.Net Foundation Design Reviews, 24 Mar 2020

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: .NET Design Review: GitHub Quick Reviews

Description

00:00:00 - Approved: HTTP version selection https://github.com/dotnet/runtime/issues/987#issuecomment-603384663
00:16:50 - Approved: Create APIs to deal with processing ASCII text (as bytes) https://github.com/dotnet/runtime/issues/28230#issuecomment-603439786
01:42:07 - Approved: Move appropriate parts of Microsoft.DotNet.PlatformAbstractions.RuntimeEnvironment into the shared framework https://github.com/dotnet/runtime/issues/26780#issuecomment-603448231
01:55:15 - Needs Work: Add mechanism to handle circular references when serializing https://github.com/dotnet/runtime/issues/30820#issuecomment-603451353
02:04:50 - Approved: Add a copy constructor to JsonSerializerOptions https://github.com/dotnet/runtime/issues/30445#issuecomment-603455704
02:15:16 - Approved: System.Text.Json ignore only when null API enhancement https://github.com/dotnet/runtime/issues/30687#issuecomment-603461964

A

A

Friends, we are now live again, hopefully with audio that does not suck quite as much as my last time and yeah. So the few issues you want to go to before we do the backlog review I mean. Maybe we never get to the backlog because there's like six issues or something, but you, let's see so Cory this one is yours.

A

B

C

C

D

Or the since her services- and they know exactly what version they need I.

A

A

So I was about to say, like this sounds familiar because we already added the default version. They should be client, I, think recently right and the version one yeah, and so this one is effectively controlling.

A

It gets on the default one. It makes sense to me, but that I don't understand is what that means for the requested message.

D

So the request message version right now is also a maximum. So if someone wants an exact version, they have no way of specifying that right now,.

A

The maximum of what.

D

It's the maximum HTTP version that we will use. So if they ask for HTTP 2, we will, depending on the URL, we will either connect with TLS and get HTTP 2 through version negotiation, or we might get HTTP 1 through that version negotiation or, if we're not on TLS. We just won't try and we'll go to http 1 immediately.

D

So this this enables people to have more control. Currently they can find out which HTTP version the requests ended up going on just by checking the response version, and but sometimes you want to make sure your request goes out in a exact version.

A

Yeah I have to say, like this is going into the direction that I was afraid. We would end up with and the Edit the two properties of like ok. Now we have more more policy and all the version number is actually being selected, but I guess that's kind of what needs to happen.

D

Yeah this this becomes either more useful with HTTP 3, because that one is not over TCP. So we can't use TLS to negotiate the version. If someone has something that they know is HTTP 3 and they try to connect to that, they will always end up going through one or two right now before they get to 3. So this enables them to say like I know, HTTP 3 exists just try that and exactly that and pre-negotiated clear text. Http 2 is something that James was asking for with G RPC I guess some of their customers.

C

Are interested in that yeah? It's.

E

A clear text or h2c: are you HCP over clear text where there is no negotiation channel and again I? Think a spec are not try to discourage that, but a lot of people want to do it for micro services.

E

For cases where, for example, the GRP sees use cross a process, communication and I agree with you more, it's kind of tricky I wish. We have some easier way how to do that, but the Kalama is set on a path on being the maximum and sort of being australia be smart and it doesn't work in all cases.

F

If it makes sense to have a version or lower, does it ever make sense to say request version or anything else, I'm I mean there's no option like that.

D

A

F

Requested version first yeah requested version first and then either go higher, lord, whatever it would make it the most sense. I guess yeah.

D

So the way a user could ask for that with this, if they set a version of, for instance, htv3 and say request version or lower, they will get the correct behavior of starting on one or two and then upgrading to three, if possible, so that that kind of accomplishes that but I'm not really sure when a user would want to say just give me any version where the defaults would not be fine for them. Currently.

B

So, looking through the issue description, um it's is there an actual scenario written in the issue, description for where you would want to cap the HTTP version at a maximum value like a the the minimum value scenario, totally makes sense, especially if you know for a factor using grps thing like you it's done previously, but capping, though.

G

Crappy I mean capping, is the current default, so it's.

D

The current default and it might it also be used if stay a server advertises that it supports HTTP 2. But you know for a fact that their implementation is broken. You can say just stick on one one, a sort of thing: I say: ok,.

B

That makes sense yeah.

H

I mean the alternative that I would think of which would be hard, given the current property names would be that we want to express it as max, but if we go with this, one I would suggest that request becomes requested because usually the agita is the past participle and not the not. The active verb form.

A

Why? But it is the gnome right, like the the requests version number, if you will.

D

Yes, now reddit is the verb to the request. Message version, but requested by I can see that making sense to it.

A

Either one works for me so.

H

Yeah that that interpretation did not occur to me at all, while reading this I read it as the active verb request.

H

English is hard you're.

I

J

Did you know about English nothing.

A

Well, you were English, you speak the wrong.

B

Sorry and I my audio cut out for a minute earlier, like did. Did someone bring up the idea of saying I'm requesting exactly these specific versions and any one of them is okay, I, guess that would be mid max almost and.

D

We did think about that, but couldn't quite think of a a use case where someone would want. You know just one and three, for instance, skipping to that sort of thing. Buggy.

B

Iot devices yeah.

H

But I feel that on that one, you would just like it's work. You can work around it with caller. Behavior of you say that you request exactly three that fails. You request exactly one. You remember that one is what worked on this channel sure.

H

Man, given that I think it would be hard to add min and Max or to change the semantics to min and Max, because then we'd have min Max and default, or default and Max. This is probably the the cleanest enhancement.

A

Yeah I mean one question: I have is like: are we piecemealing more and more of these properties in? Are we eventually better off just having a method that allows a little bit more? You know, parameters or whatever, rather than individual properties, where it becomes progressively harder to understand. What's going on or I'll be confident, and this is it.

D

And I'm I mean I I'm, trying to think of what you could do in such a method that you couldn't do when you are constructing the HTTP request. Method message: object.

A

I'm thinking of like methods give you a bit more flexibility on shape where you can have different kinds of operations with different overloads right and then whatever you call it last is the effective policy right like the polymer properties, is around three or four of them. Then you have to reason about what you set them all like. Who is winning over whom right? We have individual metrics kind of clear, like what combination of data is used a particular point in time.

D

Catch it I missed interpreted that to mean a callback you're just saying an overload instead of a property yeah.

A

Because I mean, if this is it I think I'm? Ok with that, but if we're adding more and more policy properties where eventually you're like well, what if I said this combination of property is what does that actually mean? It becomes harder, and if you have different methods where you say I, don't know, you know, use request policy, and then you do something like this. Then it's a little bit easier to understand. What any given point in time is the active policy right, but I mean if this is it I'm?

A

Ok with that, but if we add more and more than eventually becomes a bit of a.

A

Yeah, although maybe.

D

um Yeah I mean I, guess we can't know what will come in the future. Everything currently is specified through properties, so it kind of fits the style yeah. We have right now.

A

Is there any app context setting for these? No right? The app context is just for the global policy right there.

D

Is currently an app context, switch to enable unencrypted HTTP to and putting this API in will allow us to get rid of that. Alright.

F

Did you say earlier that today we we do the max.

D

Today version is a maximum version. We might end up sticking your question in order. Okay,.

F

So here it shows that the default would be version or lower. Is that gonna potentially break anyone? That's what it means to be the max version or lower is the max okay. So the max thoughts at the current version, so you specify a version. Two and you're saying give me version two or lower okay, so you can never specify a version greater to them.

G

Whatever right, basically, today, you specify one version as the version and then the system might use that or it might use a lower version, but it'll never use a higher version of what you specified.

F

Okay yeah: this is confusing max and higher I mean it kind of was the same but okay, so so the default is what we have today. Them I've been done: Sean I'm, not API. Here, yes,.

A

All right, then, in that case, unless somebody's any concerns, I will just improve it. As this yeah.

H

I mean I think that perhaps the question the emo was asking earlier in a different form is: do we think that there's going to be enough complexity going on the HTTP version policy should be a class and that someone can implement one or do we think that the enum is going to suffice for the next at least two releases.

F

And just to chime in a lot of the version detection, you know there's a lot of times: policy like hey I'll, take a patch version or a minor version, but not a major version. I, don't know if that applies to h2p. That often, but it's a common practice anyway,.

D

Yeah the the version property already has patches in it. You can do 1.1 or 3.1. In theory,.

E

It this look in the generate, so the version structure so also I would say, looking at the speed of HTTP to stabilization and HTTP, 3 I think it's probably years and years before we even kombis HTTP 4. So I think we correct me if I'm wrong, Corey but I, don't think we envision any new versions or changes and if I'm soon right I mean.

D

I I figured HTTP 2 it lasts longer than it did so, we'll see how long 3 lasts.

B

And just wait until the next doesn't use integers, so yeah yeah. Then.

D

We've got a much larger problem, yeah.

A

Thanks all right, then I mean worsening is almost always sad right. So in this case it is what it is.

B

J

B

Until you have to negotiate extensions, not just a version not only fun.

D

Http two and three have a mechanism for extensions right now: yeah.

B

But do we expose them to a public API know.

D

What they did, maybe they're kind of just send some data, and one side knows how to interpret it right now, so it would need a bit of.

D

One thing to consider with this API is that we're treating it as an explicit opt-in for unencrypted HTTP too. So currently, we kind of protect you and say if you gave us a HTTP URL.

D

But you asked for HTTP, we just don't do it, because we would have to do it unencrypted and if you specify a request version or higher or exact, we treat that as an opt-in. You know what you're doing you want it to be unencrypted.

A

How would that fit in there be different in a member stand or movie do with this being orthogonal choice.

D

Just between these three um choices, two of them will enable that option.

D

And that is consistent with behavior of like curl, for instance, where, if you ask for http/2- and you give them a unencrypted URL, it will try unencrypted HTTP I.

E

See so this tells us to get rid of the app context or you know, get a programmatic value. How somebody can do that.

D

Yeah and the the thought is that, because this is not a default, it's it's a an OK way for people to opt into that. Behavior.

D

One way that it might be used is, if you have an HTTP 1 connection and the server advertises that it can upgrade to HTTP 2 clear text. We would then take that choice, because you're already unencrypted and you're fine with it.

A

A

Yeah, if you feel like you, need to make changes come back to us, but I have our time visualizing, the other consequence. Okay, all right, Levi, we need to do is hang ASCII text.

B

Yep, so this was a request from a few partner teams like how's, her SDK and others as well. I actually made some changes to this.

B

From the last time we looked at it and just to kind of clear up some of the scenarios a bit um one of the open questions from last time was: why do some of the api's ignore non ASCII data and why do other api's treat non ASCII data as a failure condition, rather than just ignoring it and I I added I added some scenarios that I think kind of explained that in a bit more detail.

B

So, for example, when we first reviewed this, we had an equals ignore case method that takes a read, only span of bite on the left and a read only span and bite on the right and the original proposed behavior. There was for any a through Z character in left or right normalize that that way like a lowercase, a on the Left will match an uppercase a on the right and for non ASCII bytes just treat them as opaque.

B

So if you have a non ASCII byte whatever on the left, it must exactly match and on ascii by whatever on the right. So one of the changes that I made from last time is I realized that there are scenarios where we should have char overloads in this method, not just byte overloads. The reason for that is, if you're dealing with something where you already have processed the data as text. So you already have a string or a char array or something similar, but you still need to fulfill networking protocols like well.

