like a soft handshake info, or something like that and maybe be able to pass other information like LPN or pro kleh versions or any other extensions which may exist or exist in the future, so that b-but.

C

So now the proposal was just to pass the name as the minimal bearable minimum or again a drop the name completely and rely on the hostname we adding to the stream itself. What was the plan be.

D

In the future, we decided we actually needed to build to pass in more.

C

We would need to add the new go back or new, delegate and I know. We only have like several, so it would kind of suck.

E

Yeah, my preference would be to either make a class, or maybe a struct, if you're worried about allocations and passed that instead of the host name.

A

Yeah I agree it the fact that you're already extending something that was recently added means it needs to be more generalized yeah.

C

And you'd be right now pretty easy, so we can choose what we put to this struct or a class and could be just the host name. But we already have a hello parse error which can actually give more information. So if we choose to, we can expose the information we already have and or and if you scroll down Jeremy or whoever is driving I put it into the alternative designs.

C

So one option we could have is we can just shove the raw client hello data into this. And if anybody wants through an additional parsing, they could do that without doing extra magic of buffering up front and try to sneak between the socket and the SSS stream and whatnot and so again, I, don't know how we would feel about it. Basically providing access to the raw data, possibly additionally, to respond to your reports and P populate that.

E

F

You know I, don't.

E

Know how useful the the raw bytes will be I'm sure someone would use it, but I think at least starting with some of the other ones would be good. Yeah.

A

Yeah I feel that someone would use it and then the next version of TLS will have an incompatible framing at some point in the handshake, and people will just be all grumbly that their parser didn't work. So I would hold that off. I would only have the parsed data yeah, so.

D

It also be careful. We are concerned about allocations, I'd be careful about exposing a list because invariably we're gonna find ourselves having to then create that every single time. We call this, even if it's for a very small finite known, set that we can otherwise cache yeah.

G

D

Yeah, you know consider something like a read-only span or pioneer able.

C

Would be better because we can make it up this nice array.

E

Can change it so ienumerable we could have some cached value for, like the common protocols lists that we received yeah.

D

But I know we're all gonna have to worry about allocating a new rates or.

E

C

I had sort of similar concerns. Originally this the internal keyless parser that in most cases we would see exactly two fields, but you know it's technically a list, so I think if he could come up with a proposal there, at least for a normal case. We don't need to allocate I think that we trade and obviously you can drop it from the initial proposal, but then again, I think it'd be nice to have a plan.

C

What we would do if somebody asks for the LPM, because I could see that this could be somewhat useful, especially at a point where you try to set of crypto that, for example, HTTP to specification requires you to drop certain ciphers if you do HTTP to so. You know if I could see that in this case, for example, if the server is linked to HTTP and sees that it asks for HTTP, we can set aside for policies or do some other things.

C

So I could see it mildly useful, but I don't really know how to solve the allocation problem nicely, at least for a typical case.

D

I mean for the typical case. The only thing that's custom here is a server name, and maybe one very common application protocol right, correct.

C

So typically, the LPN would have either nothing or it would have HCP, 1 and HCP. So yeah.

D

So that's if it was exposed, that's the span right. We would just return. You have a couple of those you know in in static data and we just returned the one for HB 2 or we return the one for the combination and we'd only have to allocate if there was something special so.

C

Would be so again, typically in case of HCP, who you have two fields right or two to protocol, so we'll be spam. The raw data from the hello or.

D

Well, presumably, there's the 99% case. If there is, we can predict ahead of time which they're gonna be right, correct.

C

So we would just.

D

Have one of those in memory? Oh I,.

C

See so you would create the lists up wrong and just hand them out I.

C

D

I mean if we're really concerned about allocation I, would hand out the spam yeah.

A

I mean I'd, probably do it as a read-only collection. It avoids the complication of there being a span and in an async context. Also whether or not the struck needs to be a ref struct, and these the SSL application protocol objects themselves, aren't just I, don't think they're, just blit over the or just a projection over.

E

The client hello, if we wanted to, we, could expose some Sentinel values like HTTP, 1, & 2, that we could just easily do a reference check against not have to allocate in the numerator. So.

C

Why don't we take it or pocket and I can do more research and if he are okay of creating a basically struct or a class? Let's move is the server name and maybe DSS Oracle's, which are easy to do for now and if we choose to do a LPN and five I can write up another proposal. So at least you real unblocked on this- and we can in do more prototyping core discussion- how to do the LTL unless anybody wants to push it to completion here.

C

Would that make sense.

E

Yeah, it doesn't make sense to block this on that we couldn't even oh so.

C

So, let's agree that he would run it in Pasig name, you would pass the struct and I know if anybody has any objection to the names or better proposal.

C

And partially young.

D

The only other thing to think about sorry, if we think that this struct might expand, but question to think about is whether we want to pass it as in other than by value.

A

I mean the guidance says, don't use in in public API, but we can.

G

Decide we say that.

A

Cuz you didn't like that. It was nuanced and made me collapse. The four bullets and the one.

D

I feel like there was more nuanced than that should.

C

We make it restaurant friends no.

D

No, those those would be two different things yeah. We only make it a restaurant if we needed to have it rapper, ref, struct or if we like, wanted to try and rely on that to force some particular in use tape usage. What we want to do, but we would have to delegate, because if the struct was so big or we were concerned about getting so big that copying it was problematic with just two fields: it's not you can just be by value, but if.

G

D

Expect some day this might be sixteen fields, you know I'm fine with leaving it as as not in. If we think it's gonna be you know, maybe a couple more I mean we're talking about. We that we're talking about something on the path of establishing a new SSL connection, so there are already much more expensive things happening than this yeah.

C

D

Just just something to think, sir, that.

C

Was something would leave I pointed out it and sort of our hours thinking? Maybe the allocation be okay, since the SSL handshake itself, it's rather expensive and a school of crypto, but also I could see that you know the general tend to be conservative and avoid allocations. You can or try to make it efficient, so I mean.

D

As this I mean, we can always do if we start with the struct, and we find ourselves. You know wanting a hundred fields like we could always do it. We just do a GC memory info, which is it's a struct around a single.

D

You know a single reference field and we just push everything into asset class like we could do that in the future. If we chose good yeah.

A

D

A

It it also seems like this: is you know it's used once on a connection and, if grows to you know, 48 bytes, then yeah. It's sad that it took. What is that six registers for this struct instead of passing it by reference, but it's probably not gonna, hit six, so yeah I.

C

Feel that if you have to need I would rather than maybe again exposed to raw data and say here it is do whatever you want to do visit. If there's real need and it again it might not be ever, and if we have some small additions, maybe they can just go in directly and have to fall back to the read-only span on raw data as the backup if this would be on V or super large right. So let's say somebody was the whole cipher, C or cypher list. Then maybe it'll be time just keep them.

C

You know something else.

A

C

Know so we are good for now and we they'll drop the public. A OPN for now is that the agreement.

D

Yeah, there was one more thing that I didn't hear mentioned earlier, which is uh so the the new method that's being introduced on SSL stream, authenticated server, async.

G

D

One that takes a callback that returns a task, even though the callback itself returns a value task. So there is a mismatch there.

D

So that should probably be unified one way or the other. So.

A

Wait so this one returns a task and I mean the delegate returns a value task, but this is doing a bunch more work than just this right. But why.

D

If, if the delegate itself is expected to return a cached object, basically like a effectively a singleton, then why would the delicate be returning a value task if it's just going to be wrapped by this? What.

E

D

Not it's not actually saving an allocation in the common case to do that, it's.

E

Not just wrapping it is there there's a lot more like it's waiting for the the client hello to come in from the client and things like that, so the value tasks delegate I could see frequently returning synchronously but authenticated his server would almost always be asynchronous. It's.

C

A force so the authentication- it's will almost never finish or I, can't see how it would ever actually finish. Internally I.

D

Also, there there's a difference in use case between these two in the case of the delegate, we're the ones consuming the invocation, whereas in the case of authenticated, as they think it's the consumer.

G

D

The developer, and so, if we're concerned about it being disused, like we're, gonna use it correctly, and so you know when, when there's only one consumer- and it's us I'm much happier with a value desk, yeah I, wasn't I wasn't worried about misuse. Honestly, it just struck me that there was a mismatch there. It sounds like we expect this valued has to be to be synchronous. Most of the time in asynchronous part of the time is that what I heard.

A

What yeah so the the called the sni callback, which is what that is really yeah, so the sni callback, the async state, would be it's the first time you've seen a host and you have the the cert material to use in key vault. So you had a cache miss. So now you have to go ASCII, vault and now you've got it and now you're not gonna. Ask again, and so that's the the use case that at least as I understood it when I was chatting with about this before.

A

But most people are going to already they're just gonna have a switch statement and return an object. They've already populated is probably the 99% case.

F

Just to give the scenario like people have starts in key vault and they make requests looking at the host name on the fly to pull the circum key ball. I assume they're gonna catch it for some amount of time, but that's why we had, after the tomato to make this asynchronous in the past mmm-hmm. That.

