.Net Foundation Design Reviews, 26 Feb 2019

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: .NET Design Review: GitHub Quick Reviews

Description

00:00:00 - Approved: Obsolete string.Copy(string) method https://github.com/dotnet/corefx/issues/33602#issuecomment-467551787
00:11:51 - Approved: Expose existing BitOps methods https://github.com/dotnet/corefx/issues/35419#issuecomment-467557134
00:26:31 - Approved: Consider removing Range.OffsetAndLength https://github.com/dotnet/corefx/issues/35508
00:31:46 - Approved: Remove some unnecessary HWIntrinsic API overloads https://github.com/dotnet/corefx/issues/35560
00:55:42 - Approved: Add Rune.TryEncodeToUtf8Bytes https://github.com/dotnet/corefx/issues/35530#issuecomment-467570023
01:01:14 - Approved: EnvelopedCms encryption with RSA padding modes https://github.com/dotnet/corefx/issues/34366
01:09:17 - Needs Work: UTF-8 web encoders https://github.com/dotnet/corefx/issues/34830#issuecomment-467582155
01:33:11 - Rejected: Unsafe.Unbox should return "ref readonly T", not "ref T" https://github.com/dotnet/corefx/issues/34258#issuecomment-467591284

A

Right now, we're lives.

B

C

It's always a gamble.

C

All right, so this one leave I found a while ago and I think we could not convince ourselves what to do with that.

C

Yeah I think these are the only two things that Jesus was 2018.

D

Three lenses exit to the point now.

E

F

Terrible, it's.

E

F

Three years they're, like oh, my god, hello, we.

D

F

Everything that's from 2018, so we're context, shrink, I'll, coffee right now is guaranteed to give you back a brand new instance of the string so that you can mutate it affected it, but mutated string instances is going to run afoul of a bunch of optimizations that we want to do in the 3x 4x, whatever time frame, which means that if you're calling this method and mutating the string instance, you are almost certainly going to run into problems once those new optimizations come online.

F

So the idea is, if we obsolete this method now we can start weaning people off of it and on to things that are more supportive for their scenarios. Right.

D

What would that be by the way? What would what be but we'll they were placed in the beef Odinson area? So if.

F

You're doing peon vote, there are AP eyes on the Marshall class that will give you back mutable buffers that you can pass into the piano player. If you were calling this because you intended to culture and I'll clone, you can call string clone or not, because it doesn't do anything if you want to mutate or string. If you want to duplicate a string and mutate like one very particular value in it, you can do you can use any of the factories on string that give you back the stand before the string itself is baked.

F

You can then go touch. The span change the very particular element you want and then return back to the color and so on. So there are options, depending on your exact scenario, particularly message on this. Basically something short you mentally.

A

Yeah Liz, what could save you should probably add enough W link at the end of something, and then you can put fun to whatever you want to point it to it, because I almost would have expected this message to say stringbuilder rather than if.

E

Links to like Marshall or yeah.

F

Yeah yeah, my to be honest, I kind of saw this going to in FW 1 or something no doubt, I think the reason the people use this and then use unsafe code to mutate things in places that they don't want to go through the in direction of a strength only because that is the more intuitive option right right here we have the miss fanny.

F

They they want the thing equal to the same length and passive input rating and hold it modified. If it's in that.

C

E

Probably the more forward.

C

One right all right so then I think I have no problems of that. Like I spent some time. Thinking about oxidation, general I think these kind of issues we have a single API. This is fine.

C

We have nothing else, might not to do it in and make the furnishing so I'm inclined to take it, as is any concerns wops, leading that everywhere else. No, what do you mean everywhere else? So I don't want to just do the core like if help Street the method that well Peters in the standard and then backs into the matter there. Everybody.

F

Oh yeah: well, that's why it's kind of core standard yeah yeah did.

E

We ever get the response from doesn't we did from Andy on if you can downgrade maps and obviously their work, nor it they.

A

Said the answer is no or yes. What does it change? Your assessment for this issue.

F

If it's warning, you can drag my warning more right, Hado or.

G

E

And one of them looks very good yeah one of one of them lets users work around it. If there, you say on a framework that does support, mutating strings or they're, not quite ready to make the big switch. The other one makes it so that users upgrading have to make the change I, don't think, there's a big difference and I think users should be changing it to be just work and not mute. But it's that question of like how hard is the hurdle once you do? Try not great yes,.

C

I mean you might not nobody suggest I mean like this is the rest. Maybe we should make it an arrow eventually, but right now it would not make it an arrow anyway. Directly here I mean I would never introduce an obsolete as arrow unless the API absolutely doesn't work at all ever mixed case. That seems fine, but I mean this API works right.

C

So it's just yeah, you shouldn't do it, in which case I think the warning is fine in general and then maybe in a four or just yank it out, but I would not hold my thread from that. One.

H

You know sorry I got kind of an interview and a cat came late. Was the question answered about when we choose to use absolute versus the the analyzer tooling yeah.

C

So that so basically what I said earlier is that I don't have a problem with making individual api's like this obsolete, because the analyzer would basically do the same thing right. The the big difference between the analyzer and this approach is that analyzer will give you the option to, let so, let's say, there's a feature you want to absolute in the feature is spanning multiple different methods or types or properties.

C

Right then, you can basically obsolete them all and they're all putting attributes and on every single one of them, and then you have a single ID that you can suppress.

C

So the problem of observation wants a you know: ban like twelve API is that you it print for every single one of these twelve api's x, all call sides, and that means you can at that point only disable of solution entirely for your project or pragma, suppress or call sides, which is super annoying and biltema that you can just suppress the one ID problems particular feature that was obsoleted. In this case, though, given that the feature and the api is one and one, it would basically the exact same thing.

C

Sure we use this API, like in hundreds of places, the the the the analyzer would still be better because you can just suppress it globally for this one ID versus in this case, you still have to suppress it for all call sites, but I think this API here is not very likely to be used by. You know across a wide variety of call sides right, so I think it's fine.