B

Those networking protocols are still going to follow. Ascii casing semantics, so that means that the existing casing routines built into the framework such as ordinal, ignore case invariant culture ignore case and so on, are not sufficient for the purposes that you'll actually need. um So, if you scroll back up to the proposal, all up um I went and I added a bunch of char overloads where there were previously only byte, overloads and I. Think with the introduction of these overloads. That actually makes sense for the behavior of well.

B

If it's not ASCII I, don't know what it I, don't know what it means, but I'm going to assume it's correct, so I'm just going to say the thing on the left must exactly match the thing on the right, because maybe you have an accented character or something similar which shouldn't be treated as failure. It should just be ignored because if all you're doing is is casing well, I, don't care about it. I should just skip it and move on to the next piece of data. Does that make sense from a scenario perspective.

B

They realize how it was a big word, salad right there, honk read concrete example: you have a request line coming in over HTTP and you want to know if the first few characters are exactly the words get post put whatever in any casing. But you can't you don't care about anything after that, like the the rest of the request line can have non ASCII data, it shouldn't cause equal starts with whatever to fail.

A

All right so they're just dipping it because I needed to like remind myself what the difference is. So basically you just edit yeah. Basically what you added was this guy here right yeah and this guy right, yeah.

B

And for a lot of the other API is char. Overloads were added if they did exist as part of the original proposal.

A

Yeah, the only like that's and the only thing we reviewed was index off, and then we stopped so like everything after index off. We have interviewed anyway, so I croc here so, but.

H

First starts with like, if you, if you were looking for, starts with ignore case I, guess post. That would only look at the first four things. So why are we? Why would the presence of non 7 but characters influence it like if you find a non 7 bit character in the first four things, it's obviously false, but right.

B

Right and maybe maybe starts with, was a bad example, but it was. It was supposed to demonstrate that there are probably scenarios where, if all you're asking for is an ASCII case-insensitive, an ASCII case-insensitive check, I shouldn't assume that non ASCII data is invalid. I should assume that I don't know how to interpret it and that I should just skip it as part of performing the check and.

H

That is contrary to my expectations.

H

Why, if, if I'm using a 7-bit, ascii utility thing- and it finds something- that's not 7-bit ascii, then the question no longer makes sense. That would really be an exception.

B

So this behavior kind of already exists today in string because string can contain data, that's not actually valid utf-16 and you can still, you can still perform case queries and similar operations over it, and the behavior of all of those api's today is to just skip over that data, because it doesn't know how to process it. But.

H

B

Skip is then just compare for equality for the particular bytes that it doesn't for the particular elements that it doesn't know how to interpret skip is treat them as an ordinal comparison and then yeah. Okay,.

H

But orden, your ordinal ignore case is not going to say: you're sorry equals ignore case is gonna, say false for capital acts, entity and lower case x entity because you're not doing a bit casing.

B

Sorry say that again, capital.

H

Accented, your capital e, acute and lowercase e acute will.

B

Compare as not equal correct, because it's not asking yeah.

H

Well, sorry, let me let me back.

B

Up, let me back up, let me back up there. It depends on how its represented. There is a scenario where it will compare as equal. There is a scenario where it will compare as unequal.

B

It will compare as equal if you have if the first string contains a lowercase ASCII e, followed by a combining accent, character and the second string contains an uppercase ASCII e, followed by a combining accent, character. Those two strings will be considered equal ignore case. If the strings themselves are each one character and the one character is capital e, with an accent and the other character is lowercase e, with an accent, those strings will be considered. Unequal.

H

But again it feels I mean I. Guess I can understand, is what you're saying is you don't understand the combining character, but that's.

J

H

Even in the 8-bit range I mean I, guess everything with that in the 8-bit range makes or doesn't map between the ASCII and non ASCII world yeah yeah I! Guess it just. It feels really weird that it rubs me the wrong way that we would do anything other than throw for non 7-bit values, because I think that people already you're confused I mean if you get the mismatched surrogate thing, which is the only invalid utf-16 that I know of no like that is. That is the garbage in garbage out.

H

But the fact that you can sometimes get what looks like a acute case normalization and sometimes don't from a glyph perspective that that just boggles.

B

Me yeah, it's you know, there's already a situation today, which is unfortunate where I mentioned that there were two different representations of an accented Yi. If you have two different strings with the two different representations, they will already compare today as unequal. If you use the shrink not equals API, it's unfortunate that it's the way it is and maybe an argument. Maybe there is an argument to be made saying we shouldn't propagate that behavior here and I could see that argument. Yeah.

H

I think that's the one I'm making. If we're making a new class we should. We should give it the behavior we want, instead of the carrying the legacy that it would if it was appear to the existing API. But this is new.

B

Sure so you're saying that things like equals equals ignore case and others would still if they encounter any non ASCII data, even if the data, even if the unknown data is identical across both inputs, it would still report false. That's if.

H

I was writing this. That's what I would do, because it means you have given it an input that is outside the domain of the space that it is capable of answering.

G

Do we envision any scenarios where people are gonna migrate from the existing api's to the new one and expect similar behavior I.

B

Don't think I I, don't think people understand the full range of behavior that can take place today using the existing api's. So I don't think this would surprise people who otherwise weren't interested in the space, but we're not so we're not talking about throwing right correct these. None of these api's throws not a single one of them. Okay,.

H

I was but that that's what I wanted to do for 8-bit.

B

Data or well B, these api's are all designed to maybe fail. It is in return, false or return no match if given 8-bit data, but given that, given that these api's are designed to take untrusted input like I, really don't want them, throwing fair.

H

I guess I guess.

H

Yeah I guess saying that they're that it always goes into a state of non match if it finds an 8-bit character, is reasonable. So even if you gave it, if you gave it two identical, if you gave it the same string for left and right, but it contained an acute, a it's just gonna say false: it's not equal itself under ASCII semantics, which is a defensible position. Sure.

B

It mean it means that in that case we would not be able to have an early out check to say. Did you pass me the exact same reference for both and, if so, just returned, true yep, but sure.

H

But for the most imeem, the expectation I think is you're using it of you. You've read a line from an HTTP header, so you're not gonna, be passing in like you're gonna. Have your your string table existence of the word get versus the one that you just got so I, don't think the reference of quality. Optimization is really a thing that would be relevant and you can check reference equals yourself.

B

Before calling the method well, the idea of these API is is that we don't want people to have to do a precondition check before calling any of them. The API should just behave in a sane fashion, but.

H

If you want a ski equals and you give it an E and an acute E you're, not asking so notice on it, yeah, that's that's the because it gets really really frustrating when somebody posts issues of like hey I got string, equals returned false and I print the two strings and they look equal to me and so saying like if we're saying a ski that is data you see, is data that exists. There are no magic things as long as we're not backing up. When we see backspace current.

B

That makes sense to and to be perfectly honest I imagine most consumers passing a constant on one side of this and I would be shocked if, if there's, if they would pass non-ascii data in that constant I agree.

H

And I was even thinking of if it's gonna have this weird behavior of that, we would need to write an analyzer that says: hey you've passed in a literal that does not have a ski, but if it's just gonna return false, then now we could do it, but it's just like the it's just a like hey. You have an expensive return. False here, maybe maybe just say the word false.

G

Purely about ignore case, are you also suggesting that the equals should return false even for identical 8-bit values equals.

B

Would probably return false as well, because it's not asking equal and this isn't gonna slowest animal? Well, it means that we can't delegate to stand on sequence. Equal. It means that equals just like equals. Ignore case would do an ASCII Chuck on the data before it returns. True and that's worth it if you're, if we decide that the behavior is that it really is ASCII equal as opposed to sequence, equal then yeah, that's then that's the behavior of the API!

B

Well, I, guess I'm.

G

Questioning that decision is I mean.

H

If we want to say that all of the things that don't say ignore case are non semantics, ordinal then maybe and like we, it's a defensible position but I'd think that ignore case sometimes handling combining characters, because you gave it chars and and normalizing across casing, for what looks like the same glyph like that. One just really rubs me the wrong way, but if we think that there's enough argument that equals is, this is really just spam. That sequence equals then partially there's no. Why do we have it like? If you really want?

H

Are these the same byte sequences? There's a there's, a method for that? It's panned out sequence equals how.

G

H

G

Bite on the left and char on the right and.

H

We don't have one of those right.

B

H

B

It at that point, though: it's not sequence equal like if you have bite on one side and char on the other. We actually do return false if there is any non ASCII data anywhere there, because we don't know how to do the conversion. Otherwise, why?

B

Because if, if you have on the left and char on the right like I, we need to normalize both of those into the same representation, but ASCII can't represent data over 127. So if we see any data over 127 on the right like we have, we literally have no way of decoding it to bytes. So we have to veil that.

H

The 8-bit range isn't the same between utf-16 or between unicode and asking extended. Ascii doesn't map to unicode for the 7f to FF range I. Guess.

G

I'm saying, is this all extended ASCII.

B

No, it's all ASCII Oski 0 through 127, so.

G

Then, what's why couldn't it? Why couldn't it just literally be compare values 0 to 255 and anything else can't possibly much because.

H

They died, yeah 129 doesn't mean the same thing it when it's in byte. For me, when it's in chart form yeah.

B

Okay, if we had a latin one comparison, then we could do that, but this is asking.

G

So this is, this is using char, as you see too, and using bite as ASCII.

B

Yes, proxy 7 did ask yes: okay, hey char in the.net framework, as was a utf-16 code. You know, I was.

G

Interpreting this to me interpret char, just as a 16-bit integer interpret byte as just an 8-bit integer.

B

H

Of least surprise not principle of fastest code.

G

Well, for me, it's the type is named to ask you there for anything. I'm passing in should be interpreted as ASCII.

H

Right yeah and that's the point, is you can't there is no comparison between byte and charge if, as soon as.

G

I've exceeded a mere base.

H

I'm viewing char.

G

Here, as another word for short, are you.

B

Sure yeah yeah yeah, but the the thing is there's a passkey itself doesn't define a concept of a value greater than 127 like there's, no there's, no meaning behind it. So right, okay,.

G

I think I understand what you're saying I'm I don't understand why that necessitates, meaning that anything 2028 or greater means we ignore it. That seems like a policy decision we get to choose, but whatever I.

D

Think that there's a we need to understand. Are we saying that users should pass in both left and right saying these are Eskie? Should one of them always be ascii, or should we assume that neither of them like.

B

Both of them may use.

D

B

Common scenario, for this particular set API is, is going to be raw bytes from the wire on the left and a literal string on the right.

B

And the literal string, because it's a literal string implementing a network protocol is going to contain ASCII data. Well, if you, if you pass not ASCII data on the rights like you, have you as the caller have done something wrong so.

D

In that case, are we paying the price for some easily predicted branch to make sure that there are no no bytes or characters over 127 I mean.

H

That's the most common use case, but somebody taking a bite range and then walking over every string and a list and seeing if it call starts with, and they don't realize that they're gonna match on something non ASCII like that's, that's a very nonsensical behavior, and especially if this is designed for untrusted input. It means it needs to be guarded. So if you give it nonsense that either needs to throw or it needs to return fail, yeah.

B

If you like, if we want latin-1 behavior, we should have a separate lesson, one type that gives us a lot in one behavior and that's also in the comments here. I guess.

G

I'm, spelling like Jeremy, is the if one of those inputs is untrusted, but one of them is something that you supplied and you supplied it with a trusted input that contained values above 127.