A

Make sense but the the easiest sample app is you have the cert locally or you're just pulling it out of the search stored, even with refresh and all that's synchronous operations and you're, not you're, not using the network in the middle of deciding how to use the network.

D

Yeah I feel like out of any possible use case for value to ask this ends up sort of checking all the boxes, the 99% case. It completes synchronously, it's returning, something that can't easily be cached in the general case, and there are basically there's one consumer anywhere and it's a developer. That knows what they're doing awesome.

A

Ok with the callback.

A

The two that exists, our object, sender, technically, our object sender- is only when we call it an event, but we are usually object. Sender is there a reason that this one is strongly typed.

C

People get confused about the objects and why it stopped Jake and whatnot so yeah it. It.

F

Does seem backwards to make it object at this point, I.

D

Agree yeah. It.

C

Originally kind of was objected, I got comments on whites, object and item lessee reason to be an object are aside from we've done it in the past thought, and then people are over again I've seen github discussions where people are confused and like. Why do you need to cast then? Yes, you can cast it or not, but if.

A

C

I have something specifically snarky.

A

To say about this.

D

The the reason, the reason for object sender is a throwback to wind forms in the good old days of VB, honestly, where you would have a single called out hooked up to multiple control, UI event, handlers, dark and yeah, and you didn't necessarily you couldn't necessarily have a single class for sender. So it was just kind of object by default in this case, because this callback is associated with a very specific type, it would make sense to have that type as a center yeah.

D

It should absolutely be as a South Stream, there's no I mean otherwise. Someone doesn't know what they can cast it to. If they do cast it, they pay for it, and even if, if there was some guideline that we were archaically following, we could just rename the term sender. This isn't an event handler yeah.

A

D

A

Write in and yeah I was looking through to see if we had, especially given that the other ones were object. Sender I was looking through to see if that was a generic callback guidance that we had, but the only place that it appeared was specifically under event handler yep.

C

Should be named, the name, the parameter, the SSL.

F

Stream stream- this is all stream I.

D

Agree with David Stream.

C

Okay, so the last piece on this one, it's a new property on SSO stream, and so we don't have all these circle banks. This primarily came from a client site where, if you say I'll take as a client, you try to connect to a server. And then you have a remote certificate, validation callback that one doesn't have any information about what server you were trying to connect to, and so people do all kinds of weird stuff to kind of know which stream they use to connect to what server.

C

So it was asked to actually expose the name of the server you trying to connect to and we restore this intro stream, so it basically be only exposing internal property and on a client. This would be exactly what you asked for the server name on server.

C

This stores the value client asked for so it could be either at the string. If client did ask for a specific server name or it could be the server name from the sni.

C

And there were some discussion about how this should be named. It's called target name in in the SSL client options. We used the host name in some other places, so I, but basically this would be exposing interim property as a public for the value of officer, name, I,.

A

Was gonna ask why it's virtual, but it looks like everything on SSL stream is virtual, though, that as far as I know, that class has no real support for extensibility yeah.

C

So I basically did to make or I did it to match the other properties.

F

Yeah, why is it virtual anyway? I think these are yeah.

A

It looks like all the new, so you know, comparing what's virtual and not like all of the new authenticate as server methods are non virtual. No wait, I just can't read they're all virtual, except the one that takes an authentication options, so the newest one is non virtual and a bunch of other stuff is virtual, so yeah I mean if everything else is virtual, then virtual is okay, but then why wasn't the authenticated server? Async virtual is Mike Mike. Really, it seems like if we're saying everything on this class is virtual.

A

Then why isn't everything on this class virtual.

C

It seems like most of it. You know it's sorry, we do already have inconsistencies and I know was. He can also take it out and just make it as a public string. I.

C

Don't enough I, don't have enough history on the sss unit cell. Why some are, and some are not so I, if anything, I, would look for a pay-per-view committee to provide some guidance. Don't.

A

Be virtual, unless you mean it I.

A

Mean that's the short form of the guidance.

D

Well, my other question on it would be you mentioned it. It's not available return empty string. Is that the right thing to be expected? Should it return null I'm, not.

G

C

The astronomical bag gives you at this drink, so I would follow that sort of strategy. All right.

C

Neri again, that's even how it works. Intro lady popularized business tricking empty and we basically just return whatever is in there. So people who need to worry about mobility and whatnot, so so woody you men will be keep it as a distance trip. The virtual just leave it public stream, hostname right.

A

So so, okay, so if I'm a client and I call authenticate as client, yes, then I pass a parameter, string, hostname I, think great. There is a string telling.

C

Though, if use the SSO options, SSL client authenticated options or client there's a link somewhere click the option bag class. Has this name as a target name but I. Think if you see I'll, take it as a client. I think the examples the parameters might be name and host name yeah.

A

It looks well unless we have a implementation and reference assembly mismatch. It looks like they're called target host on the client, authentication options, property and target host in the authenticators client methods that don't ya.

C

So that was the proposal I think if you scroll down again or the author of the designs, I think that was using in target nameís an option we discussed you.

D

Know the reason I suggested host name and I don't feel strongly about it. The reason I suggested is it seemed.

C

D

When used from the server for it to be called target, name I mean you.

G

Can make an argument.

D

That it's the target name, the client provided and therefore the server is getting not necessarily its host name, but the target name. It just seems a little strange. Another.

A

Thing is better yeah I feel like target name is better because it's what the client passed his target name and it's what the reserved receive? Does the client target name? Isn't it was it called inspect the I just like looking at host name? My immediate response is that they're, my gut feel I'm Oh there's a bunch of context, but looking at it, it's like that. I I expect that to be Who am I, instead of who do I. Think I'm talking to I, know for this.

A

If you're the server and you're looking at it, it's who does the client think they're talking to, but since it's really, who does the client want to talk to that it feels like hostname is either it needs. A disambiguate, er or target name target host name is fine requested host name is fine, but just their.

F

C

Name feels wrong.

A

C

Well, something was a client host name or a client target name Miette.

A

Host name would be what's the host name of the client yeah.

C

So really it is the name client vicious to talk to and that's same on, both sides right on on client. We know exactly what they asked for in on a server. That's what came on a wire right and would be empty, or it can have a value.

A

Right, so if we think that host is important so that it's clear what what the target name means, that I would say: target host name.

C

Yep I was me: I have.

F

One more comment about allocations: ha Levi. Does that call back? Is it gonna? Take a state overload? Oh thank you. Sarah bracing.

F

Object state on that call, so I can not allocate, but I mean what.

A

What state do you want to go? Oh, this is the transparently passing it into yeah, avoid.

F

My closures that yeah I think that's reasonable. That should be a guideline honestly overlook every every um callback.

C

Function, I'll see about it as well, but since we didn't have any precedent in the SSO stream, I didn't edit.

F

Well, mostly other blows, you there's no first I guess there isn't, because there is yeah I.

C

Mean it'll be easy for us to sort of pass it through. So if you pass it in, we store it somewhere and you just you know, give you give it back to you in a call back yep.

D

I'm curious damage: do you have a use case in line when they're state, including to passing I, asked only because there's this is a little sided coin in that? Yes, you would save on allocation for a closure if you did need to pass in state, but if you don't need to pass in state, we still need to thread it through. It's still, gonna end up on a sync state machines. It's still gonna, it's gonna blow the size of those state machines. So it's not a it's not like it's free yeah.

F

um I I asked so today we have a closure, but maybe maybe again.

D

If we do like, if we think that that's a common thing, you should add it I'm fine with that I just want to make sure we actually have use cases for passing. State group I mean.

E

I think this thing is intended to be able to do. I, oh right, so maybe you would pass a database context in as so. You can query a database for a search, maybe.

A

But I mean if you, if you wanted to do state and minimize allocations, you could make one object. That holds a mutable context and then you just take a nap one delegate to it once and then it's so you're not creating a you're, not creating a closure. Every time you're using the same closure with the same target with a mutable secondary target like.

F

Horrible right and hard to reason about I.

A

Mean but it's the if it's if this is a point, one percent use case, then it's better for the point. One percent to do the extra work than the ninety-nine point. Nine percent to have the complications Steve laid out.

F

Steve but that's implementation complication, not like consumer complication. Right, that's a difference.

D

Steve question: um if we pass in a state, doesn't it only bloat the state machine if the state machine gets hoisted, as in the task, is completing asynchronously yeah, wherever whatever state machines this ends up on, which presumably would include one in authenticated as a as client, async and one in the delegate callback, it would only for each of those they would end up on this. Well, it's gonna end up on the state machine regardless, it's only gonna increase the size of the heap object.

D

If that thing gets promoted to the right, that's- and that comes back to the earlier question about like is this thing almost always going to complete synchronously and only rarely will complete asynchronously like does that factor into this discussion at all. It does, but there's multiple places. This would show up. So this.

G

D

Object, it's presumably gonna I mean I have to look at the code. Maybe it wouldn't matter, but my guess is it's gonna end up on some other object as well. Okay, that's fair.