C

The biggest problem of the analyzer approach right knows that it's not baked in the sense that you have to install it right like it's not reference by default, we don't, we haven't productized it yet, so we should do it because it keeps coming up, but I think the problem is that neither are small. The language team is actually actively investing in this, and so oxidation is the only thing that is completely built in and we just mark it and then everybody else gets the gets. The warnings.

C

Does it make sense if it.

H

Does it raises that I think this is the first time and all these API review meetings that were obsoleting, something yep, and so it does raise the question of you.

G

Know this sort of opens the door to our being open to it. So I think in general, like I I'm, open.

C

To it so long, you know it's not it's not a disaster right so like there's, another API I think we I think we never review the OP solution, but I think we we ended up doing it this. This was the was that WebSocket dot is application targeting for fire by which was still in infrastructure. Api was never meant to be public and, like those kind of instances, I think observation is completely fine, because you just you know it's one API, it's probably never used.

C

You just make it explicit that you should never use it and you move on I think for everything that is a feature it's much more complicated and I mean I. Think Barry was pretty hardcore about, let's ban everything that could potentially be insecure, but the problem is that you can't really pinpoint it right. It's like there's plenty of features like binary format, aware well, if you were to obviously be the hub, would you even do it?

C

Would you just ban binary format or what do you say, certain API patterns around only Paula Matic and then at that point, you're kind of like really into the you know, analyzer. You know territory because you really want to give this feature an idea at that point, as if you use binary format or a lot in your code base, like you, know, disable all of our solution at that point, right right.

F

And the argument without you is, there are safe ways to use it to turn on the application. Yes, for instance, like you, you simply never cross a trust omein, whereas something like this, the the very specific use case for this API is violating a CLR invariant.

C

It's basically a rocker on right, I, think I think. The other reason why I might use this API is because you, for some reason, ended up relying on reference equality in cases you don't want to do that but, like it seems very.

F

C

Not going to work in the future anyway, well.

F

If we were to do the streamlining yes right, but there there is a there. I am seeing more of a push for that lately and what I want to do is get the ecosystem, at least into a state where we can do the experiment without breaking people to too much, because we've primed people for these changes coming down the pike. Yes,.

C

A Christian is just posting, it says, API port says 9.8% I think the problem with the API ports, that is, that I, don't have you know: I, don't access the cloud side, but I only see the aggregated results, but I am pretty sure that most of the users you see in API for this for me, either some library or or some designer generated stuff. That probably doesn't matter like okay I, for example, I view strings or copy and msbuild to get to the string.

C

Formatting thing is because you have to back them was to call the transforms or whatever, because it's the only way you can actually invoke functions in its door, but it's not actually what you want so I think in general at least it seems fine, all right. So there's any pushback on the obsoleting.

A

H

Presumably, making this obsolete will also entail fixing up a few usage. We have invented across core FX amazing that it's a correctly yes.

A

D

C

But this was one way on pink neon I think time, nothing here on lenders as well, so this is effectively the type that we already have internally I, don't know whether we had the exact type of it certainly have these methods internally. So.

E

Up until about two weeks ago, we had about fifty different methods that matched all of these signatures, spread throughout for FX and for CLR total and one of the community members went and created the internal bid, ops, class and centralized all of this as part of the larger request for that we previously reviewed and said, we wanted more information on these cases, so he centralized them all.

E

Now we have this exact class with these exact methods and Tom's basically said we should review this, because these phones are actually shown to be highly used and it answers the issue of or Hardware intrinsics, which exposed most of these methods right. What do you do for the software generic component right.

C

The video said this is also the place that we would put like crc32, like anything that we would say, callers can assume this API will always work and we just take care of using had friends with it, accelerate them right. Yeah.

F

These things you're looking at a 32 or 16 value as the world in there it's rotating.

E

F

E

Crc32 from the perspective of how the hardware in Seagate guys are exposed are essentially, it takes a previous value and the next integer value and it combines it and returns another integer. So it's essentially doing it on okay, so the crc32 primitive, not the c32 right, great, okay right, the stream is built using the primitives and potentially you would put the primitive here for lack of a better place.

C

All right so then I.

E

Think the only interesting one here is a bit ops log. You would think that we should put her on math with the other log functions, but we can't because it's a breaking change but- and in this case it is a basically a primitive bit off for integer type. So the.

C

Why that are breaking change.

A

Because it really big because.

E

The float in double overloads integers is implicitly cast to double today and now to be implicitly going to int and then implicitly casting back to double on the return value, but they think you went in who long so still implicitly convert to double and then the result, which is a you nth or integer, implicitly convert to back to the double but the uterus expecting. So it's a double whammy of bringing.

C

So today, what do you have in math? We have so.

E

Today, if you call math dot log and you pass in the literal one- yes, it calls math dot log double where the one is casted to a double.

E

If we add the UN toward the end overload, then that one will instead be treated as hewent and it will return the unsigned integer result rather than double result.

F

Theory, one can ever you it. huh It's inclusively convertible to you in the.

E

F

Litoral Litoral.

E

One yeah, the literal one, it's implicitly convertible to you in.

E

Yep, which is why we had to create the math F class as well.

F

And it was at times it.

B

F

Because this is indicating that this is giving a different well liked, because the other love too is given you know, if you give it a thing, that's not a power to its any view back up, non-integral solution and.

E

Price is saying like.

F

You're getting an integral solution, so you better hope that you're asking the right question.

E

Well, so we also in three: oh now have I log in math, which is the integral log. According to the actor police back for mm flow.

E

If I mean so, would it be better to just add I think this is the appropriate place and I was just explicitly calling out that it's here in Norman math, because we can't put it on math and it probably makes sense being well.

F

So, if do we want, if we're adding this, but whether or not we're adding this because of the double conversion already do we also want to a dialogue for the integral pipes as input so that it doesn't uh convert to double them back if you happen to give it as an integral value, if I love is giving the same solution, is this forgive.

E

It gives the same solution for exactly representable integral values. So if you pass in like one point zero as a double, it will give back the same result as integral one. But if you pass on like one point one, it will give the double integral rabbiting. Should we have the un's.