G

It doesn't isn't the most likely case that you are expecting it to do a value comparison, a bitwise comparison, no.

H

You just you have a list of string and you wrote a you wrote a custom. Whatever you give to string dot index of the custom comparer that takes everything and just calls ASCII dot equals. You don't know. What's in your list,.

G

If you want to comparing.

H

To what comparing to the data that your gut that you just got from the network, so you've got your char on the right. You've got your byte on the left and your strings. Are you read some stuff from somewhere? Maybe it like it's just because it's data that you already have in your process doesn't mean it's trusted. Well,.

G

The scenario that Levi described, which is what we have everywhere, an HTTP client and asp net, is yeah. You have a set of headers that you know about sure.

H

And and I agree.

G

It's not the indicator.

D

It's a literal, it's.

H

It's sort of a waste to check by in the case where the the char is from some other input source. Like you, don't you don't you there's no way to verify that it's a ski I mean there's no good way for the caller to verify it's a ski before calling it there they're doing it on faith and if they get a misalignment between something they read in utf-16 and something they read in bites like that's a security bug.

G

Well, whether it returns true or false, it could still be a bug. That's why I wanted it to throw, but.

H

G

Could still be a bug sure, yes, again, principle.

H

Of least surprise, this says a ski: it should only do a ski to.

G

Me that's the most immersive prize. Anything actually, but good. Go ahead. Well,.

B

So what what api would you prefer then? Maybe maybe we could work it into this proposal or have another proposal for it? I'm.

G

Surprised that it's not behaving that that equals is not behaving like sequence, equals just with different sized data types. Does.

B

That mean that.

G

This needs to be a different name to convey the different behavior. Maybe oh, well, I I'm, not sure it's just that's. What's surprising me.

G

D

I think Jeremy makes a good point if these are not designed for internal use. High performance, we know exactly the scenario it'll be used or then we should probably handle out of the ski range as an error or return false.

B

Yeah so I had mentioned passing literal strings being what I perceived to be the more common scenario. I, don't think! That's the only scenario, as evidenced by the fact that there are span a P eyes on this as well I.

B

There are scenarios where I think you already have a stream, that's hydrated into your process like a username, for instance, and you want to check that against data coming from the lock right.

B

um Okay, that's uh that's something that I can that I can take for myself and like if, if, if the name equals is, is throwing people off as to what the expected behavior here is then I can I can take that off line and figure out if there's a more apt method, name that would convey the actual behavior.

G

B

G

You can also offline. Do a survey I'm the only one and feel free to ignore me it just to me it that is immediately what I I take this to mean. Is that I'm, just looking at two values that happen to be different size than interval types yeah the fact that there's been a subsequent check to actually make sure that every value, regardless of integral size, is within a specific range, my mind didn't: go there I mean.

B

I could make the left as fight instead of item and I would do the same thing. Make.

H

It uncommon anything yeah yeah,.

B

Okay, no! That's that's good feedback! Thank you for that.

B

um Index of is index of sure, I don't have what is it? So if so, it sounds like index set up just like equals. Wood on wood then do the return false. If either, if the one on the right contains don't asking.

H

B

H

B

Which one's the other one which one's the haystack the left is the haystack and the right is a needle yeah.

H

So if the needles not ASCII, then you can't find it and I with my expectation is this is greedy from the left and if it finds it and then that's followed by not a ski date, non seven bit data that you return to the successful match, because you didn't necessarily look ahead. I'm, okay, with also different negative one, but.

B

Correct this, this does not look. Past were the first successful naturist yeah.

H

But as long as it's a ski up to that point, that I would say it would find it it last and that's of assuming we have that I would say it's greedy from the right yeah I didn't write the last one that's off here, but they do. They do exist for the Holland and cuz yeah, so that if, if the string or real expand of char contains non seven, but they then you clearly can't find it. I mean like that.

H

That's an early optimization of your needle is invalid, but like there's just literally not projected sequence of bytes to find sure.

D

So we're talking about just checking the the char version, if it's not SD, but what about the bike version as well? My input could be that one and I might be comparing it against utf-8 things would appear equal and perhaps they we're not yeah.

B

And that's that was an open question here, like if you, if you passed non ASCII in hooked, I assumed that I was supposed to treat it as okay binary data. But you raise a really good point. This, it's almost certainly more likely, but the color did something wrong yeah.

B

So that should be an indication that we should. We should treat it as in the stuff. That would be my preference, okay.

A

So just to clarify so.

B

Same behavior, equal them off.

A

Index off so Jeremy I think you said: when the needle contains non ASCII characters, we would expect to return minus one I'm a little bit unsure like what does it mean for if the haystack contains non ASCII characters? Does it matter like if this matches anywhere or would you say it returns only successful match if until the matches from non-ascii here, because we're not encountered that.

H

That I mean that like either either is acceptable, but I am certainly willing to believe that what we're doing is a linear validating scan from the left or the right, depending on if it's index ever last index of and that you you ignore data beyond the match, yeah I would.

B

A

Non-Ascii character, you just stops the scan, or would you just say you just keep scanning until you find a match a.

B

Letter I would scan until I found a match, but.

H

That means that left of your return.

B

Value you could have non ASCII correct, but if you call index of you don't care about. What's on the left of your return value generally, but.

H

Are you validating that with to the right is asking there you go validating that the needle was was ASCII and you found a match only.

B

Validating that the needles asking you found a match, because when you, when you eventually slice out the needle and then use the remaining api's to to inspect the rest of the buffer, eventually one of those AP eyes will tell you something's wrong.

A

Well, I would say I kind of agree with Levi, because if you have this rule of like.

J

You're, the left.

A

Order to the right like that means, we are now stuck with a particular implementation of the algorithm that is forced to go from left to right, rather than some paralyzed Cindy stuff or something else right. It seems well.

H

Yeah I can I can buy that argument. As long as we validated that the needle is ASCII, we found the bite pattern.

K

H

Maybe that's the start of your string. Maybe it's the end of your string you're, the only one who knows, but you know you now know where the needle is I, can accept that sure.

B

We might actually be able to take advantage of the fact that the data is validated to be asking, because that means that we might be able to start smuggling data through the the high bit of whatever. So the operations we use so I don't expect that I don't expect that it will be a performance hit to have these checks. In fact, it might actually be at her from provement to have these shucks, but I would need to mole in that a bit more yeah.

H

And the string, the overload of index of that takes string, you could use the never actually returns. True is ASCII property.

B

Yeah, okay, all.

A

Right so then, so we talked about index off the very cover, get index off first and ask you but I.

B

Don't we did not index herbs where we stopped all.

A

Right then, I think that's looking through your comment going that.

B

B

So this API I actually use all over the place internally, I the reason that I use it all over the place internally is it's used heavily as part of like utf-8 utf-16, counting because there are optimizations in the transcoding pounds that that try to count how many bytes or how many characters would not change, and this is a workhorse method to do that. Counting I, don't know why anybody aside for me would ever need to call it, but I do call it all over the place.

B

So I thought that someone else might need to as well.

D

So this is just for optimization purposes where you want to skip all the ascii, get to something that you actually have to handle and yes, yeah.

B

So trent transcoding utf-8, for instance, um it gets there are parts of it that can't easily be cindy optimized and those parts are the parts that deal with non ASCII data. So this implementation is a cindy check to skip over all of the data that can be optimized and then quickly get to the data that requires manual processing, at least that's how I use it internally again, I don't really have a scenario for making this public, but I figured that, since it has so many call sites internally, it might be more broadly necessary by people.

H

Yeah I mean I think even the the like Jason stuff, because it's trying to stay super low-level was like. Has this as a private implementation already so I think it's fine.

A

So I just noticed that you have string overloads for some of them, but not all of them is that intentional? Oh.

B

It was just convenient.

A

Yeah, because some of them, you only have we'd, only spend off char right so like should we add, string overloads for those, so you can call them from other language. They don't even do the conversion. Where do we not care? um We.

B

We could but I I only add so at the moment I only put in string overloads where I anticipated that somebody might actually pass in a literal we could add string overloads more generally if we wanted to but I. Imagine, while you're walking through this API review. Imagine that if you see string, one of the key scenarios is passing in a literal house.

A

B

Automation interpreter and.

A

Little was equally convertible to read, only spend off char than on the true votes right. So how did that make a difference?.

B

B

It's generally easier to reason about an API that does that, at least in my mind, so.

A

I think adding overloads for string mix and I'm trying to understand is why some have them and some do not. It.

B

Was just a marker basically to say I expect someone to pass in a literal here. I think we can. We can remove them. We can propagate them more widely. Yeah.

H

B

Don't really have a strong preference to be honest,.

H

I mean the the guidance in the effectively being. If you expect someone from let's just say VB to to call it, then you should overload between read-only span and read-only char, but if you think that it's only going to be called by things that are already doing a bunch of low-level work with spans, then you don't need the string based overload.

H

So if there's no scenario for someone who has a system that string to ask when did I see non ASCII data because they should probably be working with spans, then it doesn't need the overload, but the the equals one seems common, etc. Yeah.

B

We could I don't have a strong preference on whether we add them like if, if best practice is to add them, you should add them.

A

Yeah my argument I mean my concern is less about whether we add them or not. It's just. We should be consistent. It seems odd to have them in some places and not others like either we make a decision as Jeremy said that we don't expect we'd, be people or other people from other languages to ever call them, and then they don't exist anywhere, or we just add them everywhere, but, like I, think we should make up one line which one it is.

H

The only the only cost of the one that takes string is you have to decide what you want to do with all.

B

Maybe now that would be an argument exception at that point, right.

H

But you sort of have to look at it individually of. Is it normalizing to empty or is it or is it an ane and either is fine? You just need to reason about it for the API and.

A

The implicit conversion do when you convert is another string to we'd, only spend off char much.

H

To my dismay, it turns into the empty spam yeah.

D

A

Maybe that should be the behavior that we have for empty string. Otherwise, maybe if you freeze always.

B

H

Yeah I mean the empty string becomes the empty span that the null string becomes the empty span, but that that creates. Sometimes it creates problems and definitely it creates mild inconsistencies. When we already had the string version first and it was already throwing in A&E, so yeah.

A

That's that's kind of why I'm leaning toward making it an empty string in general, because otherwise you can never add an oval of the TIC string, but you can. But it's weird right like well.

H

You wouldn't have been able to pass the word null before.

B

That you could, in the past a non literal string that evaluated to null. Yes, all.

A

Right, another cost is trained right, yeah.

H

All right yeah, so it's just like if we, if you think that it looks like an argument exception, then ane is appropriate in context and if it feels like, like you, wouldn't I guess, I don't even know what the what index of does. If you give it the empty needle like return, zero, yeah and then it's the question. Is that what you want for the null needle but.

B

We could also say that this is a specialized API designed for people who just want to parse values coming from the wire. At that point, it's at that point like you really should be in the span world anywhere right and that's the I think.

H

He most true or nothing but I'm willing to say that method group levels of if we think some of them are gonna, be common to call from general code. But then we would really be adding an ID array overloads as well and yeah. So it's probably just.

B

Okay, so let's let's leave off string for now, and we can always add it later if we needed to with the caveat that nobody have to normalize to empty I'm.

A

Okay with you're not doing string at all, because my literal should already work at least from the languages we care about, and you know it's not that this thing does anything that we can't do already it's just. That's that's meant to be a low-level API, so we can always add support later.

A

Alright, then I think so we were still so you just covered get index or first non-ascii by it. If you already cover, is all ascii know what I guess that that's the same thing right except it compares the resultant -1. I suppose.