D

But I mean to the discussion earlier like: yes, we should strive to reduce allocations. Yes, we should strive to reduce all forms of overhead. This particular code path is already very expensive. So while we should strive to do like the important things well, I expect everything else. Gonna be noise, I.

F

Think there's like APR expensive, like basically, this API ship has already hindered what you can do and there's implementation expenses and I think the latter is is reasonably it it's reasonable to say that the implementation it's expensive, therefore like why bother having why bother trying to save an allocation at the same time, I'm super concerned that, like I, mean we can always add an org, looks like end of the world, but I hope that in general we we don't disagree that, like this is a case-by-case thing all right.

F

G

Because lately.

F

G

F

Mind and we're in this meeting trying to decide what everyone will ever use this for and we're kind of making the assumption that that is cheap enough based on implementation, details right now. What it feels like at least.

D

Sure I mean there's a whole lot of network back and forth. That's required in almost any implementation of this.

F

Isn't it just the handshake piece that happens before I, even the full handshake, but just just pressing the ice tonight to get the hostname before this gets called.

D

F

D

Happens as part of the handshake, my point is the whole Sharik.

F

H

D

Is this is this? Is this delegate invoked from anywhere other than the handshake? Maybe I misunderstood? No, it is not and clearly I'm fine I'm fine with the object state I just want to make sure like yeah. You actually believe it'll be useful if it's just if it's just like we, because we can- we probably shouldn't, but it's because it's useful to we- should I.

F

Guess what worries me in general with that approach is what is up happening? Is it gets used and there's no callback? So it's? What is a closure and people deal with it because we didn't think to either to add a stake or to add a state overload, and we are inconsistent across the stack about that right.

F

It's not as though we we, we have this level of care across the stack for every car back API. We just kind of like decide which ones, at least from from my point of view. It seems that way. We don't like say: okay, this one deserves day, because this clock is expensive versus not it.

D

Also called it sometimes, it's just very, very difficult in the implementation of flow that through as well, which is why some api's don't express it. Yeah.

F

And I didn't ask for a generic because, like that, that would be worse right. You required like generics everywhere, but yeah, but.

D

If you'd like to avoid allocation, then then.

A

It becomes a negated server, async of T, which uses a.

D

A

Callback of T, which the.

C

State object, I assume it will default to normal yeah.

D

Yeah pretty much every object state we have in the platform is object. Question mark yeah.

B

D

C

And again, in that case, the penalty should be that bad right.

D

I think what you have on the screen. Jeremy looks correct right.

A

So in the callback it obviously wouldn't be defaulted because defaults in a delegate, don't either if they're, even legal. They don't really make sense to me, but do we want that one to be defaulted to null the I think I think if you're well? This is.

D

Poorly chopped, because I was.

A

Trying to get it in front of your face with syntax highlighting so if.

D

You're using this overload I think it's fine for you to have this passing Knoll in the first place. The goal is to avoid closures for these advanced cases and.

G

F

D

Gonna be like kind of being your face uselessly pass. No yep.

E

Yeah I agree, plus, if you're using a lambda, you're gonna have to have a state parameter passed there too, so the symmetry but everything yeah.

D

That's a good point like the curve said you have to handle it anyway in the lambda.

E

The SSL stream certificate context, it's called create for server right now. Is this something that is server specific behavior, or can it just be called create so that we can use it for client stuff later on.

A

So when I was chatting with Tomas and I suggested the name create for server where it can go ahead, and so, if it's called create for server I, think that it should probably throw if the server is not about or if the search doesn't isn't valid as a TLS server certificate. So it gives you client verification of you've picked the wrong services. Like your code.

A

Sign starts not a valid TLS or don't don't pick it, and so that would be the difference between create for server and create for client is which verification path does it do before accepting the certificate and returning the results? I.

D

Missed the beginning of this discussion, the idea behind this API is: we do a one-time, expensive, set of all operations and then returned something, that's checked everything and then that could be cuts through yeah.

A

Right, it's essentially the json encoded text, but for a searching, wait.

F

Wait wait but but don't know, isn't this API trust whatever it was passed in no and that other step happens before this call.

A

So I mean that so there's multiple ways that all this can be done, though the way that I had suggested when chatting with Tomas is it.

A

It takes this state here and then it builds whether it's you know offline first and then warn and then move to online, but that it builds the chain to make sure that everything's normalized and in the right order and there's no redundant information, that'll confuse the other side or etcetera right, and so it's now it turned possibly unordered input into pristine input, and that holds that data and but one of the checks it can do. While doing that normalization is.

A

Does the target certificate claimed to be valid for TLS server purposes or is missing the EKU extension altogether? All right so.

F

I, like that, what I don't like is that policy is wrapped in that call, and it can't be changed by the caller and there's no way to create this context. Outside of this call right, it's dizzi only call that. Does that context all right, can you be a bully, don't do stuff, then, if it's a default does stuff? Is there a default? That says, like don't do the policy and let me do it and then you trust me dangerous right.

A

So that that was discussed in particular for scenarios with cross certified routes, where maybe for compatibility, you're trying to present a more complicated route than the one that the platform would build to. The I would say that the problem that I understand what that one is, you can say, build this thing and and I promise it works and then on Windows we're gonna completely ignore it, because we can't actually pass that data down to as a cell stream. Sorry too, as channel.

F

Sales channel is that okay, so is that true, so the issue, my only concern is like the policy wrapped in this call. That does IO. That has timeout said that does different things like how do we control that behavior? My.

C

Expectation was that we would not do the IO. We would use the offline below here.

F

So if, if, if I passed in garbage and you did a build and it failed return false, how would this fail? Would it return null daddy intent? Cuz exact I thought the code was the code calling this was supposed to do the build and pass in the resolved like things now,.

A

This would like the in target. You would pass like sir topham unified with the cert key, and this is you pass in chain, dot, pim and you've. Never looked like all. You did was load those files and you didn't do any actual work. It does the work for you and.

F

This does io, potentially it.

A

Yes, it would potentially do I owe to find that you didn't pass additional certificates here when you should have, and then it will either find them so it'll find them from the store which is offline or from the the internet if it needed to unless internally it says it isn't.

A

Gonna use the network, but if it does I, oh, it would only do IO once and then, if that failed, when doing this, then it would be a in a in a crippled state until it the process we started or the next time you tried building it.

F

So the policy has to be tweakable. Somehow then like something has to be said about like timeouts. Do you want offline check all like should return null arthur offer for if it couldn't retrieve the actual intermediates.

E

Yeah, I think it would be nice to have a policy that said like if I didn't supply everything you needed then just throw. Instead of doing I, oh yeah, I mean.

C

You can already do that a prong. So the example right.

C

You could use the ex-59 chain with the offline flag set and build a chain and verify that you have everything.

F

Use the HUD I see so so. I read the code that this thing has internally myself and I pass in the result of that chain, build to both target and two additional certs, and then it wouldn't do it wouldn't still do a chain build again internally, just a check.

A

It would do the math, but if you've, given it all the data it wouldn't, it would stay offline. It.

F

Would be cheap.

A

F

A

Like 10 milliseconds super.

F

A

D

It's only 10 milliseconds.

A

D

Solving this one, how often does one do this once per server started right.

A

Now the way SSL stream is currently implemented, this happens once per handshake. The purpose of this thing is to make it be once per time.

I

A

Felt like calling it, which would probably right once so, there's.

F

Two two scenarios: one is a server startup where you pass in app info achayan file, and that happens like before the sequence even runs, and then the other scenario is where you really don't notice. Certain till you have a hostname, so you run this in the callback doing sync IO in that callback once hopefully per host name and you catch it there.

F

In that scenario, I'm guessing that there's still be kind of this, like flood of of connection, is trying to get the trying to connect to the UM to call the callback, and they be all stuck on this on this call until that was the case right, is that work right.

A

So in the case of the so you have a pfx sitting in key vault, so you get.

G

A

S and I call back, you call key vault, you get the pfx it doesn't you have it where it only contains the leaf cert and not the intermediate. It goes to build the chain. It has to go figure out how to normalize the chain, but then again, assuming it all works with all the like. If everything works, then all the stuff gets cached in various places, and so not just the in-memory cache.

A

If they have this strong reference now, but that you know the normal chain rules or the intermediate will get cached into the system and things in 98% of the cases and then the yeah.

C

A

Is copasetic so.

C

There's still option right, even if you fetch it from a wall, you can build the chain offline and if you see that it's missing certificates, you can do async IO yourself to fetch the intermediates. If you know where and how and you build a chain and again whatever you pass into this, it's the complete set right, it'll be more complicated code, but I don't see that possible. Well,.

A

Who's who's determining the intermediates are missing, and how are they doing that? You.

C

Called the x.509 chain build of vista of life like before you call this their a forty qualities, yeah well,.

A

If you call it with the offline flag and it fails with partial chained and the way that you would get the intermediates in the average case, is you you remove the offline flag and you call build again, which is the same as you just called this. If it's.