H

And you own overloads to ILOG, so that your integer values don't get converted to double. To get I loves to get turned back into no.

F

They were just sideways.

H

Call this I don't think.

E

Just because the I I log is a specific I, Triple E function.

C

I mean I think we could put lock to math because basically, the the usage is non-existent like there's like as far as I can see. Nobody calls log to well.

E

That's because we just added.

A

E

Yeah, adding it right now would not be very well. It will lead to confusion for users who are expecting it to work the same way as every other log function. Yeah.

F

I guess again the fact that all the other ones are you learning that, like this is 2 to the 2 points, o-73.

C

Now, cuckoos, no they're saying they're not just different in five, do two different in behavior.

E

Or so, to weigh that long, so integral log returns the power of to the nears power of two and trims off any fractional bits. That's what this does. Basically, oh because.

C

It does need a bigger tool, so if you go to the number five right.

F

Well, cool would be some reflection on Tuesday there, two or three depending.

C

On which way it's going okay, yeah I, guess that makes it that is fair.

E

So I guess indicates this.

C

Is behavior for.

E

This log to is basically find the most significant bit set and return that index. What's the behavior for they rotate functions and put offset exceeds, we.

F

E

Matched the C sharp. What C sharp does, because we implicitly just use shift some of the cover which is and the offset with 31 for 63.

F

G

The it mask as if it's a unsigned we do.

A

Exactly what C sharp nose.

F

With a regular base and 41 yeah, which means that if it's a negative value, it ends up being like 32, minus n value.

C

Yeah, what do people think of a name and names bit.

F

We don't normally use off syllable, reuse operations. That is an in type names. We do prefer to to extend them I'm just thinking we had cryptographic operations right that went in for three up I. Think.

E

Bit operations is probably the better name. It's a bit verbose, especially for places where you're just wanting to do things like rotate left and such and I know mo and I talked a little bit all flying about the name. Space I think basically, there's two options here: they're system or system numerix system makes sense in the fact that these are really primitive operations that just aren't exposed by the language for the most part and they're fairly commonly used. But system is also not just the drop box for everything.

E

If it was system numerix, it would be the only type and system numerix that is not part of the external system. Numerix.

F

One argument for the Unison systems that we already have things like the Kindle in there things that already do manipulation or primitives. This seems like a sister of plastid up yeah.

C

We do also only have system for commonly used stuff right. This is I, don't think commonly used and it's commonly used by people that do bit manipulation, which is a very small so.

F

Like we'd have to be a little bit Wiggly with the turn converter, but just a dissolve isn't over yeah.

C

Look at the converter is super busy, though, and like epic converter is really also more about creating new pages out of other pages like this is really I mean if you squint. That's, why I said Oh? Could you see me double baby out of two bins, but like it's not really what yeah it's.

F

It's bit converters basically like a blessed reinterpret cast. Yes,.

C

Yeah I went to the to maximize its Jeremy, like I, try to avoid getting a new tie, but I guess I, don't I think I think that probably should be in you. Try, because we were also probably grow that list all the time. Yeah there's.

E

C

E

The other proposal, which isn't the these are ones that are actually being actively used, includes things like insert. They extract bit compliment bit.

C

Yeah yeah I'm, not keen on system, look like I have to say: that's the only thing I mean I I want to see I think what merits would be not bad yeah.

E

I think numerix would be an acceptable compromise and it would be someplace that people might go to look for this type of stuff and honestly I. Think if we went back and did it again, math would probably be better in numerix as well then system, for example, so.

F

You you mentioned you mentioned earlier right.

D

F

That was that yeah you mentioned earlier. The system numerix didn't have enough types in it, maybe justify this going into that namespace. What was that concerned? I.

E

E

At least from what I've seen previously, we don't like having a namespace that contains a single type and all the other types that are in system. Numerix are like big and dieter and stuff are not part of the standard package, but I.

F

Don't it looks like that? There's the system numeric saving space when that standard is split across that standard dll in systems where inspectors right.

E

Which is basically what.

A

We've already split the namespace across the sender that right.

E

It's basically things that are used by core lab and things that are completely out back.

F

Hey, that's fine too, like if we're shoving it in court of it, because that's what we're gonna use! It didn't.

E

It's fine to be insistent, yes, I.

F

E

Is growing with things like the tenser types and there's requests for other pegs to be added, which would probably go in numerix as well. Yeah system.

F

Numerix but I worry yeah.

C

I think, honestly, that seems right because there's also a vector of teeth being is rich. Henna was the old version of other intrinsics, we're all being hyper specific mm-hmm and I mean this is similar in the sense that it's also not Hardware specific is just hardly accelerated. I think that and it fits see, even any objections to that I take silence is agreement just in.

H

Numerix good operations, yeah no I, like it.

E

Someone in chat suggested bit primitives as an alternative name, just just run out like.

F

Kelly have binary, promotes yeah.

D

System, poppers I, don't.

E

Think it makes sense in buffers now, though,.

F

E

H

F

H

C

Never prevented precedent.

F

To use that primitive, yeah, yeah.

C

I, don't think we use to turn the primitives.

H

F

It looks like what we have in there is that I want to read things and I want to I know what Indian Miss the data is and I don't necessarily know what Indian is the hard way right. Yes, yeah.

C

I think bit of version is a bit more septic voluntary or what it is. The primitive seems a bit like you have to be the bull. That's right. There's.

F

A good primitives.

A

F

The bitten right today, we rotate a bit vector.

B

Primitive on a bit.

B

F

G

Did it again system image.

C

C

You just answer that, basically it's just, but it's not the trim. You want to use.

B

For this and binary.

F

Primitives as part of the system, memory package and yeah doesn't feel related to that.

B

F

Binary primitives also is again like a reinterpret cast, exchanging representation.

C

Right so they'll images to prove.

E

Is it for not standard two one? That's the question. Yeah.