B

But it's it's a behavioral equivalent to calling the method and comparing it against -1. Yes, the implementation is free to be different, but behaviorally. It's equivalent.

A

Makes sense to me.

A

Yeah, they all make sense to me anybody else. Any concerns also, what is a ski buyer makes also sense. I.

H

Mean on that one I had the mild: do we need the type suffix but I don't care.

D

It's their reason we wouldn't just have is a ski taking a spin by insurer I said so. It is all.

B

A ski and is a ski binding is ask each are look at the same method, name yeah,.

A

Yeah as logically, you check the entire put for being extradited.

A

Yeah I think I, like that. Yeah.

H

We can reason about, it is essentially the single will input or the the single inputs are helping you when you have the list of length one. Yes,.

B

I did struggle as to whether to even have the elemental overloads there and the the reason that I eventually included them is because there is still, in some circles, confusion over what the exact definition of a ski is. Does it include the eighth bit? Does it include the value 127 and I figured having these elemental api's says? Look the rest of the api's on this type are opinionated as to what exactly an ASCII value is.

B

So this is a way to get the exact same opinionated check without you having to reason about it inside your own code, yep.

D

H

Yeah, so yeah I think we can read it and then Malta is asking is all a ski is all a ski? Is a ski bite in his escuchar I think they can all be overloads.

B

If is asking for in general, should we even have the bite in char suffixes on some of these methods, so going back to the get index of API play? Should we drop bite and char just call it value I.

A

Would not even call Beatty all right, I would just say well.

H

Ending with an honest, is weird yeah I guess that's true here can.

A

Be say character here: where do I get shot down by the Unicode police? You.

B

Could you can call it character if you want to like us.

I

B

Worried I think.

A

Barry just come in the chair. Well.

H

I think, even using all the various interpretations of Unicode definition of character as long as you're limiting yourself to 7-bit ascii, it all maps unless they exclude control, characters.

B

H

Centers the c0 controls.

B

Do fall under ASCII, yes, I.

H

I would leave those as they are I, don't see a reason to mean, like yeah I mean changing it to elements so that we have a noun there just makes it longer with no real game. Okay,.

A

Yeah invasion seems like we try too hard to make it the same thing right I mean yeah.

K

Okay, Levi for is ASCII. You said a lot of this has to be quite opinionated. Do we have do we have to define those same opinions for room, dot is ASCII and it's.

B

The it's the same, okay, cool thanks! No thanks for bringing that up, though,.

H

So I'm, assuming we're moving on with to lower invariant you don't have crossways, is that intentional.

B

I have by Corey Corey. Were you actually pushing for cross voice? Yes,.

D

We use that in HTTP, okay,.

H

So you also want byte to char and char to bite.

D

At least charter bite I, don't think that we use bite, tutor yeah. We.

B

Should we should have three- or we should add the fourth one? Yes ob, so two notes are first off we should probably drop invariance from the name here, just because we dropped ordinal from the other api's. Okay.

B

Second, if we, if we're doing crossways, we now have to figure out what we do when we see data that can't be converted.

B

Does that throw it? Does it become a try, Muffet bring got to lower and every other to lower so.

D

Some of those samples that I was able to find for that in our HDTV code and we have both just truncating to bite whatever number it gets wins and we also have throwing scenarios so.

B

Nice, so you can. You can always go from bite to char. Like that, that's fine! You can't always go well. You can't that's a problem because you're you're now changing the representation of it and you don't know what the representation means as soon as you leave the.

H

Seven bit range: it's there. There is your nonsense as far as this class is concerned. Well,.

B

It's specifically it's the combination of leaving the southern bit range and doing a transcoding operation at the same time, if you, if you do only one or only the other, you don't have the concern as soon as you do. Both you have a problem. Well,.

D

Can we be consistent across all of these api's and I would rather not have some of them that return false if it sees an esky and other ones that just ignore non-ascii or things like that I feel like they should all error out. If they see not asking that.

B

That's not the behavior, though, of any other text. Processing API. We have anywhere in the framework yeah.

J

Like there, there.

B

Should be some consistency there and when, when you have an API like this, which is going element by element by element and processing the element generally, the easiest thing to do is to skip things that you don't know about.

H

And I: don't like it: okay,.

B

D

Mean if we're saying that this should skip, and that raises the question, should all of the other ones also skip they.

B

Do except for the ones where they have to make a yes or no answer based on the element value, all the other ones do skip more by other ones. You mean here down because.

H

Well, no, like it.

B

Index on, for instance, we said index of skips non ASCII data. It doesn't fail if not ASCII,.

I

Data appears in the haystack.

B

B

And this would be similar, it would skip non-ascii data, assuming it doesn't have to transcode transcoding with frozen wrench into it.

B

That's only when you start having crossways, though mm-hmm. If you don't have crossways than you can literally just take the input that you don't know and copy it to the output, because you're not changing the representation.

B

A

So should we just if we sorry go ahead, should we just try to just pool what we currently have and then move on, because the problem is, you only have 15 minutes left and there's like six more issues that we need to unlock people on yeah, yeah.

B

A

Because I mean, if you want to do crosswise, we can still design that bed, then maybe, let's do that outside this meeting. Sure.

B

Sounds good but.

A

You suggested that dropping in variant from the two lower ones right and yeah.

B

Because we we dropped ordinal from equals, so we may as well drop in variant here and just document on the class and not a single one of these api's is culture aware yeah.

H

Ten minutes, there's no ASCII cultures, yeah yeah.

A

That makes sense for me.

H

Yeah the destination being too short returning negative one sounds weird.

B

That matches the pattern we already have on things like spanned all to upper.

H

Okay, I, don't think it matches what we approved for buffer guidance, but.

A

What would you do instead? Well.

H

I mean you know as a precondition that the size won't change. So if you pass destination smaller than source throw like you should have dealt with that, it's.

B

Not you have to be the right thing that absolutely is a precondition here we will. We will write exactly as many elements as we read yeah most.

B

H

So I would say if source that length is greater than destination, that winks throw argument, exception, destination, yeah.

B

So we would have the caller, so we would have the caller check that yep. We would so, even if we had that behavior, we would probably still want to return an inch, because returning an inch makes it easy to chain these yeah. Damn.

H

I, like returning the interest, is goodness I'm, just saying the what we had for simple buffer writing methods was if destination is too small, that's an exception.

B

Okay, so so we wouldn't follow the behavior we already have on things like memory extensions. We would say these it's up to the caller, to check this condition and then handle it appropriately and.

H

We wrote a bunch of stuff prototyping, effectively prototyping spam through the VCL and then shipped it, and then we wrote the guide and said what we wish we had done so: okay, okay, that's that's fine!.

A

Alright, so do I have to scroll I guess: I do yeah.

B

And then their elemental to upper and lower, oh and in place to upper into lower Christophe, was actually asking for the interplay streams, or maybe we just don't leave these. Maybe we just leave these methods off entirely and say the existing to lower to upper can handle in place.

A

I'll put the handle of that because you can pass in the same spend on both sides. Yes,.

H

Yeah I: are you.

B

In their mind it, it just means that we would have to complicate the logic of the other methods in order to check if their sources and destinations overlap, which I was not planning on doing but existing in for this methods don't operate on bite correct. There are no existing in place puppets.

D

And by that I mean calling to upper with source and destination equal to each other. You correct they were Qatar.

B

D

You can't do that. It throws oh yeah.

C

B

Great because I think it I think it was intended to work, and then we made it not yeah.

B

So maybe we say then, that the the original methods don't handle overlap in the end place ones. Do the in place once clearly can't do cross conversion? You.

L

Need right, damn.

H

Fine with having the emplacements okay should.

A

Be but do is there value and having a different method group for that, or should it just be all over loads.

H

I mean it seems weird to me if one method in the method group is non I, guess there it's booting into this writing to a destination, but one of them has effectively mutate the input and the other one is read from input and write to output, and that feels different to me for reasons. I can't explain.

A

Yeah to me it's that kind of, like I, see I, hear what you're hearing but I hear what you're saying. At the same time, it seems to me like it's, but it's very cleanly over though the barity right, maybe but I'm. You.

H

Know we don't often overlooked by arity, because frequently the arity is just complexity and options. Sure.

D

Yeah, so we have an existing problem with string where new users will call just to upper thinking that it's mutating the string and you're a little bit like yeah and I, like that in place kind of tells people that, yes, this does mutate your buffer. It doesn't return.

J

B

Yeah, okay, so leave in place them.

H

Yeah I think it adds value to be a different method: European, okay,.

A

All right that works for me then get hash code you'd.

B

Hit hash code yeah because you might need to shove these into a dictionary and we need the ability for you to be able to calculate a hash code. Based on the data. I mean you, you can't take spans and shove them into a dictionary, but these the spans that you're providing us might be part of a larger buffer. Where you are managing the dictionary keys yourself. You'll have to provide a compared to that dictionary, but how.

F

B

You build such a compare these api's. Let you build a compare works.

A

I mean the only thing I would suggest is probably remove Auto neligan and just go. There get hash code in all case, yeah.

B

I messed up sorry.

G

Where else do we have such static? Get get hash code helpers I'm wondering if it should ever just be be a compare so.

B

On string, we have get hash code as a static method. Oh the the problem with making a compare is we're now opinionated as to what the input T's should be on the comparer and I I.

B

Don't necessarily know what actual data type you're going to use taken to represent your data, you're, probably going to use a byte array or a memory, a bite or something like that. So now do we say we have a compare of memory of bite and if your data is represented in any other fashion, sorry.

G

Okay, what about do we have get hash code? Ordinal, ignore case methods, I thought for like on string, I thought it took a string comparison as an argument. It does yes.

B

The the reason that I didn't take a string comparison in any of these api's is because the only two values it would ever be valid are ordinal and ordinal ignore case, and they behave differently. Then string docket, hash code, ordinal an ordinal ignore case, behave differently on a string get hash code. Ordinal ignore case allows a ski and non ASCII characters to transfer between each other, whereas this would not.

B

It would actually produce a different value.

G

So you're saying this, this will produce an identical value with string tight like the soft get hash code. Overloads will produce identical values with.

B

String Beckett, add code correct, but ordinal ignore case will produce potentially a different value than string docket hash code, doppelbock ordinal ignore case because it's doing a different normalization under the covers. Okay,.

G

For what it's worth I, don't think it matters whether it's that's coming from a string comparison argument or ordinal it or case in the name. If we're concerned about that difference, we should be concerned about it either way. Okay, with.

H

The string comparison, the problem is like four of the options, don't make sense and we either mapping them down to the two or we're throwing. And then you know the similar to advice you gave about default up default arguments versus overloads of like there's no branch. Now I mean the branch was in the color code and it's actually not a branch. It's you you're, calling the ignore case, one or not so.

G

I'm not suggesting they be default, I'm suggesting they call it get hash code in the ordinance.

B

So we have good hash code and then take a string comparison, yeah.

G

Same for methods just renamed again as per normal, ignore case and how to string comparison to argument I guess it's really.

K

G

The well hold it before that. Don't make sense, they really like there's nothing you can do for them or what I can't do anything for them. Why I.

H

G

I guess you could say.

H

That alter is the ASCII. Culture is non ignore case and so you're just mapping for things to to, but that's a boolean that be boolean or case.

G

But but that doesn't mean like, are you saying you know if someone passed in buried culture ignore case you're, saying there's literally no way we can.