C

Doing it or whatever or you could add more code, and you could be unstuck and do the writes right thing, we don't prevent you from doing it. The API. Doesn't I just less convenient yeah.

F

This seems this seems fine. We just have to figure out if, if, if, if I'm in that case, where I'm doing I/o in the car back or trying to do things in the callback that ends up doing isle for everyone hitting that same hostname at the same time, right and I have to guard myself for a hostname. So I don't have like duplicate audio for that same cert, for example, from key ball right, you would need.

D

To build your own sorry, even if it's a sink, you probably want your own mutex anyway, right.

F

D

C

So I think some of the problem worry exists and you kind of have to solve it. If you are it's, it's sin re, oh yeah,.

F

To make sure that we aren't missing anything so yeah.

C

F

Sounds make sense? Fine, okay, so this will do a chain build for you. But if you do it yourself, you can avoid uncontrolled IO potentially because it will, it will resolve I.

A

Mean we can certainly add a boolean parameter to it of you, know, offline and yeah.

G

A

That means essentially- and that would mean it doesn't do the the fallback at so at least earlier code had try building it offline. If it got partial chain issue, a warning and then try it again with online and offline would mean don't do the thing after the morning, yeah I think that.

C

In a case, though, right chain.

A

C

A

C

A

What will happen right now if SSL stream gets partial chain, it just sends the data it has yeah.

C

A

It would, and that's presumably what you would have it do in in this case. If building the chain, even an online mode failed and then you would be like well. I I have only the target cert, so I'm gonna do nothing or I have the target certain an intermediate, but I couldn't figure out a route for it, so I'm gonna send but I have, and hopefully the other side knows what to do with that.

C

Right possible or how the platform will do the same, that the Maya wanna be control, happens in all cases, but.

D

A

A

We have one for revocation no.

D

Its replication, okay in.

A

The specific context of chain building, it's there's online offline and no check for what should we do for revocation, but this one's not going to care about revocation. So it's only okay for the new boolean disabled certificate, retrieval in Tanzania, the yeah yeah yeah.

C

So, are we in favor of adding the buil.

F

D

Yes, I think we should, but is there before we could call it a boolean like? Is there anything else we can possibly think of that would need to influence the operations this performs. Yes,.

F

A

If, if offline is false, then theoretically there's some time span, timeouts that could be edited, there's other stuff on the chain like contextual routes and and things of that nature, but the the the simplest mode would be. This has offline, and if you want any of the other complicated stuff, you build the chain yourself and then give it to this with puffs line.

A

D

All the complicated things got one direction. Are we gonna be here in six months, adding a new API I forgot that six just know then great I.

F

Think we could maybe avoid that if the flag was called trust me.

C

So the they talk Steve about adding OCSP, for example, for this object. So we might be back for other reasons, but.

F

But you don't have to be if you, if the flag is called like just use, what I gave you and then it's up to someone? If we add other guys to create the right context, then it may be fine, but if it's just offline, then maybe we're at risk of having to add more I mean offline is kind of a flight that can represent OCSP and CR ELLs and intermediates, but I'm not sure do we wanna do we want to well.

A

Because the OCSP stuff, at least what I had in mind, is, would never be public. That would just be that this is using it as a as a cache object, internal to SSL streams, consumption right, but it would have no public impact and the offline here that I for great for server would only be about building the original context, not saying that this whole context is offline, because the context is offline. It doesn't do anything. The fact that it right this one object happens to be the no CSP responder cache is a hidden detail.

D

The Jeremy I understood, said earlier correctly. Your assertion was basically, if we give this thing, a boolean vote that says offline use. Only what I gave you, then any other scenario could be achieved by doing what you wanted to up front and giving the results to this and then either it would succeed or fail, because you said only use what I gave you yeah the.

A

One thing that would fail is a scenario that Kevin Jones pointed out with regard to multipath chains is and again on Windows. As far as I know, there's nothing we can do about this, but I'm, not an s channel expert. Is you pass the thing you pass? You know create for server a target, here's a bunch of additional stuff, including all the stuff. It needs to build the longer path and offline.

A

And then the system says: hey I found a trusted path that was shorter, so I'm throwing away half of your data that there's no way that you could there's nothing. You can do to override that behavior if we wanted. If we thought that was important, even though we can't make it work on Windows, then we would have to do something else, and that could be that we have two different things here. One is the create for server, which does normalization and the other one is. We had.

A

You know, create and/or like create unverified or whatever, which is you give it the stuff in the right order, and it just goes.

A

But again on Windows, that's the same as you pass target cert and nothing else, because that's all s channel accepts is what's the the pointer to the thing with a private key and then it builds the chain inside itself and I. Don't think we can influence it interesting.

A

So that would be the only thing that you couldn't do with just the offline parameter.

D

Seems reasonable, that's me with the with the bool, okay or in heating, but really wanted to, but a bool seems fun. My last question is just back to the name. Is there a client equivalent to this that we can imagine in the future or is the four server just to provide more context about what we're creating I.

A

Mean there are client authentication certificates, and so this is back to the does create for server verify the server EKU. If it, if everybody thinks it's silly, that the server is verifying its own state, because I believe right, I think Tomas said it doesn't right now it. The client will just reject it. With this, the cert made no sense in context, then it would just be create.

A

We would drop the for server and share it or when we wanted to figure out how to plumb the API through meaningfully for client authentication, then create for client would be added as well.

A

I think I closed. All my parentheticals.

D

Would we consider a model? What if we, what if we started with that, just create and it doesn't do the server validation and then, if we really found we needed the server validation. We did one of two things: either we added the crate per crate for server overload, or we added an instance method that, basically, you know, took the instance validated it and returned to new instance or more validated in place and updated the state of the object or some such thing.

A

It seems fine, it's just about how much helpfulness do we want in the in the system for pre validating.

D

I'll defer to others for that it just so. It just feels weird li asymmetric to me so.

F

Cash flow does the looks, not happy gay. You so wouldn't really help, because we I think we well I. Guess today. If you pass it indirectly, we don't check of it. We could check in cash, which is fine, I, think it is not to be in this API cuz it. It seems fine that the server itself would check for a server scenario.

C

At least good behaving, except for us right, a.

F

Nice, a nice.

D

Server yeah, it really.

C

Goes between the worst.

D

That happens, or that happens if it doesn't check if it's buggy right, yeah.

A

D

A

No way, that's yeah, so the if I, if I understood the way everything is plumbed right now. If you call the authenticated server and you pass in a certificate that does not have that is not appropriate as a TLS server, we're still gonna push it on the wire s. Channel I may or may not push it on the wire. It may fail. I, don't know I, think Tomas said it worked. I could let him speak for himself, but apparently I've decided not to and then the when the client gets it and it runs the verify.

A

It'll say this: this made no sense, it says it's a code sign, sir yeah.

D

Yeah I to me, it I think we should just take what the user gave us and pass it along, and if they want to do additional validation great if they don't well, they you know it's what they get. I'd only do additional validation. If you were introducing some sort of security problem as a result, yeah.

C

And you basically ferret to the platform to handle it as best as it can, and if the bathroom cannot be just throw and if it platform can we just do the best work we can.

A

And to clarify here Steve when you say just accept what someone gave us, do you mean that we should just take the target additional certificates build? This is opaque and not normalize the chain in any way, shape or form, or that we should take it and figure.

H

Out the icing ambient.

D

Context like I meant like if, if the developer really wanted to give a non server cert, we should allow that and then maybe they're using it to validate their clients and so they're validating that we may get a add cert like they do the right things. I, don't know it just feels like it's unnecessary, additional enforcement that would generally be done by the server itself if it wanted to, and so we don't need to complicate, be a bad boy. That's my opinion, but I'm, not an expert in this domain.

C

So one thing we discussed one option is that if you give us the collection, we use it as it is, and if you don't give us the collection, they do the best thing which means between the certificates. I know it might be more hard to describe, but basically take the collection always as is, and if you don't, if you don't do it, we we do it for you.

A

Yeah I mean it it. It is still a little strange and that that would give you again. We can decide garbage in garbage out in context is all fine, but if, if on Linux or Mac, if I gave a target and then in additional certificates, I gave a bunch of completely unrelated data, then on those systems it will send everything the way that it was presented. Possibly it's the right data, but backwards and that'll possibly break a client, but on Windows we're the only thing that matters is target, then everything works.

A

That is pretty so it's you end up with a weird like defenses like on Windows. That essentially is always treated as if this was null and on on other systems. It's not, and that's actually still true, because if you I guess yeah no matter what that'll be true, if you pass in additional certificates now that Windows can't find with its chain, it won't send them because we can't pass it to s channel. So maybe the distinction of null means use ambient context and non null means use.

A

What I tell you is fine, there's a question of: do we care about ordering and removing duplicates and removing redundant and etc, but I.

A

Don't know if that would be better as a as a null input then or a proper overload or a different named method or or what to convey that this thing being null has a very, very different behavior than it being empty.