C

Like that's, why I take that as API is thinking it for the Senate, but one of the things I think Yan was pushing for this, because it's a community see internally usage of you want to centralize this and it sucks when it's in a library- and you can use it right and since we're effort to say that right now might as well just put it there not making anybody's life substantially harder. There.

D

You ready right what you got sounds like what we.

F

D

Pretty clear, Vivian wait.

F

You it sounds like what we need to do, then is the engineering team is just like: do a scrub through cor cor, cor effects and see, if maybe there are other patterns that it would make sense to bring onto that time. There's.

E

A few, but there were so like insert bait right bait. Oh you! Imagine those earlier. Those.

D

E

Also used, there was some discussion with yan on whether or not those were considered simple enough to implement. Look just in mind that it was worth exposing.

E

I was I was at the side. It is especially since you can support those with intrinsics like pts, BTR and stuff, but it definitely required for further discussion. Yeah.

F

So what is Harry I'll just say? Yes,.

C

Yeah, so that is sort of awesome context, so immediately edit the ranch time, but we needed this method get offset in length right there. Basically, you close the range over the size of the container right, because the range can have two indices that are open-ended by they're basically says: oh I'll start from in our logical n, minus ten to logical, and but you don't know the actual bounds at that point. Right and so there's been basic basic gives you back the actual amount in the way we design dat nice?

C

We said well what you really want to do is you want to basically get twins out, but like we looked at out and looked ugly, and then we said okay, but whatever you do be able to use it up for them. Is that all we don't want to use tuples for various reasons? And then we said well, whatever just have it try.

C

There is a deconstruct method on it, so that you can call the method and immediately constructed into two variables right, so basically get kind of the the best of both worlds, and since we just reviewed the the other thing with the language team, one like how do we name tuples, we also kind of talked a bit about versioning a bit more and I. Think this is a clear example of something where well. The ep9 is called get offset in length.

C

It would never ever return more than two values right so like in that sense, there's not really a concern regarding versioning like it's just two things you will never ever add to it and if you ever were to do anything with that, but we want to return more layers. You probably just get an overload at that point. Tuples are probably the thing you want to use and yeah. We may also talk about naming. We decided to go with the book, the past, the casing right.

C

So the way the API would look like would be like this, which I think makes more sense. Giving her er I mean.

F

The only the only thing that I can think of is that we could. We could precisely we wanted to use offset Malik because which means there is a input type on things right, and we can point to people.

F

But really we have to stand in memory for that so or.

A

Just range right, I mean the whole point will be. That range is the.

C

Exchange time for or ranges father that has lost it in names, but also the length is just the the thing you would expect the receiver to firm convert you right, Jeremy. Why couldn't we? There was a tuple.

G

F

When I wrote down for the proposal, guidelines for tuples is never accept one as an input parameter, not that we've not yet but like because as an input parameter, it makes more sense to remove the parentheses and just pass an offset in a path of length. You know they're not really related.

H

But in this program the same thing except for the offside and might offset in like the struct right, yeah.

F

Well I mean so then that was my my my statement for that one is, but it would be weird to sort of pass one of those cuz. You would just pass a a memory. Is it's clearly related to something at that point, and then emo said before you were just pass rate.

F

I, don't really see it like I, can't think of a case where we would want to use the offsetting link as as a type in a method. Signature therefore, switching to to listen.

C

Mabel combination.

B

C

What are you, thoughts on that or your side? I'm.

H

Fine with it I think it makes more sense to use it to pull. Now that we've agreed upon naming convention I mean we, we were basically going out of our way and making things more more type. Intensive and more. You know instruction intensive, more CPU intensive just to avoid using a tuple, so yeah.

F

So now that we've approved a thing and they're going to put this in that standard, it's probably reasonable to resurrect the naming a new mobile, not dip. Yes,.

H

Yeah I already talked about that emo. It's I need to revert the revert of the reverb.

B

I like to see the git commit message for this: let's be ridiculous. There.

H

Might even one two or too many perverts in there, but yeah yeah I feel, like you said, a number that was going to be taken out again, maybe should be reverted. The revert yes.

C

All right, so this is one where hopefully somebody else understands it, because I'm not entirely sure I did the honors. He did not so so for context. So we added these API, so I, the added runtime feature is dynamic. Code compiled and one-time feature is that any code supported and the intent of these API is was to say. Well, if you only care about the fact that you can generate code on the fly, then you only check for is dynamic code supported.

C

However, there might be implementations of the runtime where the way this is achieved is by Il interpretation, which might be slow right. So if you want to use cogent at runtime to speed things up, what you really want to check for is is that I'm, a code compiler, which is basically it implies it's supported, but it also means it's actually cheated at one time rather than being just interpreted at one time and so for the idea. Do you know the two common scenarios?

C

Are you want to do mocking where you don't really care about the performance? You just care about the fact that you need to generate a type at runtime that implements an interface, so you have no other choice but to use it the only thing there you would care for is you know: do we have dynamic code at all? And then, if you want to do things like read text, we want to use code gen to optimize. You would just use the other API.

C

That was the intent now Yonex pointing out that the way I understand his his comment, the monitor and tell you I do. But the way I understand it. Is that because these API is our net new, you don't have a way to reliably detect an older version of the runtime where dynamic code is supported or it is compiled, but you don't have an API to query for that.

C

So, basically, if you use the dynamic light of API is that we have, which is the runtime feature, dot is supported and you pass in its dynamic code support that it's anime code, compile you know in a runtime that doesn't know about those two ways. It will just return false right, even though the answer is wrong because you know core seal our internet core OH, one two clearly supported it's clearly compiled, but given that they do that, I don't know about these flags. That would return false.

C

um My opinion is that that I, don't think that's fundamentally bad. It just means that it's unfortunate because we didn't envision the API as before, but I I think the the fix for that is you just do platform checks? If you really insist on supporting the court to X right and then moving forward, you just use the new API. It's the way yon proposes the fix is to say: well, maybe we should just invert the values so that when you write code, it looks like that. I.