B

Produce the right answer: yeah, it implies that there's going to be some kind of culture, sensitivity taking place which we can't provide in these api's.

B

We can't provide yeah, so your if you have so say that say that you pass in something that is cultural, where in varying culture, current culture or something like that, the data that you're passing in the the actual text that you're passing in we would have to somehow convert that to utf-16 before we can before we can call the get hash code routine on it, and if the data that you're passing in is not ASCII data, we have no way of tell us transforming it to utf-16 before we perform that comparison. Sorry,.

G

If you're saying that these are happy to accept non ASCII values,.

H

G

Have to be yes.

H

Yeah there's because there's no sensible fail.

B

Mechanism they would have. So that's also one of the reasons that we might want to reconsider the equals method, not allowing non ASCII data right, you can see the hashcode and equals should behave identically. That's.

G

Not true, thank you.

H

If equals is true, get hashcode has to be the same result, but if get hashcode is the same result, equals does.

B

Not have to be true, it doesn't have to be, but one of one of the behaviors of thinking of this get hashcode routine is that it is collision resistant. If it's trivial to well, actually would that work I take out that never mind yeah.

H

Equals should still be defensive, but get hashcode, there's no way that you like returning negative one, because there was an invalid thing that that is itself a dictionary attack. So it has to do something. Yeah.

B

No you're right, you're right anyway,.

G

B

Saying Steve, sorry well,.

G

Maybe that's if we have to do something on some of them. It seems like a good reason to do something on all of them. It starts to feel very arbitrary when we're doing something versus not the.

H

Comparing things is equal is, it has to be ASCII and the things that return numbers, which is get hash code return, numbers I, know.

G

That's your opinion. Yeah.

B

So for things that do case comparison or for things that you have code calculations generally, what they do is they're iterating element by element and then mapping one element value to another value. The way that the mapping works, at least in my mind, is if you don't know how to map the value potentially because it's outside of the SP range, like it maps to itself and then move on to the next piece of data.

B

So that's why I was saying that to upper to lower, ignore, not ASCII data get hash code would ignore it as well, and by ignore I mean it treats it as ordinal and then it just moves on.

H

B

H

Work if we allow it, if you will allow someone to pass in the word invariant, because that has different, meaning I guess.

G

And I'm still so two to lower invariant. We can make work well.

B

It, let's drop the word invariants, because all that means is we any byte that it sees this upper case. A through Z gets turned into lowercase A through Z every other byte stage, the exact same.

G

Okay, so what do I mean? Sorry, okay, so why wouldn't the same semantics apply and get hash code if I pass in ignore case invariant.

H

But we're getting rid of the word invariant in the method name. So it's just two lower okay.

B

Are you, are you passing in a string comparison or a pool ignore case you pass so I'm trying I'm.

G

Trying I'm trying to like it feels very weird to me that we've got this one off where we're taking a value from the string comparison and putting it into the name.

B

Which is why we should, which is why we should remove ordinal from the name, probably and just call it ignore case, because it doesn't directly map it doesn't actually directly map to any of the string. Comparison values it maps to its own thing, I mean.

A

It's not really the same with the relative of equal tonight. We have equals in because any more case, but.

B

To your question, though, Steve, if, if we did try to take in a string comparison, then presumably any culture-specific string comparison value that you passed in, we would need to honor in a culture specific way, because these trim comparison, enum says I. Want you to interpret this as if it were in whatever culture I specified. At that point, we now need to unambiguously transcode from ASCII to utf-16, which means this API would have to fail if it sees non ASCII data, and we can't have a get hashcode routine fail in by fail.

B

I mean throw an exception.

H

Yeah, but if we feel that there's too many method groups we could make it be a billion ignore case, but watch I would be totally fine with.

G

And I don't have a strong opinion on a boolean ignore case versus a court case in the name.

B

But yeah there there's also string comparison hasn't been used in any other API here, specifically because it none of these API is matte cleanly to the existing string. Comparison values, I used ordinal, ignore case in the method names, but it's it really doesn't map to string comparison. Ordinal ignore case. That was a mistake on land.

G

To step back for a second, is there a higher level gap here? Should there be a string comparison ASCII?

G

J

G

Here that we're trying to encode in general, that does isn't representable with the values we have today. Does that mean we're missing a value.

B

Possibly I, so we can't add values to string comparison like we would rate be too many. Things have weeded out.

B

Sorry so string, compare so string. Comparison ASCII ignore case is what you're proposing right. I'm.

G

Asking I'm not proposing I'm, saying like her, you you've outlined like a bazillion api's here that are all exposing a semantics that you said doesn't exist on string comparison, which makes me question whether there should be that semantics on string comparison.

G

H

Feels to me like that would be the I mean I wasn't on the team. You know in 1999, but it feels like dotnet string was really trying to be Unicode and expressing character data so saying that you want to do an ASCII comparison on a system. String is weird, but if you say that you are limiting yourself to the ASCII space and you want to do a comparison that that feels like you've accepted the limitations now. So it feels like this relationship is right and putting it on the string. Comparison is awkward, I.

B

Could I could actually see that justification like I? Could I could see an ASCII ignore case value working on shrink comparison like it would? It would be weird in the sense that it wouldn't have and it it wouldn't have a case, sensitive sibling, to live on the same type and it does kind of go against the whole strings. Are you cs2 data, which is what Java and.net both did, but as far as the actual behavior goes, we could make it work if such a thing existed, I guess.

H

We're just changing what the case transform is during normalization. Yes, that's the.

B

Only thing it would do thank.

H

You hey yeah, I, guess I guess it makes sense as a as a comparison. Yeah.

B

That's literally, all I would do it. Would it would change the case normalization so that anything outside of a through Z gets ignored? Honestly.

H

It's what everybody writes if they try writing their own to upper anyway, so yeah.

B

This is basically the two upper you would have written in CS, 101, yeah.

B

B

But the the scenario for these particular API so was again that uh you might be storing. You might be storing some larger piece of data into a dictionary or some other collection, and you need to be able to create a comparer. Somehow now it's one of Steve's suggestions was. Maybe we should just make a compare, and that would be fine in my opinion, but we don't have a comparative standard. A and I.

H

Don't know how to make.

B

H

When we get all in addition to this, we could add the string string, compare accelerators for a ski and a ski ignore case, but we're just asking ignore case, but because yeah absolutely but yeah I I agree with the notion that we've that the span things and the compares don't play well together, do.

G

We have any I asked earlier about get hash code overloads. Do we have any get hash code methods anywhere that takes bands today, UTI.

B

Yes, string has good hash code but expand as a static method. Utf-8 string has them as a static method as well, and what? What kind of comparison.

G

To those use a.

B

Full string comparison, but it's they can actually handle all six cases easily and and unambiguously so.

G

In what situation would you choose to use this versus that with ordinal with.

B

Ordinal, sorry, could you repeat I've you you cut out. Yes,.

G

Sorry, the hash code overload here that takes just a buffer in what situation when I. First, this string cat hash code with ASP and level you have the buffer. You.

H

Haven't read only span of bite instead of reeling span of char. Let's.

G

Say I have the chart? Is it like what.

B

Is it is this this get hash code? Routine is not necessarily guaranteed to be the same as the existing string, docket hash code routine. It doesn't have an algorithm, it's it doesn't have.

G

B

G

It did that would just be an implementation detail. Okay,.

B

G

That, aside, like, why would I choose to use one over the other.

B

Just the same reason that you're using equals, instead of span dot sequence, equal for bite and char and stuff, like that, like you you're already using this class for text processing and it's just the methods there, it's convenient, you want to call it because it's the easiest thing for you to do. I think.

H

It would come down to the question of the guarantee if we're, if we say that get hash code read only span of bite, does not guarantee that it would produce the same answer: a strain decade, hash code for the same semantic content. Then you need the charm, and so you could balance.

H

Because you, the only that would be the only way that you could compare a slice of a string to your ASCII buffer is, if you, these two have to give the same answer for the same semantic input, but they don't necessarily have to match the the one for string unless we say that they do have to match, in which case we can eliminate the charm I think saying they.

B

Have to match ties our hands. They.

G

Have to each other? Yes, they.

B

What do you mean they have to match each other? These.

G

Charm, which, by topic you might have to provide the same result for the same inputs.

B

Yes, correct that was correct and.

G

The bottom two ones have to provide the same results for the same inputs assuming.

B

That the inputs are ASCII Austin in all cases. Yes,.

G

Even if they're not asking if.

B

They're not ASCII than the places where they differ could affect the hash code.

G

But like at night in, if all the values in the read only span of char were 255 or less you're, saying those still wouldn't produce the same thing, they may not produce the same results. Correct.

B

Well, it's what whether it produces the same result is not a guarantee. That's the thing, because it's not as detail.

G

This asking the end of hey this feels that feels wrong, but I guess it comes back to the same thing as my equals earlier. Yeah.

F

Just a general clarification question on Knowles I'm assuming.

J

Any embedded or.

F

Trailing all's repeated as a billet element and not ignored or treated as terminators. It's a correctable.

B

Little relevance key.

H

But I think is your question: Steve: are they treated as string terminators or they treated as content? That I think they're treated as content, because you gave us an explicit link.

F

Yeah, yes, my question I guess there are three possibilities: one is krita's content, they're just ignored or treated as a small Terminator and I mean I. Guess if you have a buffer, you know that's populated with zeros, for example, and you try to compare against a literal. You'd have to use index so instead of equals. You know that type of scenario.

H

Yeah I assume that essentially this is that there be serves of not know s Z's, that's my guess. Levi, oh.

B

H

That again does does a null be produce the same hashcode as a B or the same hashcode as a or a unique hash code. It produces the.

B

Same hash code as a MLB like an anal, a mole B by itself is a sequence like Rex.

H

Equivalent is different.

B

H

And it's different than a be correct.

B

Okay, yeah a shrink string, pocket hash code by the way can, depending on your platform, treat all three the same: yeah fun, I.

H

Haven't open at all three are the same: a and a B are different.

H

G

Or for culture sensitive, not, yes,.

B

Right, but if you had in very if you called in Burien culture, then, depending on your platform, all three could return. The same thing. Sure yeah.

H

See also the SA yeah.

B

Which is one of the reasons we don't do any culture specific stuff here whatsoever.

H

Alright, so did we decide, are we doing bullying's? Are we doing to method groups.

D

What context you lost me hash.

H

Codes, the last thing we talked about I'm trying.

B

To move us on so it equals this to method group like all of the other. Things are to method groups like it. Hash code probably should be as well just for just for symmetry yeah. That's my tiebreaking position as well.

H

B

We're almost done email almost almost on writing in there well I'm.

A

Okay, should we go over a widened to utf-8 16.

B

H

That's the next one yeah.

B

Widen the narrow, so on the utf-8 class we actually dropped Weidman arrow as prefixes I, just forgot about that for this proposal and we literally called it to utf-16 from utf-16.

H

Presumably you want a lowercase T and usually have 16 on narrow, but yeah.

B

B

This doesn't perform substitution if it sees any non ASCII data, it returns, operation, status, error or invalid data. It's.

H

B

Operation status, API and they're.

H

Called read and consumed I can only see up to by thread and chars V. So it's.

B

Bytes red chars rotten, but we would we would use whatever sorry.

H

Yes, it's bytes consumed, so red should be consumed. Is our our naming guidelines for operation status.

B

Bytes consume chars written is that right, yeah well.