C

Dentures, don't do this or role just to make it very explicit.

C

So anyway, what are we converging.

C

Single method or just create visit boolean.

A

Yeah I mean here's what I've captured so far.

A

Yes, apparently I only chop, some of the things.

A

So rename it to create which means drop, the server, verification and add offline and I added it as default, because I did, but certainly people can object to that.

C

I think defaulting doing the right stuff makes sense to me right because there's nothing.

F

Default more likely to work and.

C

Make it possible for David who's, gonna poop? It always be yourself.

A

G

A

Final question on it in the proposal you have here, this is a very opaque container. Should the target cert be exposed back as a property.

A

The collection can be complicated, so I'm not suggesting the collection, but so that somebody in a debugger without looking at private fields, can can figure out which object. They're. Looking at I.

D

I think if it, the concern is the debugger, we add a debugger type proxy to it. I wouldn't add public fields unless there's a real need for it.

C

A

That cool Jeremy, yeah yeah I mean like logging, could make sense, but you can.

C

Be seeing someone long praying something nicer.

A

A

I'll see how well I've learned from email and okay. Here are the notes that I have, which I've chopped more things, but still not all the things.

A

So did I miss anything or get anything wrong.

A

Actually I feel like I did miss something because obviously the new context has to be added to something: yeah, the SSL server, authentication options right, so yeah, I guess a question about that is at the point that we are in the handshake.

A

Can everything that's on SSL server? Authentication options still be changed since your new callback for sni is you're. Passing back the full option set. Will all options be valid? There.

C

So, for example, there's a callback to call on yes and I was expecting. We would not call that right. So if you set it in a callback, the new Comex we're gonna be cold, but I think rest of it should be still suitable. As far as I go.

A

Okay, so at the at the point that we are in the handshake, when the s and I call back fires, you can still change the protocols. You can still change the cipher suites you can still. Obviously we want you to be able to change your certificate. Yes and.

C

We actually do the parsing can redo the callback before we call any platform. Yet we just parse the row buffer be received, and so it's still kind of early very early in in the state. Okay,.

A

All right so I think that addresses this. Unless anyone has any final feedback.

I

Those good to me.

A

The namespaces are for chumps.

I

A

Right are you here still print off, okay,.

H

Okay, Nathan got is also here.

H

So this is mostly a like a difference in implementation, slash bug and how the compare a compute works today. So, if you look at the annotations, the the validation result. Object has a property that says what member was and what member had the error, and this is useful if you calling like posit a turn on an object graph and then you're trying to figure out which property or member was like incorrect.

H

The bad attribute doesn't set this value, so you know when validation comes back out, like you can't sort of reason about what property isn't like was invalid and blazer like relies on like data annotations, for it's all this validation. So you know it uses it to highlight like input boxes in the UI.

H

So when we get like this comparator, B result that comes back with like no value greater we're like well, we don't know what it's associated with and we end up like associating with like the model itself, and you know we were trying to like figure out how to fix us. The hesitation in this areas that, like a lot of this code, hasn't been touched in ages and I. Think we don't. We don't really know how a lot of these api's get used and what sort of assumptions people make about it.

H

So we were trying to like make these changes in a way that is minimally breaking to users. So.

B

I can give I can give a quick overview of the kind of breaking non-breaking aspect here. So so these annotations, the the data annotations.

B

You know this is a set of attributes that it's basically used in two different ways. One is directly as a validation framework, which is what Blaser is using it for, but it's also used as a mechanism to theoretically in an abstract way, annotate your model with things that then can be consumed to determine how to deal with your model. So, for example, EF will really read these data annotations and take act.

B

You know, take actions based on that and there's a lot of code out there that reads the annotations and if we make updates to the annotations, then, unless the consumers are also updated to understand those, it becomes very confusing, so we're basically not evolving these api's forward as a way to annotate abstract things on your model. On the other hand, I do think it's reasonable to bring changes to their validation function, since that is a kind of well known single way of using these.

B

However, we should be careful not to make breaking changes because again, there's a lot of people already using these for validation and it can easily break applications by changing simple things. So we have to be very careful on the breaking changes from.

H

Thanks happy right, so in this case, like my like you know, I basically came up with a proposal which is something laser shaped as an experimental package, and we kind of thought was okay, but the idea was we shipped like another attribute that did basically what compared attribute does but fit like the fixed behavior, the other all it was like.

H

One was like we could just go fix it, but, like you know, clearly there is a risk that we destabilize some library that it relies on the old behavior or we could add, like an or property to the, not repeat itself. That says like. Please do this. You know basically a context by such so it's a certain to the new, correct, behavior coding code and I'm, basically open to all of them. This was my like the first. The API proposal simply happens to be because that's what he ended up doing.

A

So compare property attribute would be the exact same thing as compare attribute, except the return. Value from validation. Result is sometimes different.

H

It's always different, but like it has like the validation result, member name is non null. Basically.

B

So pronoun one we talked about this before you, we had discussed actually not creating a new attribute for this case. I believe did you investigate updating the existing attribute to also have this behavior? Potentially, you know behind a behind on option, if necessary, to avoid breaking, because it feels like in this case it's not really a new attribute. We just want to add a new kind of validation behavior into it right.

H

I listed that as an option like I'm open to that to so, but it would be something like you know. It.

B

H

Clear to me what the flag would look like yeah.

B

I think I think we should try that first I can't tell you know I, don't know the code well enough and I know there's a lot of. It is not obvious in the in the way it does particularly around you know. Attributes in dotnet framework had some particularly specific rules in the early days due to the way designers work.

B

So some of it is non obvious where it validates stuff and everything so I think I need to see like the PR to see how it integrates, but I think that would be a better way to go initially on this and then only to bring a new attribute as an API. If we really think we need something new, as opposed to just kind of adding functions of existing one. Okay.

H

So go from try fixing that implementation maybe also add another property to like sort of talking.

B

Yeah I mean, ideally, we don't need to add a add a flag or anything right. Ideally, we just do a bug, fix and and we're confident with enough reviews that we're not breaking existing scenarios. I suspect that fairly unlikely I suspect we can still put something in here. That's relatively plea doesn't involve a new attribute. Okay,.

H

I know so maybe we skip past this one because selecting the other one has like an actual API. Okay, yeah.

A

Like I, don't even know that you would need a competin, it I would probably just fix it. Yeah.

B

Except we know from history that we have to be careful about breaking changes in this and I'm, not generally a freaking change, averse person, sir. So.

I

What is it, what is the behavior difference that somebody would see.

H

So if you look at like the validation result out there, that would return like a non null property. Basically, that's the difference so, like validation result, has a property that says member name right now. It's null and like after this change, it'll basically have the property on which its applied, which.

B

In if itself sounds fine, it's just all the issues around initialization and how those things are. Actually you know when they're, initialized and consumed, because we did look at changes in this area a couple years ago and then it rolling back a couple because of the non-obvious way the attributes work. So ideally, this is just a bug fix and that's what we should go for. First I guess.

I

What I'm, what I'm trying to assess is like so the so within core I, think we have a bar for breaking trainers. That would probably be low enough that we could say you know, let's give it a shot and see where it lands. But my concern is, if you write portable code that works on full framework and core, and then we don't fix it on framework.

I

For these reasons- and the question is like do- I have a situation now, where code that works under that framework stops working on core or coded works and courses, for example, Famer yeah.

B

I

Of the the the problem that I'm trying to wrap my head around agreed.

B

So in general, that's another reason why we haven't wanted to move this forward, because it does get very confusing when it works one way and dotnet framework and one way and our net core I think this one may be okay, especially given that it's primarily for use by blazer and and so there isn't any dotnet framework component to it. But again we should be. We should be careful about not making that too confusing I agree make.

I

A

So this needs work, email.

I

A

I

Back to it needs work and probably assign it to enough, because he bother has to look at it right. Yep.

G

Arthur had a question: sorry I jumped in a little late here so I, were you able to find through any archaeology why this one doesn't get member names because I think we had it on all the other ones right? It.

B

Gets them, but they're, not initialized in does weirdness around the initially pronoun probably has a better idea of the actual code. Here, I, don't remember the details. I haven't looked at the implementation, oh well, okay,.

G

H

Look I tried looking at the history and I think it cuts off after a point like this is like here is dotnet code like dumping it in the repo okay.

I

G

This one, all all the other attributes I, think, were consistent in applying member names at the right time. They weren't they. This is this one anomalous. It.

B

May be, and if so we can just fix it and say yeah, it's a book fix and it's all good, and you know when we're confident that the compact is not impacted. Then we should definitely do that. Okay,.

G

A

I mean the only thing that I would sort of say by looking at the code here and what things are doing is: maybe you want an analyzer that says: don't use the constructor of validation, result that doesn't take the member names if you want it to be null pass null.

B

Would be useful, yeah.

G

A

Okay pronounced other issue yeah.

H

Right so next up is violation. The validator object on indeed annotations only validates like a single level of the object and like this is documented in the like API reference, Docs and so on and so forth.