C

Don't really sure I understand that reasoning, though, because to me the current behavior is conservative, meaning if you under one time that doesn't know about those things, the worst that can happen. Is you assume it doesn't have a feature which seems better than the reverse, where you assume a feature is present because it doesn't know about the values.

F

Having the notion of we don't know about this for this one, so it clearly is supported. Well there there may be places that your code can run, but they don't support it and they don't understand the question.

B

F

B

F

I'm saying I disagree with your business. Okay, I agree with the Conservatives you.

E

Like his proposal be there, which is no API change and update and supported stuff to recognize the not prefix, so you can still write it logically right for people who want to do it that way. So.

C

Let me just extend this because I took it all so, what's understand, why proposed it so basically ideas you mentioned, you have another API like runtime, future dot is known, and you can basically ask do you know about the value and then you can interpret the falls. I saw, you know about it, but you definitely know the answer is false versus oh, it's unknown because you actually don't know them the value. The problem of another API is that the other API also does make this.

C

So if you're running on a 2, X, runtime you're still screwed, and so the unfix is well, we just we just basically make it recognized the pretty sure doesn't make fine. You api's, however, I'm not entirely sure how this would solve the 2x problem, because you will still have to serve as it somehow to make it recognize the the exclamation mark. We would say false.

F

But this Oh it's not compiled. So yes, it's compiled.

F

But again, this would be. This would be a problem on the house thing, because the API supports this, because you know on all the existing runtimes that don't know about this right. They would still enjoy going forward it. It makes sense, but that's created right.

C

Which to me is I have to say, like I'm, not entirely sure.

H

So I sort of feel like the answers.

E

E

The thing that I don't understand from Yuans proposal is why can't the user just put the not in front of the runtime, is supported call. Why does it have to be a part of this so.

C

That this is the light-up scenario right, so the in support of the API exists forever in 55. This exposed right. This property only exists in platforms that happen to have it right. So if you're targeting that Center tool, you can still call this thing here, passing in the string and then when you're running on an on a runtime that would support this property.

C

It looks to return true, but you wouldn't be able to express it in a net sender to a library right.

E

But so my question is: why does the not have to be part of the string? Why can't you put it? He has a truth table in.

C

F

C

Call yeah because it's basically the wrong thing right. So basically, when you say is supported right. So let's say the runtime doesn't know that string. It will return false right, which means it will, but that doesn't mean when you're negated, that it's no supporting the feature. Alright, alright, sorry that it that it no means the code is interpreted right because basically it's a it's a it's a tri-state right. It's true thoughts and I. Don't know that Keith. If you scroll down to like the first comment, there's the other two out there.

F

So in particulars the two one and over one time so is dynamic and not is dynamic, return, false.

G

Because it doesn't understand the question, but why can't we just happen is not supported. That's something that you can't.

F

Call them to them, I think, is the argue, but I think that, because we're introducing a new concept to an existing API, the existing runtimes, don't understand it, which means we're now interpreting the not wrong on those which mean shouldn't change in semantic behavior and show that care about the older runtimes, DVD yeah, okay, I.

C

Just understand ho Yantra positive fixes to 2x case yes, because the idea is that you don't have to touch 2x. 2X just will return false for both of those strings and then the three or runtime would effectively interpret the bang and given moment in for resulted, and so it's it's all. I did.

F

That assumes that any future feature that we ever had we had I mean. Maybe that is the thing that we will do that that we add it with crew? Is the happy path so that we never accept the version. Am I.

C

Have to say, given that the only run absolutely good for the polymers. In my opinion, this goes back to the same discussion we just had with the vendors on how to solve. Malleability is that we are basically now signing up to support a bang for all runtimes moving forward for all new keys when the actuality, the only places we care about is the 2x runtime and these two strings right, which seems like a fairly heavy machinery for what amounts to be a point in time. If you write.

F

All that we really care about is is to one services.

C

That, or just tell the person by the platform, jack yeah, the only problem with platform shakes is: did we bring that out of our way to make them impossible right? So.

B

If somebody had to write the code in Hollywood, detect it to 1 1 10 reliably, but right.

F

But if my paper out is making his dynamic code compiled return through on that core 2.1 and that's that's a service.

C

Yeah the thing is, though, you can't lie on it: vulnerable, bugfixes right that gets in like well. If you happen to one on the server's runtime, you get it otherwise, you're screwed right.

F

But again it's the if it returns false because you don't have the update. You get your figure. You know your other all the.

C

Dead space, it might my general argument of the future use. If there's a lot that you're running into a crash, it just means your code doesn't run as efficient as it could have, in which case I think the behavior of the API is is correct. Is then you know you think the worst that can happen? Is you fall back on something conservative but I? Think the question then is well: what do we do for all new values? Right I mean we just we.

C

The way the feature was expressed is that the outcome happens to be conservative, but that's more happenstance than actually design of these two properties. So the question is when we add something else, but you could run into similar problems so.

F

Then what you actually want is like an is understood. Oh yeah, yeah.

C

So what I proposed down here was something like that where you would basically say this, you would say: runtime feature is known, blah blah, you turn on blah and then you say: well then you have the api's. Then ok, let's cure them yeah. If you don't have them well, then let's do some other crazy cut from checks too, but I deduced it because.

F

C

Can't make this work on the tri-state. Can we it would be a very constrained like.

F

C

F

C

F

Annie- and you know like I, supported I, don't support it. I have no clue what you're talking about what we watched in market yeah I, support it and I don't support it and I have no idea. Yeah.

C

The other thing, if you were to do that improbable question mark, would be better Becca, I, honestly think like, given how you write the code. I honestly think this is also not horrible. You just have someone duplication on your strings, but you.

E

We could, instead of adding a new, is known API and an overload of is supported that takes that has a pool that says is known as well notable yeah. Basically so, if it returns true, then it is assumed that is known as true, but if it returns false, then you can check is known for.

C

So what do you pass in? Basically, the question mark no.

E

No, so you would pass in it support it is dynamic, could support it, and then you would have an overload that returns. An out bull is known. So if the API is supported, both are true. If it returns false, then is known. Tells you whether or not that version of the wrong time understands the API.