H

The thing is consumed in this case for the first one there would be bytes consumed and chars written and the second would be chars consumed in bytes written really that's what Christophe wanted.

B

You're right: okay, no, no you're right! We just uh I guess we just followed the pattern inconsistently before we had proper guidance. Okay, that's fine! It's consumed so.

A

It's by it's consumed in chars.

B

Britain yeah yeah um any problem with the operation status. Api is I mean there. I would think these would be the most straightforward ones. Yeah.

H

Than the other than the names of the of red to consumed, looks yeah.

H

Yeah Moe doesn't look happy.

A

No I'm, just friend consented, makes sense.

M

Did it ever return needs more data? No.

H

Because the data unit is one unit.

M

H

M

The one day you take a real nice pair of charged input, correct.

B

It can, it can never return, needs more data. Thank ya.

D

Surrogate pair would be a error in that case right, yes, valid data.

M

And presumably narrow owner might I was thinking like that men believe you returned destiny too small, but that couldn't happen anytime, yeah.

H

Yeah to string seems reasonable.

B

So to string I wonder if we should actually just kill that and say you know what you have in Cody, not ask you just call encoding ASCII get string calling today D.

H

B

Let me just double check and make sure that actually does exist. It does get string, read-only expand, byte, yeah, yeah, just call encoding, not asking and it'll. It also allows you to customize what happens if it sees not ASCII data yeah.

H

So since you're, the one who proposed that trim type method should return range, why are you returning expands there.

B

Is a get trimmed range at the bottom, the trim that returns families just for convenience, but we could. We could kill all of those so emo. If you scroll down just like one line, you'll see it yeah.

H

The team's control bar is blocking the methods I want to see. Oh sorry, yeah.

B

Sorry, for that, that's all good, so trim works. The exact same a string trim only. It only looks at ASCII whitespace characters. It doesn't look at higher Unicode whitespace characters. It will ignore them if it sees them right.

H

So, with get trimmed range you can implement all three of these correct have.

B

Absolutely everything can be built on top of yet trend range, yeah.

D

B

Only downside is that get trimmed range is going to inspect both sides, so it might be doing more work than necessary. Good yeah.

H

Yeah I mean if we have. We think that the others are gonna, be common enough to really want span to span I mean we can leave them as accelerators, but I would suggest cutting it to just the one operation. If we think it's sufficient sure.

B

And and I did look at uses of string, dot trim throughout the framework trim start and tremendous Tring or called very rarely full-on trim is called very commonly yeah. So maybe that is a good argument to kill all six of those are religion or.

D

We could make those six overloads return range if we think there is a valid optimization there.

A

D

Salami name, get trimmed range is weird, because the trimmed range would be the whitespace right. I.

B

Guess the range true: this returns arrangement results after trimming. Your.

H

Head which really means this trim? Yes, so we could leave the three method groups, so leave the six methods and make them return range or we can eliminate start and end and return the range, because you could just take the lower or upper bound yourself.

H

But the start and end will half the work it does BAM so based on the I, would suggest, eliminating span span and make it span range and then get trimmed range goes away because, as Corey pointed out it, the name suggests it's returning the opposite of what it's returning. So okay.

B

So the trim methods all return range rather than spans.

H

That is that was out your gardens. That's my recommendation. Based on your suggested guidance. Yes,.

B

That that does mean that these four methods have an API signature different than all the other trim methods in the framework. But it does make them somewhat more powerful and.

H

It solves our six way overload for me, I mean we.

B

Would so how many overloads will be end up with in the end just six or we.

H

Would have the six because you're varying on byte and char, so.

B

H

Each method group, but then we don't need we don't need, read only span, writable span, read-only memory and writable memory and then to decide if we want to do something special with string as well. Okay, so.

A

Okay, so let me summarize this so what what's the so first of all Institute's jinhwan. So this is ching when we decided to draw because there's not overload that we can get already.

H

Encoding that ascii can provide that encoding.

A

H

You that get string of the buffer all.

A

Right that one and then for the trim one. What was the conclusion on that? One change.

H

The return type of all six trim methods to range and delete the two methods that are, after them all.

A

H

A

Makes sense in this.

B

Yeah I I do get a little nervous just because the pattern is different than every other trim method, but maybe that's okay. Well,.

D

The other trim methods on string return a new string, so it's.

B

Memory, extensions from.

H

Methods yeah, but that's where we learned that we were having the the Oh God I gave you an array and you gave me a span and what do.

F

I do with that yeah.

H

B

Okay, at least this solves all of that at once, yeah I.

H

B

It makes a call site a little bit weirder, but at least it's doable indeed.

H

It's just it solves our our implicit and our implicit conversion to reduce the number of methods we need is still good, and then we don't have the problem of you can't figure out how to get back to your original data type.

M

Are the performance characteristics of two string and encoding or a Skeeter guessing identical.

B

They're about identical they're within the range of noise of each other, okay.

F

Can the trim map that has just returned the same buffer? No trimming occurs. Yes,.

H

Well, I mean they're gonna return range now, so it would be start to end yeah.

B

They'll return the entire set of data by the way I think we would probably want to call these value instead of buffer I notice. After all of the other API is here sure, yeah.

B

Sorry, which one is that on trim, we would want the inputs to be called value instead of buffer same with github.

A

B

Just miss out what I've worked on this originally and then to lower in place, probably as well right or, let me say, I I would actually keep that as buffer, because it's both the source and the destination value implies something that's only being read. I guess that means is asking should be.

N

A

Yes, you sometimes you refer to it. As text, though, should be able to change text and to mail you as well, no.

H

It's just an index of yeah.

B

Text in value on index I believe is correct. Okay,.

H

B

H

I mean I know that it doesn't match our naming guidelines, but honestly I find needle in haystack much.

B

More clear, we call it span in value on memory, extensions whatever on.

H

Memory extensions isn't span this parameter, yeah.

B

Yeah and these are intentionally not extension methods, because we don't want extension, but that's hanging off a very close fanta bite. No, that would be.

B

B

Thank you for letting me monopolize your climbing on all good, so.

A

Lets me just pull the remainder.

A

All right so then it just pays all of the stuff and now.

A

In the hopes that I didn't get anything thanks.

B

Everyone for slogging, through that with me and Steve I will I will get. I will ponder what you had said about about naming and latin-1, but not ASCII data and see. If maybe we can come up with a more concrete behavior there. That would make more sense, Thanks yep.

A

Sweet all right so.

A

This is mine. Hopefully this is an easy one, I hope so too. So, basically the gist here is you just want to expose an API that gives you the rate that we're currently running, on which we do not have there's no easy way to do this today,.

I

We have this method, but it's not like an official system. Api yeah.

A

It's in this mecca so don't have second abstractions runtime environment, which medicine is gonna, kill the whole thing right, merge it basically what's useful in there with the rest right.

I

That's kind of the proposal is to get rid of platform abstractions the library and anything that's necessary in it. Move into the quote: unquote shared framework, the only thing we've, the only thing we've discovered that is absolutely necessary is the give me the current rate of the of the machine. There are two other api's that aren't that aren't exposed elsewhere, but they aren't we kind of don't think people should be using them. That's like they get.

I

The name of the OS, which is like, will return a boon to Santos, etc, etc, and then give me the version of the OS, which would return like sixteen point: oh four, eight, etc. How.

G

Does the? How is the current method you're talking about? How is it implemented.

I

Currently in platform, well so there's two raid calculations in our in our world Long's written in C++ code and one's earned in c-sharp code, and they both have the same logic, which is if you're on Windows. It starts it with win and then gets the current. You know asked Windows what version it is and then slaps on the version number if it's on Apple it just slaps in OS, X and then gets the version number and then for everything else.

I

It goes and reads: etsy slash, release, OS parses, that out to see what's the what's, the distro and what's the version so.

G

It it won't actually be the actual read it might be, pretty close to it, but it's there cases were the actual read and what this returns are different in the current one. I realized the new ones, gonna be more accurate. Note.

I

Well, so both of those both both implementations, the C++ one in this and the c-sharp one- have the same logic, and so in the cases where we have special cases like red hat or Alpine, where we don't want the full version like we don't want Red Hat 7.3, we just wanted to say, read at 7, right or maybe I might have been doing patch versus minor. It might be 7.3, not 7.3.

G

That's an example of what I'm talking about like it sounds like with the current implementation that ends up producing a different value than what we would want it to produce with.

I

The current implementation, we have the logic twice, and so, if we update one, we have to go update, the other does that make sense. I.

G

Guess: I'm not following so today when I access, this runtime environment, I get runtime identifiers. Are you saying it's special cases? Rell? Yes,.

I

Ok, both both in the host right in the c++ host.net, dot, exe and because that's the thing that actually gets used at runtime to pick assets, okay for portable apps, and so this is an implementation detail. But the way this the managed api is gonna be written is basically there's only gonna, be one implementation and that's going to be the host and the host is gonna pass through a new app context. Variable for the current read: okay,.

I

Which a point of discussion is, if you're, using a different native host, who didn't pass in that app context? Variable this. This API is gonna return null because that app context variables not there.

G

So this is minor but null or empty.

I

Probably whatever app context returns it when you ask for I, guess it's try get switch, so it's gonna be false. Yeah.

G

I'm just asking, because you know, we've started using knowable reference type annotations on stuff, so is questioning whether this proposed API is actually a string question market-rate.

I

It should be string, question mark I think it should return null if, if that's not there, so I can update that test. String. Question mark.

I

It is, oh sorry, go ahead down, I was.

J

Gonna say if this is going to be used for like switching based on what the current rate is. Is there any need for us to not only return the current runtime identifier, but also the the generic runtime identifier or possibly have a method which says? Is this read compatible with the current read because, like that, like will return wind 10, X 64, but people might want to care like when X 64? Instead, if they're like trying to resolve an asset from a native library or something yep,.

I

So this proposal, like lower in in this whole issue it talks about those scenarios and I've scoped them out of this initial removal, because we don't have those api's today. Well, let me back up. We do have those api's today by somebody would have to go load the deps json file, that's in the shared framework, which has the full rid graph for the current machine. So that way you can tell when 10 X 64 is compatible with Wayne X 64.

J

But that's probably a future API. It sounds like right.

I

It's not part of this proposal this this proposal. The reason this one's code uncle blocking is because I want to get rid of platform abstractions, but we need this new API, because we need this new API in the official place where it's gonna go before. We can delete the old one because, like things like, the SDK relies on this API and.

H

Platform extractions does not currently define the is rid compatible, correct.

I

Okay, correct it.

J

I

Get current read and then give me the OS name and OS version and then other junk like what's the what's, the architecture which we already have API is for those.

F

So Eric is this gonna be um do exactly what it says here in the proposed API for the implementation and just look at the that context. Variable that's.

I

The the current proposal is that, like I put in there about the discussion, should we maintain a managed fallback yeah Jung Eun's point yawns opinion was to not do that.

A

I mean it seems, like the host has to know that anyways right, and so we have to have it in native. Then it seems better to put it in the managed right. So it's easier to just pass data from the host to the financial environment right.

I

I mean my only issue is: is the other host right, like people who are native hosts that are loading up the CLR and then calling into it? Anybody who calls this API is gonna, get null.

A

How many do we have today, where we had custom holders, there's a common scenario as a more like Cookie Monster area,.

I

I'm, it's I, don't know, I'd assume it's. It would be growing, for example like if Visual Studio ever wanted to be on that neck or they would be, they would have their own custom hosts. They wouldn't have. They wouldn't use dotnet Exe well,.