H

So if you have a like a person object and a person object has an address and addresses like validation on its properties, like none of them get validated, and this Emmy see like you know, we used to eat animations, but like the babies, he uses it as that, like it basically walks the graph and then in you it calls validation on like individual properties rather than relying on validator itself, and we kind of wanted to like avoid like repeating the same code over and just move all of this into the framework.

H

So the suggestion was basically we would add, would we would add a property that say is that, like that, he would apply to a you would add an attribute that you would apply to a property that says Taos ivali validated that you need to like step into this object, and it's only one level. It's not recursive and then basically validate those. And that's pretty much all there is to it because, like the idea is that this is a new property, like consumers are like opting into this, so it's kind of non breaking so.

B

This is a case where it does feel to me that a new API is appropriate here. It's also generally useful. I think this is not something. This is not something that specifically for Blaser. It's definitely something that people have asked again. I, don't think updating the existing thing makes sense here, because it will likely be breaking and it will be less clean than just adding a new thing that does this so I'm in favor of a new API here.

H

Like so like, try validate like the object, so I think there's a property on there is there I think. Maybe I might have said that. Can he scroll on the API dev like for that one yeah when.

B

I say new API I was making yeah having a validate complex type attribute, or something like that, rather than trying to use the existing the existing attributes here.

H

And like I thought like even for the like, and do it it either and they give you up absolutely needed to was basically like you would annotate it, but also, let's say like my like: the caller has just asked, opt into this behavior.

H

B

Know you learn.

B

If it's a new attribute but um yeah I'm, not sure on that one, because people would have to be using the new attribute to opt into this anyway, right right.

H

Like my, what is that, like in case you've, had like something like I'm easy like if it happened to call, try validate object and.

G

H

That you know you wouldn't recurse into the object, and now you start seeing this new behavior, it's problematic, so versus if you added a new API. That said, like, please walk into the graph when you see this thing that that seems, but we wouldn't walk into it unless.

B

We found the new attributes right right. Nobody yeah.

H

Right, nobody thing is just like: let's see, I have an existing model in my application, and you start applying this attribute just because you happen to share it, but in place in an MVC yeah like me sees understanding of like that Model X. Now.

H

B

This is basically what you're, basically coming back to is: there's lots of different consumers of this all right, they're already they're not going to update to change what they call, and so we probably should add another method so that people who do want to understand this, you know, can opt into it.

B

The that does, though, start to produce the comp, the the confusion that we can have if you have two components like MVC and blazer and they're, both using the validation framework, where MVC doesn't understand the new attribute, because it doesn't call a new method and it's never been updated to do it. So now you put validate complex type on your members, and your boys are out validates them and your MVC app doesn't or whatever I'm using MVC is an amble, but any existing existing thing out there.

B

So in some sense it's better to have new apps off into doing the validation that you've added on to your types. Otherwise, it becomes kind of obscure why validate complex type works in one place and doesn't in another.

H

Sure yeah like that, that seems fine, like the one thing I guess like just as a part of an implementation, is that, like you, now have like additional like context and state that you need to store if you, if you want to walk a complex graph, just to avoid like recursions and things like that. So you.

A

H

Arbitrary users are now going to be the cost of it. So it's just more.

H

And I've wondered like, and does this effect like, so the thing that is kind of unclear to me is that if now in their mind, I find they're like the property attribute, only shows up it's like if you have the right package, so you have the right implementation.

B

Yes, but it requires somebody calling validate object to call the right method right, mm-hm.

H

B

For example, the f6 is another example: F record doesn't do validation, but in a f6 we call validate object in certain scenarios for you, and so again that would be a case unless we updated ef6, you start applying, validate complex type attribute, and nothing would happen and it'd be. Very you know weird right.

B

This is this is basically the general problem with obtaining this space is multiple consumers which are not going to get updated and the producer and consumer have to kind of be aligned in order for this stuff to make sense, which is kind of why we've moved away from it in a lot of cases.

B

But we could explicitly documented as it's opt-in for a new stuff and it won't work on. It won't work on all the validation platforms. I, guess, I, don't know what other thing.

G

And I'm trying to rewind my brain and even look back at some of my blog posts when we were first creating all this stuff to see if I had laid out the rationale. One.

B

G

Did I did my tour I know you know one pardon just for the random brain dump of details, I'm not sure if they they coalesce into a coherent thought or not, but and one one detail is as we're walking that graph kind of, depending on the context in which you're doing this validation, you might want different graph, walking algorithms, so I know. That was one of the reasons why we held off. On that you know doing client-side validation. We had one graph walk and server-side. We did a different graph walk, so different implementations there.

G

Also when you're walking the graph as you go to do, validate object on child objects. You need to construct validation context around each one and there were cases especially server-side where construction validation context using service provider. Implementations could actually get expensive, so it might feel like this is going to be a relatively cheap operation to validate a complex type, but it could become unexpectedly expensive, especially when you've got a raise and make certain collections of items in the mix.

G

So we wanted to make sure that was understood by the caller to and as and we also had had validation context, the construction of that and the asp net and of CF ria services scenarios. We had extensibility hooks so that you could participate in the construction of that context and make sure all the services you wanted were plugged in there too. So I don't know you know ten years later and how much those extensibility hooks have actually been used outside of dub CF grey services.

G

But that would be something we want to make sure to take into account too. So.

G

B

Sounds like there's actually quite a lot of potential complexity under here in terms of the the behavior and compatibility yeah.

G

I think you I think the way you talk through the compat I thought it seemed reasonable, but the behavior isn't.

B

G

Not clear to me, they won't compare.

B

What the or the right thing to do is and now you're also introducing the you know: there's behavior complexities. We need to consider here yeah.

A

Because one thing you know that the mention of this new attribute would only really be understood by the new overload and the new colorways are gonna get. It is if you're supposed to be interacting with the any of the colors that don't understand it, in particular, if you're going between core and framework, because you're a nougat library like, what's the what's the way that you're supposed to write, something that it works best with compat. Or is this just a if you want the fancy thing, your net five and only blazer and like it?

A

It feels like it's. This isn't adding something on to a system. It's adding a system next to the current system. That kind of feels like an add-on to the.

G

Current system.

A

And but that it to me granted I've never used this to me. It just sounds confusing and that that it would be better if there was a way that this new attribute just meant that the existing path would call into the existing object in a different way, but granted. That means we still can't fix that on framework. So I, don't know how important that compat is. This may be a you know, and we have to wait another 10 years before we can make this kind of change. The.

B

Other option here is for blazer to do something that is specifically for blazer and but that then starts the divergence of saying validation. Is you know, application or app patent, at least specific and I? Don't know if we want to go like down that path, but that's basically what we're doing for the other use of this, where it's like abstract types like EF, is just introducing its own attributes for that kind of thing, rather than rather than doing new stuff, but for validation. Again, it's I think it's not as clear-cut as that. Yeah.

G

We tried really hard to blade anything that was app framework specific. When we were creating this, so we did, we did put a an escape out you're. Looking back at some, my other notes, we did put another hook in here and it was part of what we used in resources at the time to pull this off, and that was the interface of I valuable object that if I recall correctly the way we implemented the graph walk for our server-side validation.

G

Andrea services was the we extended the domain, the entities, the entity types to implement, evaluatable object by default and we then had an implement, relate default. Implementation of those that would do the graph walk with the algorithm that we wanted and then consumers could. Your users could implement the interface themselves if they wanted a different graph walk algorithm. So we gave that default. So I don't know. If maybe another approach to this would be for Blaser to do something similar where we just use that existing hook of my validated object, I mean.

H

So the problem there is, though, like he suggests, like users, poco type like their their app models and now you're sort of pushing the problem of doing the class walk onto them, which is kind of why we wanted to like solve this in the first place. So like the one possibilities are like I kind of, like wasn't sure is: are you able to like make this attribute like net 5? Only like can we cross, compile this library and, like this? That's all this problem to an extent.

A

I mean certainly, it can be added to the net five only ref, and then that would tell someone who is trying to compile for a lower target that they need to figure out how to deal with the fact they can't use that attribute.

A

It removes a source of user confusion while adding a source of user complexity, but that may be, if you only care about validate complex type, and you only care about net five plus, then there's no real complexity for you other than perhaps the scenario where you've written a thing using this new model, and you interact with an older component that still works on net five, and it doesn't understand that you have an attribute here at all. Slightly like EF. Six would be classic example of that.

H

Right but like isn't that kind of an incentive to like sort of say like yeah, it doesn't make sense to put this attribute because the implementation isn't going to work anyway, like this kind of makes it a compile-time decision. Well.

A

But you you, you said you were building for net five. Only you put this attribute on your time on your type. You verified it using whatever component of blazer is going to call the the new method.

A

Here everything looks copacetic and then you happen to also send it through I think someone's at EF six, and you don't realize that EF six is not aware of this attribute and it does not do what you expected it to do and even though you did a make sure a negative thing happens, you only did a negative thing in the case that it mattered and you forgot to do it in the case that you weren't aware that it didn't so.