F

That's not bad actually, as you have it hit a hit. A case label irritated people, yeah.

E

And then to one will still always return false, but 300 and forward people can check. Does the runtime understand the thing that you're querying for.

F

Technically speaking, our support thing says the moment that we have released the dot release. We no longer support the previous, not releases of that membership, and so we don't support 2.0 2.1 punto. If we bug fix this, we only support the version that in the book accident, but.

E

Then we also don't have to worry about what does not mean for every single API we just have to be like. Does this runtime understand this flag.

F

And so I do think I heard that we should fix the problem going forward, but if 2.1 is the only thing that we actually care about here, we should solve the problem going forward and dump it. You know I think.

B

The pond yeah the owner, determines.

C

Approaches that you can pre-record right, because you can only require like major minor what this is, what our support matrix says and.

F

We only support the latest release, I.

C

Think you viber a cell phone provider. I would probably just write a platform check at that point. Say like you know: I, don't know whether you guys are fixing 2x and even if you do I, don't know how I practice so I, just I just write a platform check for that. Then I.

E

Think the downside of the platform track is, though, as then platform support expands and no.

C

That's why you use these API there's a platform, support, expense.

H

They were significant.

C

Right and so that's why exactly the scope was written basically say: if you don't have the capability API is then I, then I light up on the ones that are happy to care about right and then years. Everybody else just goes to this path, because you would expect that whoever shows up after the fact implements the latest API I mean it's not guaranteed, but that's kind of the same way of doing things, which is why I wouldn't care for new things.

C

That's why I think that phone I mean I generally, don't like platform tracks with the reasons, as you said like. If things expand, your your branches expand right. The other problem is it's not reliable, because different platforms have different ways to detect them so also non-uniform, ID, C code looks always like somebody got a PhD in path of detection versus these ones are effectively uniform and they always work the light up. The only downside is well, it doesn't work for the past, but the past is by definition, a fixed set. So it's yes.

E

The platform check API is may not work in the past as well, depending on how things expand in the future. It wouldn't cover everything anyways yeah, so I like the is known the best but I think, rather than an explicit method having you.

C

Would Spacey say is supported with with.

E

An outbreak strain future.

C

Name but feature, and an hour is no it's I'm just is it.

E

Is known, he's understood, cells.

C

Known feature some.

E

F

Then all the implementations just use one guess which they set the out and return, and then we change the existing API to I think the out parameter off this way. It always works and so and then yeah you're, not you don't have to make 92 different switch statements or anything else like everything so you're proposing returning true for I, understand. True, is I, understand this and I support it return, false means, I, don't understand it or I understand it and it's not supported and.

E

F

Out the parameter which sounds I did or did not understand.

E

Alternatively, we.

F

Have a convention for this yeah.

E

It's all a try, try again to support.

F

It yeah or trying to support it yeah, which returns to her fault returns. Whether or not it understood the question and I would see whether or not it's supported.

E

You know because.

F

That I was gonna, say this is exactly just dictionary target value with this right.

F

And but we could say that in this one, what people really care about more is the if it says false, that probably means or if, if it doesn't say yes I support it. They're, probably going to treat and I, didn't understand the question and I didn't support. It was the same, but in a few cases and like that, okay, we could invert the relationship on this yeah.

E

I think the inverted relationship definitely makes more sense here, because it if the first, if if the result is true, then the other one is implicitly true as well. Yes, but in the case that would the result is false, which is the case that you care about switching on the secondary result. Tells you why.

E

C

The homes subside from check a pianist by the way does anybody know.

D

We can find out to check it in.

D

D

Beyond it looks like it was the last person to change.

F

Yeah, let's say on parent is a part of the system. One time contra holy is I've. Been me.

C

Alright, so any objections to what I just wrote.

C

C

My natural charm.

E

I think that's fun. All.

D

F

This is a standard impact. It.

C

Probably is marked as such. The thing is, we should probably find it well, I guess it's the same issue right. So, let's just comment on this and let's just run it back to API needs work.

D

Yeah, how is this.

E

Yon doesn't have objections, then it goes to approved without you didn't come back here.

C

Well, if he has objections there, we should write.

E

It yes protections, but if he agrees then, if.

C

He agrees, then, we should, after the proposal to reflect that they simple basic, reflect my feedback. Isn't like okay, who post this new API, and then we approved that one and then implement one.

F

This life did make it is in that 71, so we should update it with this better pattern in phenomena so that we don't do introduce some stuff. Obviously,.

C

All right, so this snow comes back to technically.

F

We could remove the one without the other as long as the one with the I was present the next night. If you want people.

C

All right so any particular issues that you wanna look at ten. Are you read one three.

E

Five: five: six zero.

C

E

No three: five, five, six.

D

Leave a good one. I have two three five: five, three zero.

A

C

We zero what hey? What's the other one, three.

D

Four eight reason.

F

Mm three: four: three: six, six, no I'll have it.

C

Four three.

F

Six six: we.

E

This one's easy.

C

E

48 eggs in a 2x2 class we expose insert vector, 128 and extract vector 128 and, unlike most of the other api's, we are currently exposing. Both forms that take in address and forms that take a vector directly, there's actually no need to expose the address forms. We can just support the address form via containment in the JIT, like we do with most of the other hardware intrinsic api. So the suggestion is we just remove these. To avoid confusion, help make the code more performant and reduce complain. Then.

C

The feature hasn't shifted, don't care anybody that says. Opinions on this I would just prove it. It's.

F

Fun, so the idea is, if you want to extract from a particular object from an index of a particular address. You get a vector 128 point here, for instance, immediately directed and then pass the Dirac version into the other overloads.

E

Know so, for example, if you were to do extract to an address, you would do extract value index and you would pass the result of that too.

A

E

And just like for load, you just say, load vector and then pass that vector to as the value and the Jade knows to say. If this is Fault immediately followed by a store. Yes, then just fold it so that it does the efficient encoding it. Basically, this leaves the only things that actually take an address are the explicit load and store instructions and the broadcast instructions.