J

They wouldn't use dot exe, but presumably they would use our official hosting api's and presumably, as part of that, we would support setting these common app contacts variables right, I.

I

Think that would be up to the host team.

J

That seems like something that we'd at least want to talk to them about, especially as the number of these app context variables grows both now and in the future. We're not going to want every single host to have to redo placate. All that logic. True.

A

Yeah but it seems like a code factoring h2 more than anything, there's nothing fundamentally problematic of sharing their code enforce and make it available as an API.

J

Right and and that's what I meant like it, there should be some way for them to trivially like move to the dotnet 5 runtime host api's, and be able to support this without having to like manually copy logic out of our host or anything like that agree.

I

So then, as long as other hosts either use that functionality as long as they use that functionality, whether it's on by default, somehow or just a simple method, call that says, add runtime identifier, app context yeah they should just be able to get it for free is.

A

That Justin, who owns that.

I

I

We've we've changed recently, like I. Think in three: oh, we changed. What's our recommended host, api's I could look, it might just be on I'd have to look, it might just be on by default. Steve barter. Do you happen to know, since you have some.

F

Okay, sorry, can you repeat it with.

I

The new hosting api's in do these app context. Variables automatically get put in cuz I know like I know the new ones aren't like Ike or CLR hosts they're, like you load up the host policy, I.

F

Don't know I mean I have to look at the code: okay, Russell's guys, I'm.

J

Assuming they need.

F

Some other context variables that we flow to the runtime. However, so I'm assuming they're set up properly, would okay.

I

I can look I just and.

F

You and also just a quick question here: um I, don't know about Linux processes, but I, don't know if we should say it is that there's a guarantee that is always gonna return, the same value if somebody messes with with this environment variable at runtime and do we cache it? For example,.

I

So the host, when it loads it reads that environment variable, if it's there or not, and then it doesn't like it, the host won't pick new assets. If you change that environment variable wallets running right,.

F

No I, just I just meant for callers. If they call this, um you know and if somebody wanted to mess it up and you know try to set at runtime change its environment variable um which I don't think you can do in Windows, but I don't know about other operating systems that it could return different results in some weird. You know attack that case or something the value, and then you know get it once and return it great.

I

The managed code won't respect the environment variable at all. The managed code just says whatever the whole sent to me on the app context is what I'm gonna return.

F

I was thinking we looked at the environment, verbal okay,.

I

No, you can set switches an app context, yeah I.

H

Mean it might be reasonable to to cache it to a static on the type anyway, just because then it avoids going to the dictionary a second time great.

I

Should we change this from a method to a property, then.

I

A

Okay with that I mean it seems like it's a really. You know one access, yeah.

H

Every everything else on runtime information is a property actually, so runtime identifier seems like it would be a reasonable property as well. Okay, that.

I

Returns string question mark.

H

A

Okay, that seems said: okay.

I

Well, that's what we returned when the app context variable isn't there I know.

A

But honestly, there's one of those areas that we just said that's about as the host like this seems pretty fundamentally flawed, yeah.

B

If I were calling this API I would I would be surprised if it didn't return a default value but applicate the string unknown.

I

Which is like the root of all raids.

B

That's reasonable, actually I, guess I'm, just having trouble wrapping my mind around the concept that the runtime itself, as opposed to a library, might not have an identifier yeah.

H

I, like that was discussed, it's the if custom hosts but yeah I think it actually would be reasonable. The first time it's accessed that it's ASCAP context, question mark question: mark the string any because that matches our read semantics is.

D

There could an application using this API do something valid if it returned null.

D

Yeah I mean: should we just it's not specified in a context which, because I'm trying to think of what, in what a user could actually do with null the.

J

Same thing, they do today, which is fallback to things like runtime information, OS and other things, and try to manually determine a read I honestly.

A

Say that we should do this in the runtime either the host automatically said read so the eff context thing is never empty or we should just have. You know the fallback mechanism baked into the thing like I. Don't think we want people to write code like if now then do crazy gymnastics, but what would they do with any crazy.

H

Dude nastic, or they.

K

I

H

Ask is: do you have an asset, that's good for any, which would just be standard, not platform-specific, io, I, I.

J

Think, rather than any, it would be much better for us to have a simple managed fallback that says: if your windows return wind, you know x64, x86 or arm or arm 64, and if your Linux of any kind return Linux whatever and same for OSX yeah,.

B

That is an application developer. Tanner's proposal would surprise me the least.

I

But then is like new platforms come online right, like iOS Android.

A

I would say our hosting API should make sure that we do not do that right. So this would only carry the case where somebody has a crazy host, and you know they. If they don't do something sensible, then too bad right.

H

Like essentially the returning any should be a none hit path, because all hosts should do it, but it does. It does mean we can take the question mark out of the API signature, but having the managed fallback means. We just have a lot of unused code, which seems wrong well,.

J

It wouldn't necessarily be unused. It would if nothing else, give people who were on like 3.1 something they could copy and reuse.

J

A

J

To be LTS, anyways.

A

Right but I mean like it seems to me bad that people copy and paste that stuff right, because I mean the only case we would have this likely is you put. The new platform shows up right, so any hard-coded single, it's probably not work there, either yeah yeah.

D

I think I would prefer. We either do some of the gymnastics ourselves as a fallback, or we just throw saying the host did something terribly wrong. We.

N

Don't like throwing out a property gets, though.

A

Yeah I mean honestly at that point, make it I, don't know it's fairly early in the run time just crash the process early on, but initialize correctly. This.

B

Would be that's misconfigured, though so I I think that would be okay for Charlie, fair.

H

B

Alright, as long as it's.

H

B

That it through so.

H

I think the general consensus is we don't like return null, so we can get rid of the question mark and then the the details on that or either any throw or a fall back, which is not exactly relevant when we're technically over time. Yeah.

A

So far, I got three thumbs up, just give it a quick vote and then we can go the other issues, because you have three more issues. That thing will probably take hasn't had 20 minutes to finish the other ones. Sure.

A

D

A

I guess that makes it easy alright. So that means therefore, are they close this one.

A

All right, David you're up.

L

So I would like to start saying that I would like to defer the review of this API I mean at least the final, the final design for a subsequent meeting, but mostly because I checked this API to reflect the changes that Wisconsin was previewed on on the latest AP review about this, which was to address changing from a class to any new, which totally makes sense when you had two options which were default and preserved, but for fun in a funny model after you play today after you bladed the description community member came and said that he wanted to use another property.

L

That is part of the Asian. The network is not correct, it is, it is not already exposed to our API, so I would prefer to change this into a class again and maybe maybe convey that with another functionality that might justify. Please be in a class so tank we can review it. Does that make sense.

A

H

A

H

If there's a scenario where it makes sense to allow the reference handling to be a user implemented type, which of course means you have to design all of the inheritance models of it and and have a sensible plug-in system. If that's part of.

L

The feature, then that's fine, sure well, I, was not thinking actually on on I was thinking more than a shield class and making making maybe more customizations just just added to that type as a property and but I guess that should also fall into the interesting area to review later. Yeah.

H

Because then I mean the thing to keep in mind is, if you end up with like two boolean properties: that's just for flag or for values. You know, and so it's it's about what what makes the most sense to as as a call mode but I am but I think we can easily defer like if you weren't.

I

Here when I talked.

D

L

A

Then we need to follow up then for this one, or is it just good, as this.

A

It seemed like you wanted to follow up on this one.

L

Well, we already discussed we array, others that on the pure CPI, so I guess there's nothing yeah I did there is no need for follow-up. Unless someone has some concern with a with the final implementation.

O

A

Seems like you said they wanted to look at the something at the beginning, but maybe I misunderstood what you said because I mean this looks reasonable to me. Faced some with this man, yeah.

L

But yeah, but someone came up with any with any scenario, about adding our reference resolver, for example, which is something that we currently don't know expose we implemented internally, but it's not part of there. I mean we don't expose it anywhere. So it will be nice to turn this into a class and add that and expose that into them into their reference handler type right right.

M

So I think like save it, we probably still want to review that new design right so once you're ready with the new design, let's bring it back and we can look at it right. The class worried about the pay yeah whether.

H

The extensibility is in a type called reference: resolver, the extensibility Xand, a type called reference handler like it sounds like you're wanting to design an extensibility mechanism into this, and we can't, if you, if you have a scenario for it and you design it for it, then that's fine. We can review it, as is I.

J

H

Only moved you to the enum in the previous one because it was described as being closed and not really growing again. Yeah.

F

This is a scenario somebody wants to implement the equals.

L

Method, no, they want so they want to implement the their own reference resolver in order to keep track of references and keep us and keep the state and save the state across multiple civilizations. So they just want to make the dictionaries static instead of non static and.

M

They want to use goods rather than oh yeah IDs, as their ID.

H

Yeah so I think if the question that you were coming with David is: would it be okay to go back to a class if there's a reason sure, but we we can't really talk about it without a concrete proposal. Good.

L

A

Thank you all right, I just wanted back to API needs, we're gonna, do comment and then yeah, that's totally reasonable. I mean input, as Jeremy said when originally we just want you to know because it seemed like over-engineered. But if you have a concrete scenario for that, let's review that and then just have that extension part. That seems also cool all right. So then, two more and the copy constructor to Jason see other options.

L

This okay, so this one right now is Pinet needs he needs to copy. It needs to create a copy based on based on their Jason's utilized adoptions, just to set one property, which is the encoder, so they they can eat. They can set the relaxing coding on their other options when the user does not pass its own encoding right. So right now this they need right now to implement it, a method to copy all the problem of the options that they want to into the new instance of the Eastern civilization options.

L

But such thing needs to be maintained every time we add something to the JSON serialize directions, so, for example, for for then for the for the newest previous that we expose reference handling the reference cally feature on the options they they're, our customers, their users on asp, net and embassies that are not able to use reference handling because they code on asp.net is not copying this new property on the option. So, basically, what we want to, what we want to bring here is to create is to is for us to provide dncls adoptions copy constructor.

L

So they know they don't need to maintain their own logic, to copy their the option and.

H

This is going to flush the like it's not gonna copy. The type cache and it'll go back into an editable, no great yeah.

L

No I I think Steve a comment regarding that, but I think we shouldn't copy the the cache yeah.

H

Because this is.

B

H

Yeah because I can see this is a reasonable scenario. You have some options. You've already see realized with it. You want to tweak it to one thing, but the the type is now locked because you see realized something so.

F

As long as it's.

H

Ignoring the cache and unlocking itself, it makes sense to me yeah in.

F

Addition to the caches you can add converters on before you before you, sir lies. You can Adam at runtime through the converters ahead method, and that requires a little special logic, but I I talk about that later. In.

H

F

Or that issue I'm not sure which right but.

H

Effectively, it's copying all of the public state, not.

F

The state- and it goes back into writable.

H

F

H

I mean that makes total sense to me.

A

We have preferences whether we like a copy, constructor or method on the type that just gives you back a new instance with all the settings copy copy constructors.

B

Tend to be more a more common pattern, I think.

I

Like this, isn't a sealed class says seal it seal. Well, there. It is it's not right. Next to class, as.

H

Partial has to go before the word class, or are I like, after top in my head, I can't think of examples, one way or the other, but since it's a sealed type, I think the copy constructor doesn't have inheritance, complications.

A

Yeah I mean even if it were to inherit it, you would just change it right. People would have to opt into this, and then it would actually be somewhat nicer to copy constructor rather than a clone method. Player.