A

It like this is one of those as a user of systems I hate when they do magic like this, because you can't tell when the magic doesn't happen and I mean.

I

This problem about to be it's all right, because attributes are always basically some part of the stack has to honor that and if some party doesn't honor, then that's what happens, but in that sense, there's no different from let's say, malleable annotations that ef6 poly will also not be lightin Timon right. So in that sense, it's kind of the problem that we have a bit of every attribute right. This one is not special in that way. Right.

A

But this in particular of a very general feeling system that is used an independent source or used in independent contexts in independent ways of you, may have learned that EF respects these attributes and don't understand it doesn't respect the new ones.

I

Yeah I mean I agree with that I mean that seems, I mean the nice thing that this kind of thing does is also pay-for-play. Well, you could say we don't update a year 600 enough people complain every may consider doing that. Right, I mean.

B

That's, which is also.

I

Fine I mean, like we said many times: it's it's um it's in its maintenance mode, right, so I, don't think that's unreasonable to say you know we innovated in an area that not all existing tech minor fits from right, but that shouldn't mean that we don't innovate in that space. Yeah.

A

I'm, just making sure that we think that the value that we're adding is exceeds the confusion that it has the potential to cause right.

I

I mean to the earlier point of it if you were to lose attribute for an ED Center to Oh library and then existing stuff doesn't work with it. That's more problematic because we basically go out of her way to retroactively. They make it available to the past of what updating said past. If we only edit in at 5, then well yeah. If something old runs on that 5, that might still not know about it. But that seems easier to explain than the reverse.

A

Since it says, validation, complex type, attribute, extends validation attribute. Does that mean that things like ef6 are going to do something here? Just not the something you want it to do, and is that good or bad.

H

Basically, yeah probably ends up like looking at it and saying like it, has nothing to do here because I don't like the validation itself prophecy. It's like.

G

Could it be implemented such that this validation? Actually it's validate method implements the algorithm they wanted to implement it. It sort of by nature becomes recur and all callers of validate object on this with and yet that behavior. So.

B

I thought that I think was kind of what I was suggesting the beginning. Although I like the idea of putting the implementation in the attribute, that seems that seems good, but I think I think the issue there is that you're going to like MVC is going to start doing different things. When you put this attribute on which pronoun was pushing back on a little bit.

G

B

Think that's actually ok, because you started putting this after me on and so you've opted it into it doing a different thing. By doing that, yeah.

A

Yeah I think it would be. It seems like a better thing if the the validator class engine, whatever thing is what understands this attribute and or like that that effectively the attribute is what says the thing is different. Instead of you had to call a different method to get it to to have meaning, and then that that means that, yes, you slap it on and and it it's consistent in all the places that would have previously walked it with the the about the base validation attribute.

G

Yeah I think we'd have to find find all the places where we are calling that, though, and see if there what differences there are in their approaches.

G

B

I think validate object is there's a raise reasonably generic in not generic in technical, in the sense that it's it's a it has a broad applicability. It does. It's not very specific, I can't think of way implementations we get horribly messed up by this new attribute. Certainly if it's not there and nothing should change it feels like probably having it. There won't also be problematic. I feel like that's. Okay, I feel like if we start validating this after view as a new attribute, and we do it with.

B

While you still call the old API, then as long as that validation in the attribute works and is reasonable, it should get every consumer should get it right.

B

Just maybe that they'll do additional stuff like NBC, because it thinks so I also still have to drill down and do my own drilling down into the graph and do things so you may get.

G

B

Stuff that you didn't want, but it shouldn't be problematic, I, don't think right. It would just be really redundant work. Not yes, yeah.

E

H

So I think one of the issues with like using the attribute itself is that, like that, if you can only return, like one validation result at a time, yeah yeah, if you wanted to get there like an aggregate resolves like something needs to be able to like on graphic yeah.

G

And, and that then leads down the path of what do you have? Is the member name for those uh those deeply nested validation results? We had that as a problem as well train.

G

You know and I think over the last ten years, we've in that we've solved this problem across every platform in the world of what syntax do you use to show that graph walk down an object, hierarchy right and you know saying: oh, this particular validation error occurred on the third address in the collection on the you know, this particular property of that element of that collection on this subtype you've got to invent that syntax and solidify it, and we completely dodged that problem by saying it's up to the callers to do this and then figure out how to serialize those results.

G

A

G

Yeah like if the.

A

Existing method has to figure out how to essentially be aggregate exception on this validation result, because it's complicated and it and it has to make up some sort of syntax to express what went wrong or that multiple things went wrong or whatever, like that at least means that, like failure happens, when failure should have happened and success happens when success should have happened and then maybe there's room for the other overloads of I.

A

Don't want the aggregate exception style, bundling I want the the details, give me more than one result per per potential call and then that's the caller gets to opt into I want better details, but no one's confused by the validator didn't run yeah, because I think it's more that that what is more important is that success means success and failure means failure and success doesn't mean I. Wasn't aware that you wanted me to fail. Yes,.

A

So that would be sort of what I would strive for and I assume that's what the I guess I didn't compare, how different the signatures are, but I assume that's. The idea behind the new members is to get the the details instead of a notion of aggregation, yep.

B

So um so then, if I understand what you're saying we introduced the new attribute and we update the existing calls that people make so that it will honor the existing attribute and fail if validation of that fails, but we may also add new API that lets you do something more specific. If you know that you're also consuming there, some want to and want to get special information out right.

G

A

That absent knowing how any of this actually works and how someone uses it yeah.

H

Like that seems like having it like feeling like like having it like, not know or like not to anything, seems very problematic. I I agree.

A

Okay, so now do we need to talk about the new, validate object and try validate object.

H

A

I have over here a.

A

There now they kind of fit on the screen.

A

Now they fit on the screen.

B

So we already there's already a validate object right. Yes, it's.

H

Got that extra property that says duty aggregate unwrapped, the aggregate I see so it's basically one additional property dr. on. We could make a different word load instead, I'm kind of so given.

B

That these things already generate a collection of validation results anyway, wouldn't we just add: one validation result for every complex type thing that we that we validate it or would that is that not reasonable.

H

um But wouldn't you want the individual like, so the idea was in one case like you would have one validation result for like the aggregate result right like, in the other case you sort of on the details. This is the API that says Kim. um Give me the individual evaluation results for each of the details, because, like the consumer has no way to like unwrap it, there's no API, I, guess.

B

What I'm asking is is the differentiation between the details and the aggregate is that is that necessary, or is it I mean, given that you just treat this as a list of validation results anyway? Can it not say, okay, you know at the top level is failed and then at the next level this failed as another result and then- and you know, on this one- it failed I mean you may get lots of things in your results, because you know basically validating the entire graph.

B

But then you I'm not really sure why we need to differentiate those from other results that you get so.

H

I mean like it isn't talking about like what NBC and laser, like you kind of reconstruct what UI element triggered the validation failure. So you want to like people to like, highlight it or put a CSS attribute on it, or something like that. If all we had was like well, something in address field like it's kind of useless, because address might not even show up as a UI element.

B

Right, but doesn't that go to what Jeff said earlier about? You need to define a path to say: it's not just a dress that failed, but it's this the dress here here here. Are we basically punting on that by adding this flag and saying we're just gonna give you them and you figure that out right.

H

Yeah, like I'm yeah, I, necessarily think about that, because the thing is like I was just trying to like, at least in the the example that we created, like you, kind of constructed the pouch as if they were like C sharp parts, but like obviously, that's not that doesn't.

G

H

Very well like there are points, but it sort of falls over and even if it does like even can construct it. Like correctly, like you, consumer does have to parse it like, like it's a it's, a format that you have to understand. So it's interesting yeah.

B

So that's right, so people trying to pass that format which they probably shouldn't do but I'm sure they are doing, would potentially break.

G

Yeah I was I, don't even remember if we ended up with a string based representation or of that or if we were actually just admitting a true object graph that mimics you know mimic the shape of your I feel like that's where we actually landed was having something that movie the shape of your entities, but instead of it being the entity data, it was the validation result. Data feel like that's where we actually landed, but that was all generated code.

G

The chilling for ria services would produce that. I don't know what envy would and I guess. Asp net doesn't yeah pronoun what it? What is asp net tune for building up the entire graph right now so.

H

Amy's he walks it, um but it thing is just like it does reconstruct the path from strings.

G

H

You know obviously like there are a bunch of code just trying to like hypothesize what the path could be and then like crime map it back to a property. Pleaser does the same thing, but, like so far likely it hasn't had the problem of like doing anything more than like looking up member names, because that's all you get from validation today, it's just kind of what we're trying to solve I see.

B

So having the pool flag here definitely makes it safer in terms of not changing the existing behavior for anyone. I'm not calm, still not 100% convinced it's necessary, but if we want to err on the side of not breaking, you know existing expectations and breaking the front page of applications. Where now the validation fails and nobody can log in and you might want to go there yeah.