E

The broadcast instructions were originally part of this proposal, but we can't actually remove all of them, because AVX for flow to double only expose a form that was reg mem and then for a B x2. They exposed the red trench version and then so for consistency. We agreed just just keep exposing both forms for all of the broadcasts.

F

I'm just gonna ask the devil's advocate question. Since all of the idea of this is this is just great hardware. Instructions are we now having opinions about which hardware instructions were included? Construction said it was this a convention or a convenience, overlook something. No.

E

These aren't convenience overloads. These are actual underlying instructions that are commonly.

F

And now somebody has two: instead of one construction to instruction, it was just a different way that still lives. It's a different way with the one.

E

Structure yeah, so so, basically, someone will say, store the that they'll say a B X store and inside. As the pram builder they'll say, a B X extract, blah blah blah and the JIT will say fold that down to just extract with the memory actually.

D

Of a Pierrot that uses others.

F

Thank you better I might need to change my PLN ow.

A

All right, so that means.

C

A

E

C

G

Deletions it's great to wait long and immature, but.

F

It is Eric on the phone, no, no okay. So what happened here was um there was a there was a public API on room that was not exposed in the rough assemblies and during some refactoring in core effects. Recently it got exposed in the rough assemblies, even though it had never been approved, but so this is, let's review it the like. Clearly, the name is such that it was never intended for public exposure but yeah. Basically, given a rune write it out as utf-8, we already have the utf-16 overload, it was already implemented.

F

The reason it was never previously approved is because we were waiting for Tory to come online.

F

Just going to use an overload of trying Hoda, but at the last review we said that we're okay, with putting things like utf-8 utf-16, blah blah blah, is suffixes on these methods. You.

C

Think they're I think what we said earlier ones that for most of our cases, it's obvious from the tide yeah or from the method name already what's happening. In this case it seems it would benefit her to say, try and code through utf-16 and the previous one. You just have clarity, I mean.

F

We can always rename it like it never shipped in too long right here.

E

F

We get chari, we can have try encode span of char and try and code Spanish re, and then the type information tells me what it's a code. This one is because it's a four bytes tell you what what bites it's right again.

F

Third Suze like a mild wart, but everywhere that we're doing utf-8.

C

So sorry, are you suggesting maybe leaf trying code without the suffix, because we can add the overload later or do you suggest to rename trying so.

F

Try the first ones already there in the rough assembly today. The second one is the one that has never previously been released in any preview right, the second one we would definitely want to have utf-8 somewhere in the name. If it's a span of, why doesn't read it.

D

F

Think that we would, though, that I mean either we renamed. So there's gonna be a third thing. Let's call it X release right, so we get Charlie online of what it's called so we're gonna have to try encodes with your span of char and span of Charlie's or the current one. We make a breaking change to rename breaking from the real pregame the rename, try and code to utf-16 right and then we have two overloads and try and code to utf-8, which is span of bite and span of char eat. Granted.

F

We could also add try and code to PS 16 span of bite because, as we code I, don't.

E

F

And so it's a question of where do we want the you?

F

Where are we getting the type inference of overall resolution, and where are we getting the good name so I think that trying code is fine, for that, like we have a span of the rich type thing, but phantom fight? What if I have any information, so it breaks you.

C

So you're a citizen leaving the previous month, as is then and then later on, potentially at the overload I. Think it's fine yeah I mean that's I mean that's reasonably for again.

F

If we had, if we were getting carried this release, we would have just to trying codes right and we wouldn't have anything with it, and, and that is why we never approved this API in the previous review right, because we're cooking yeah because we wanted the overloads. So this is a deadly that works for me.

F

If we defer drew until we had Chari and not have one that had you or you can hate the name, there's always to say this methods being added early, so it's so letting the control code as utf-8 them instead of to utf-8 bytes.

F

What is utf-8 then, if we wanted to the other methods that wrote to bytes would be trying code as utf-16, big-endian, little-endian and utf-32 beginning there. So there would be five. If, if we took this to its logical extreme, there will be five methods that had the trying code as block pattern. Well, assuming that's, we usually have 16 with a boolean parameter. Do you.

E

Think I, like the name just fine code, as you.

C

I think if he study, but even he was gonna long to go anyway, I see we only hit a better call.

C

All right so said: try encode as utf-8 and then strip divides because it takes.

F

Bytes as a parameter, why duplicate the name in the method I.

E

Don't think we're ever going to have like trying code to utf-8 string.

F

Orkut to you, so they will show the the feature. Branch implementation of this type has a two utf-8 string method on it, which, just yesterday.

G

F

D

F

Fabricates, a brand-new type and then or 22 or under instance- and there were there- are some comments and the threat to about when it returns true and false and waterfall, but they don't know one other thing there is and I didn't I didn't really call it out here. Unfortunately, uh there are two properties hanging off of the rim: tag right now: utf-8 sequence, length and utf-16 sequence length that tell you the number of code units in utf-8 utf-16 respectively.

F

Do we like those names, since this is our opportunity to rename them? If we need to it, seems because the other like if we wanted them to look closer and intellisense, it would be like plane, boat unit length, utf-16, which just feels awkward. No, as a property I think it's fine, you can't they put unit length in you can see well there. They are called right now, utf-8 sequence, length and utf-16 sequence length like we could remove the word sequence. We get to a place of a code unit like we could just know off.

F

Don't do anything! You just said you know it. It's a sequence apparently needs back.

F

Let's see of you unless you're arguing that you want to rename the thing with you there's health products. No, it's the the spec. Does you sequence just in terms of life, it is all strings. Are sequences of movements yeah so.

D

F

Say and I know we haven't been very consistent with this, depending is that we focus man.

F

Do we want to have than on try burdens of this, because our technically our try guidance says if you have a trial, you should have an on try, and in this case there is a property telling you exactly how long so you could test it or you could simply do the assay like there, possibly some places that I'm going to want to try and do some wait, three auditions of where we've added try API for crypto and all that I do literally everywhere. I call it is, if not try or it returns the wrong output.