H

It's slightly weird, if you say new Jason's, to realize your options of derived Jason to realize our options, but that's you chose to be type limiting in that case, yeah.

B

H

B

Things I'm thinking is that if you, if you knew up a collection like a list of tea, a dictionary of tea stuff like that, you can pass in to those constructors another list and other dictionary things like that, like we do have precedent in the framework for copy constructors as opposed to copy methods. That's.

A

Fair I mean yeah, I mean it certainly is easier for other things right, because you can just again just miss connection, there's other syntax. If you do a constructor and yeah make sense like this was just.

M

Wondering what people.

A

M

What are we doing about the convertors list? Steve you mentioned that we should also copy the convertors list. Why.

F

Otherwise, if we're not cloning, the converters that were added before the first serialize but.

M

That converting this grows and changes right.

F

So I proposed only adding the converters that the user originally added before the first search as soon as you do, the first serializer. Do you realize we we, you know.

J

F

It immutable and don't allow you to add anymore converters so but.

M

Since we are unlocking, it I think it'll be surprising to think that the previous and waters they were really added, get back, get it again, because now this could be different right. It's.

F

The same it's exactly the same, except the converters that will be added later at runtime um could be different based if somebody decides to add some more converters. I mean converters are part of the options right at least the ones added before you start. Cyril are seen when you start sterilizing, you get, you might get more converters added. Those are derived from the ones that were previously added so.

A

Maybe I'm missing something, but are you saying that we, like the system itself, adds converters doing serialization I'm.

F

Like a factory converter um returns.

N

Another converter, which is cached, that's all.

F

A

See so basically, but you see your basis suggesting that when in the copy constructor wants the only thing that copies, but the user added to the original j6l as options- yes, I, think that seems reasonable, because otherwise it would be very weird if you would only copy half the object right. It.

M

A

Be weird: if it's missing it.

M

Might be weird, though right like I, can see why your Steve's point that, yes, we should add the converters that the user explicitly added themselves, but it's also weird to say, high speed, a copy constructor not to undo all the additions, the converters that someone has had it by adding my own or somehow are you moving them? It would be strange.

B

To me, if the auctions class reflected data that took place at runtime when the serializer was already running generally, options are how you create the serializer they're, not supposed.

D

B

A state on the serializer instance itself, I.

A

Think the question that Allison is raising is, should we copy any converters and I would argue, would be very surprising to me if we didn't, because if I want to clone- or you know, run a copy, constructor I would expect the entire state that I put in there to what to be reflect to give the new object. If.

N

A

This, yes, you have to undo them, I mean, but that's not different from any other setting on that object. Right.

F

Encoding, the encoder property is another one. You know which you can.

B

F

The encoder we copy that specify converter, we copy that yeah.

H

I would expect everything that they has already been set via public API is reflected in the in the clone, and if that that includes a list of converters, and so it should be. Essentially you do some stuff you, you run it through the copy, constructor and now one of them you set Pam, okay, start acing to camelcase and the other one. You said to a snake case, and now they are identical, except for casing: anyways Steve.

B

Can you clarify something for me so when, when the when the serializer is already running, is the actual converter list, as reflected in the public API being mutated, or is it just an implementation detail that stuff is cached.

F

F

Only the cash is mutated at run time, so I think I might have missing heard that need to be cleaned up when I when I said, we need to enumerate that list only up to the point before it occurred. So.

B

What I would I create the options, class and I populate the converters list? I use the serializer a little bit and then I queried a converters list again. The list that I get back is going to be the exact same as illicit went. Incorrect no entries were added, removed or mutated. Yes,.

F

Yes, you try to call the add method and that's not allowed, and the underlying list that we returned is never it's not. We use a cash, you know dictionary and not. Okay, yeah I was just looking for clarification on that.

B

F

That yeah, like I, said I, think my comment needs to I need to delete a sentence in my comment. If you're looking at that, I'm not sure but yeah.

B

Yeah, that's what I was basically my question on yeah.

M

Okay, that make sense I just want to bring it up, because I understand that in either way someone could be confused, but I buy the argument that, having everything mean cockpit surface area makes more sense. It's cool because.

A

Of the only copy part, then you have to explain what part and why everything seems very arbitrary.

M

To the only and only knit I would have is became the parameter name in the constructor to be options, not source.

H

Yeah and I didn't I didn't have a strong opinion like an option, seems better than source yeah.

L

What what name, though, or a copy constructor.

A

Well, the it depends on what you have, but if extension method said, we don't really have the one parameter name for extension that sense or copy constructors, because it kind of depends on what you input so I. Usually we name them after a good time, all right.

M

So I added you for that could be, let's say the list of T constructor, that takes an ienumerable of T and that thing has the parameter being collection. Yes, it well.

A

That often is called source like it's also similar to, because that's just how I knew mobiles are often named right. It's not necessarily because it's copy constructor.

M

Well, yeah: no, the parameters for election not source there.

H

Is also not list of T, it's like election.

H

It's not exactly a copy constructor, it just functions as one.

H

A

H

I, just can't think of any I.

A

Mean you can't do right, I mean you can totally pass in a list of T to a list of T. It's just happened to be typed as I collection, because that makes more sense, but effectively it operates as as the copy constructor right right.

A

All right so then, that's just close this guy here and move on to the last one, which might be easy or hard.

O

You see so so right now we have a way to ignore no values which is globally applied. That means across the entire object crafter. Well, she realizes not sure, isn't every property. So we can, you know, ignore no values in the adjacency realized.

J

O

No, it's boolean option, but there's no way to correctly include only subsets to be associative properties to be or in or exclude them so based on the value of ignore no options so.

J

O

That what to be true, there's no way to include some properties which are no and I know in the opposite case, if it no no values is false and there's no way to exclude some of the properties from from that room, and that applies to both serialization and deserialization. So this is about providing a way to property, ignore no values, so there is no net, has no value handling which is applied on, which is in a global option on the adjacency relation settings and also no value handling in owned as applied per property.

O

So the semantics seem to introduce by this issue, give us some parity. So the reason why I want to do this. This is a popular request by customers and also by yeah, and a lot of people need it.

O

So the proposal appear here is to have a condition for which each property is ignored, so the three values are always when no and never so one thing to keep in mind is the backlog for 500 includes a proposal for default value, ignore so giving indicating some default value for like a struct or even if they valid reference type see. This is default value and then ignore or ignore that so a future extension point for this for five most likely dependent on all the other features would be to had ID.

O

You know a when T fourth property in non-volatile, so a signal condition and then well. This would be to be a new condition: property on Jason, ignite reboot, which would be default to which default value will be just an ignored condition always which gives us parity with the current default behavior. So if you don't specify that condition, then it if also always- which is just slapping on to Jason, ignore as we do today- and them is always ignore this, regardless of what value it is in the Jason or what CL our value on serialization.

O

Okay. So so that's the DPI included, if you like usage scenarios where I always like I said I always always be just like. You have just an ignorant thing else, so one no considering two properties where we both say that this should be ignored. Why? No so during the Jason we have a value for the class property. So we do not. So we just realize it because it's not it's not no yep, and then we said the value can set that value set. Then we leave dictionary because we said to ignore it: one.

O

No so notice we didn't have to specify global option here to say you know no values so yep turning on series, so we said to ignore it window. Then.

O

So the value of simple class in this example yeah. So yes, the way we renew your object again and the class values no and then so we skipped on sterilization basically. So the other interesting scenario here is when we have never so I mean as any questions so far about to behavior. If.

A

So, logically, the global settings that we have various is the default right. So if you have adjacent ignore property, so it should be present and always take precedence, no matter what to global setting it's correct. Yes, okay, that makes sense man just.

F

For clarification, the attribute is already existing, we're just.

O

F

O

This actually already exists today, so they never allows us to say, regardless of what the the global option, the you know, no valid option on GCSE layer. The option is never act like ignore this value. So so, if you have decent, so if you having no no values true and then the property, you have the condition, never ignore it, then it will be T C less than C earliest. So the example shows the same. So here we have the first property which doesn't have the we doesn't have any. There is an acknowledged with condition.

O

So when that is no, we do not see realize we do not dis realize it, because we're not trying to digitize no not into that. But you know the next property has that it should always be deserialize decirles. Then we set that to no so, and the same goes for serialization so realize the other value class values now and then dictionaries. No, we realize it because what auto is realize it so a few considerations here so no cannot to value types right now. This DC liza through is just an exception. Even when justin ignore value.

O

No no values applied, so we have an open issue to relax this behavior and when you know just like don't attempt to disarray or series well it if I lit up can happily Nolan serious without that issues only on this realization, so we fix that and we circle back and then make sure that we ignore no values for value types on the property pieces as well and then so you're having lasting already mentioned. This is the concision to ignore default values and sterilization and serialization.

O

So well-liked literati. It's no default values: option on desensitized options, then for pirates here, because people likely need its Tsukuba can add a when default property I mean in on condition on this in ignore condition.

O

So yeah, that's about it for this proposal. Are they any questions or.

H

When the Novick.

O

H

Seems to me the same as if I give a string to a value type of it makes no sense, so it should fail, but that's a different conversation and I think this works. Fine. It is, should.

M

We add ignore default now, like I, don't see why we should buy. We will wait for that. It makes sense because.

H

If we do it and don't have the other feature, then it sounds like it won't write down a zero for an int, well yeah. We have to.

O

To start respecting some attribute to tell us the default value likely to default value attribute, and then it just seems yeah. So, based on these points, we haven't done the two Club all behavior. Yet so we should wait. So we do that because there are some open questions. I think around comparing the default values for structs as far as performance goes like um so yeah. Those I think if default value ignored, needs to be like.

O

First of all, it's more a backlog issue and feature at these points, and then it requires some extra thinking so I think that should be done separately and then we added when we we didn't. We Magister issue all.

M

Right, that's fine as long as you're, confident that the enum and current solution is accessible to that feature. Yes,.

O

We can easily add that the other thing is if people want to have a window lock default. I think is reasonable to add that as a an enum option, rather than do like some type of flag solution where we use, we.

E

Made in a man fly.

O

Fly but they want to do some like validation, to make sure you don't do like never and always so. I think 24th and when no default could be potential things that we could add new feature dependent, 2014 more likely, but when all the default dependent on, if people are asking for that.

M

So maybe this is a general pattern for enums we're. Only a couple of them can have boolean logic attached to them that we can keep them as separate enums rather than having the in a become a flag. Is that how we generally do things? The guidance.

H

For flags is, you should never have a combination of bits. That's invalid.

M

Just name it correctly: yeah yeah.

O

Yep, that's it for this issue.

A

All right so like what's the conclusion on the ignore defaults, a separate.

O

One it's just a bit when, when I, when we go and look at two default value handling ignoring default values, then in that proposal we have to include another in on value, 444, 24, 24, so future, so not including that now.

F

That's a larger feature, and it has to also deal with like whether you want to write out null. Some people might want I mean right out defaults. Some people might want to always read out defaults and so yeah splitting them up into two separate future requests makes sense to me. Yeah.

H

I assume this is only here to show that you've thought about it for for potential upcoming features and that you believe it is sensible. Yes, that's right. Yeah.

K

M

Is awesome and I like it, nice cowl.

H

If it makes a sand happy, we should ship. We.

B

Should I agree with us thank.

A

You alright, then this is it for today. Thank you, everybody for staying a lot of a big number and then I see you later at some point. I.

D

Think you might.

B