G

I'm pausing a little bit, though, if we do find that over the years we have multiple graph walk, algorithms that we want to employ. I, don't know how to it. Just having one bolt makes me worry a little bit about about them, I'm, actually I'm, looking at validation context to see.

G

If, because another approach here would be to actually put a marker on validation, context to say, hey when you were validating this object, we want the thing invoking validate object to respect this flag on context and then different app frameworks take blazer or whatever different firms could put their own markers in there who.

B

Creates validation, context.

H

H

G

H

Or Emmie's he would create it and then.

G

Joy's or MDC would then respect yeah.

B

G

B

Trying to put this in something in validation context, rather than a bull flag.

A

Yeah I was gonna, say the the boolean suggests, given that there's now two adjacent bully ins that instead of adding a boolean, it should be changed to an enum or a Flags, enum or something but properties actually does now I.

G

So yeah I didn't get to finish. Reading on that, I think it was a multipass on required fields first and then only if all required fields have been satisfied. We do the next level of validation across everything else. You can say validate all properties, I believe that's what that it was go ahead and validate all properties, regardless of required field validation, because that was cases when you could actually have validation, attributes relying on required fields already being fulfilled. Yes, actually have all refs.

A

Yeah, if the caller is building the context and passing it in then it seems like that already should be sufficient to disambiguate the complexity of the new attribute yeah.

B

I think we should try that before adding a pool, definitely.

H

So and we forget to mutate the validation context like if you're adding to it like. Would it make sense for a way to ask the caller to specify the name like construct a name for mice and then like? We would have some nominal way to like constructed for you feature creep, but yeah.

G

H

You can construct like box brackets or you, but like somebody has to like feed that on the other and then make sense of it or black.

A

H

You, don't you can say like well, I know these things actually much I wanted to do was like. Have it be validated.

G

H

I mean we can start the day easy one like and I think at least four Blazers scenarios like we can try and keep the implementations simple enough that like we would do it and hope that nobody actually wants to like parse these strings, because I don't think it necessarily say that these strings are meant to represent paths or like them like they're, just strings. It means so like if you happen to parse it. Well, it's your own fault. If we're.

B

If we're explicitly saying we don't want people to pass this, we should at least document that, because otherwise I mean people will anyway we're gonna fast, you serious at the time and they shouldn't, but we should document if we're saying this may change. So don't don't rely on this right.

I

You can say that, all day long, if, if Corky was working, they will rely on it. Yeah.

H

And I mean if we do like an ability to construct a part like if you happen to like somebody needs it, we can do it for now. So I guess I can keep it simple and just like make some.

G

Brain yeah, yeah or maybe even just kind of dodge it and say all right, something on a dress. You just say just stop at a dress say your address is invalid and not even try to represent. Why and then. If somebody needs to understand why a dress was invalid, they can come back and reevaluate. Just that address object.

A

Alright, so the notes that I've captured our new attribute seems fine, but be careful and how it gets consumed. So it's not a surprise. Users use the objects in multiple contexts, and it seems like some new data belongs on validation context. Instead of overloads to try / non try validate object with some more words, but it sounds like we don't know what that would be right now. So sounds good. So I think that this goes back as more work.

A

Alright and well, we are technically out of time, let's see if anyone wants to just rubber stamp a thing that is the wrong button.

A

A

All right, you've already rubber-stamped us, but uh maybe now that he's here, he'll care slightly. So one of the things that we talked about during the ssl stream API improvements was the way that people load certificates classically in dotnet. If you want a certificate with a private key, it either came from a cert store or a pfx.

A

We've been incremental e, adding support for how you can do that after the fact, based on like assert that PIM and a private key file, or whatever names, cert bot, uses by default, and then this is taking where you could have written utility methods yourself and putting them as static methods on x.509 certificate, 2 and X 5, and a certificate to collection. So PIM is PIM just means PIM these days. Technically it means privacy, enhanced mail.

A

But this is just one of the encoding formats of a certificate and then the idea is, you can create it from you load the contents of the file or the you pass in the path to a file. But since we're using a span of char for different purposes, this is PIM file and this is PIM and then that there's optionally, a path to where the key lives. The difference between this and the existing thing.

A

If we see a file in this format, is the current loader if it sees a certificate and then immediately after it a key, it ignores the key, and this one would not ignore the key.

A

So the reason this is a default path is, you would pass null if you think the certain key are in the same file or that there's no key, and if the key is sitting beside the certificate, where beside could mean anywhere on the filesystem, then you would pass in two paths and then these are if the key files encrypted same thing, for you have the raw data aka, better unit testability and then on the collection types.

A

We don't support key loading, it will only load the certificates, but right now the collection, if it imports from PIM that only imports the first one in this will import all of them. Ok, anyone want to say we need to discuss this. Next time or rubber stamp or anything else,.

I

The only thing is like do: we already have create methods on the certificate class know.

A

Everything it does is via constructor, but this would be introducing ambiguity of what the Constructors mean. Yama.

I

Proposed to do there, I was just wondering whether the existing pattern for our factory methods, but if there's none, then it doesn't matter collection.

A

Has instance import, and so this is, you know, adding words to say why this one behaves different than the existing instance import methods and then that the x.509 certificate to does not have any current creates. We could certainly make creates that look exactly like the constructor. If you think that once we have one set of creates, we want parody on having all the things be, creates and, and then we sort of deprecated doing new things with the Constructors.

I

Maybe I mean this doesn't strike me necessarily as necessary. It's more like the you know what Pat under your they have for factory methods. How do you call them? But if there's none than that, this is establishing the pattern.

I

I think the pattern makes sense, I mean we know we had this thing in Jason, maybe quantification over load versus parse and that confused the ton of people so I actually like the idea that we just subjected to a file when it's a you know loading from a physical thing on disk rather than taking things from a memory. So that makes sense to me yep.

A

Yeah this one is you've already loaded it or you're a unit test, and this one is load the contents and call that one, but it it saves the people from writing file that read all text.

A

Don't even our people still on the call got a couple yeah.

D

I have a quick question on the password overload, so remind me again how how the read-only span of char is interpreted. If you specify it for a password.

A

A

F

The end of char is like yeah right so.

A

Read-Only span of char is interpreted the way that it needs to be interpreted, which means it's.

A

Utf-16 le, if it's a PF, X or or it uses a KDF that uses the the PF x KDF and if it's any of the simple KDF, then it's utf-8. If you pass the one in byte form, then we don't normalize it all and we just pass the bytes as they are as input into the KDF, that this is just copying. What we did for a symmetric algorithm import, their import, encrypted, pkcs7, okay, awesome, not seven! Yeah.

D

I just remember there be like saying some weird logic behind that and wanted to make sure that that it was reflected here as well right, bytes.

A

Is the complicated one and it's for if, when we picked utf-8, because the KDF s-- work in bytes, instead of chars that if we picked utf-8, but you happen to know that you're interacting with something that did you see s2, then you can give the bytes and we'll just pass them straight through instead of failing so it's it is a very it's. In fact, it's super nation. We could say that we don't need it on the certificate. One that'd be fine with dropping it here.

D

I, honestly, just defer to your judgment on that. You understand the scenario is better than we do. Okay,.

A

And then the only thing, that's weird, which is again good- that this is from a method instead of from another constructor overload, is if you load from a pfx the default is the key file gets imported into the system, key store on windows, and this one will be as if you passed the ephemeral key set flag, meaning the key is only in memory, so this actually won't work for SSL stream. You'd have to take this load. This export it to pfx and import it to BFX because of bug in Windows.

I

A

But Kestrel could like Kestrel, will have an accelerator for like loading from these paths and it will have to do the export and re-import to to work around that. Windows SSL stream only works with key files that are persisted to disk sorry, with cert keys that are persisted to disk, so but people don't use, pin files on Windows anyway,.

I

So people don't use what they.

A

Don't use the cert that PIM and proved a key. They just need to use the pfx in the cert store. So there are some oddities it'll be documented, we're hoping that Windows will fix their their bug. They've said or they've acknowledged that exists that so that's a good start, but.

I

Make sense so then, let's approve that here and then call it a day. Okay,.

A

Let's see, are there any notes from chat.

I

F

A

Yep all right so, unless Kevin feels like arguing with me, offline I'm, gonna, remove the byte based password, overloads and then hit approved.

A

That's what a copy the whole block don't funny.

I

To be used to time offline to actually mean online off.

A

A

Yes, that's because there is no longer in offline world. I am sad that I will not be in Switzerland next week. Will.

I

Be in Switzerland next week, no.

A

I was supposed to be in Switzerland next week are.

I

You supposed to ice tea.

A

Now, I don't or won't be because you know reasons.

I

But that means you're entirely neutral.

I

Please get some Swiss cheese just for the for the field.

A

Yay all right and with that I believe we are done for the day.

I

Awesome blocking as well just for shits and giggles and then.

I

Awesome thanks.

A

Everyone, and have a nice day thanks.

I

To hosting bye.

A