F

You know simply fail or throw an exception. It's like wait. Why am I not just calling do this? Oh I mean we, we could. I I have no qualms with having an actual income muffin because and then in code technically there's two different possible overloads. We have one is it helps the length and the other one is it throws if the input was not the exact right way, so I actually have that API, it's internal right, and so this is.

D

F

Like while thinking of what I wanted with hash code that could be or try to compute hash right here with span is well with this one I know you better write. You know this number of bytes. If you didn't something's wrong, so I'm gonna throw so. Why doesn't the API just do this? For me, I mean if it follows the different follows the conventions that we've set out everywhere else with always one the problem is with span so far was fan.

F

Our rule is ignore the rule that says add an entree overbook, add an on tried variant, and that's because really we, when we started doing span, we we changed left rhyme in talking to try parse pattern, so technically we're in the territory, but this is part of trying to codify what we think the rules should be going forward. Hooker, it's a really easy change job because of the way email people even ask the question you would have to prove that is or just decided it's grandfathered under English doctor.

A

I'm talking about it so going back to the you guys: okay,.

C

You generate sequence, language, abate sequence, like the only thing I would suggest is: can we make this? They may have a common prefix rather than a common something yeah.

F

But I feel that, like the energy flow like the noun, that you're asking over is is utf-8 as a property. Putting the utf-8 on the end just looks.

E

In my head, it's kind of like long length versus yeah.

A

Yeah, it's awesome. Oh great, a piano itself, I think.

F

That the API is amazing adequately, but not happily, but advocate correctly correctly. Okay,.

C

My caramel is not super-high, just I, don't know it seems a bit on the long side for sure, and it also seems a bit more comfortable than it needs to be Broadway.

A

But I was only the greatest scientist rejection.

C

All right so then, let's go to Jeremy's. First.

C

Did you support it when you're on the other one? Do you want to order both, except what you're trying to do so.

F

I think this will just be less contentious if I.

C

F

Get something done.

F

Rsa encryption.

F

They there's a variation there's two forms of RSA encryption: one's called pkcs, which is classic thing: one's called OAP, which is a new thing and and instead of making a new algorithm, what RSA did is they added it as an option? And so then we in our classes that it is an option, and so there's this whole plumbing problem.

F

So what this is saying is for our our envelope CMS, like there's, no way that you can indicate with our type right now that you want to use the newer form of encryption padding.

F

So this is just adding an overload that when you pass another certificate, you can pass system the padding mode it'll throw if it's not RSA, and then you can also there's an overload before you can specify the subject type and then so, if you had a great overloads and then a the nullable property to indicate, so it will return null if you didn't call one of these overloads. Otherwise it would end of that first, a kid. Why? Wouldn't you just assume a very particular algorithm?

F

Doesn't a default topeka CIO's the defaults, the pkcs? The problem is, as the else isn't so easy, but you don't know if he recipient splits, it so good says the person saying I'm encrypting this you can say I want to use OE.

F

Cesare the old API has to do with the old API. Does this new API lets you do the new secure standard of it? The printing already exists. Rsa encryption padding is a type that is existed before. All of that.

F

B

Inaudible though, it's.

C

Really close hmm come back the company inaudible because it's already closed.

F

Are the encryption padding is a Byzantine impersonates, a.

C

F

Yeah, if it's a glass thing and.

D

Sears no I yeah. Okay, though I thought it was a strong spring yeah.

C

That is what you came from.

C

G

It doesn't matter.

C

If I'd be the same, like oh, it's a.

A

G

String class thing: it's a magic string, class Oh.

F

No, well, it's because it's a it's a it's a one pole or two pole.

F

Because of yes, because the haha removed.

H

In version omen.

F

And the hash algorithm is a magic script. Yeah.

C

Took a dump: oh that's! Okay,.

F

So it's a so remove the question mark because it's already like I said, is it going to be surprising to people if this property might work remote, they didn't know any input knows what they get at the Apple. Well,.

C

Strictly speaking, though, you can put the question my balloon for my right, so I didn't leave it it. Maybe that's what even today yeah and then.

F

I'm just doing one deck right now to see if any other properties on this thing that gets that but I'll just get.

A

Nobody using the new features all right. They.

F

Don't get on, he has it the constructor and.

F

I didn't know, one might have an opinion all right. This cryptography, we had opinions, we just had very loosely held opinions and you know had the opinion that, besides the class decided I think.

C

That's the best opinion. My opinions affect.

F

Okay, so this is needed for some of the Jason serializer work, that's being done, the we currently have ways to escape HTML, URIs and JavaScript strings in the framework, those mechanisms all except utf-16 data, because we have things like utf-8, jason serializers. It would be great if we had mechanisms that accepted utf-8 data.

F

There are two proposals here: one is adding a brand-new set of types to do: utf-8 and utf-16 types, but very the two shall meet. The second proposal is actually adding utf-8 overloads to all of the existing utf-16.

F

The plus side to the first proposal is, it doesn't come with any of the legacy or cruft or anything else that is afflicting the original types. The problem with that is, you now have effectively duplication of I. Don't want to say logic, but, like you effectively have duplications wherever the framework. Now one one does utf-16 one does utf-8 and if you were in asp.net and you're customizing, it saying hey I want to put a new encoder in through the DI system. Now you have to specify it, though not just one.

F

The plus side to the second proposal is that solves the asp.net problem. Just say this is the encoder that you're going to use. You only have to mention it once, and the downside to that is. It makes the API shaping that weird.

F

F

um This again is the first proposal looking at here. This is this was written in terms of in terms of charge. It could be written in terms of bite because it's explicitly utf-8 the type name slightly over okay.

C

So this is I think we looked at this earlier right, yeah yeah, so remind me again what the problem was with the existing times. It's basically the way they're like they're, going over the API. It's not the stream, whatever they, but the apps, where the actual abstract methods are not the ones you want to expose or something to explain right. Basically,.

F

All of the abstract methods are written in terms of utf-16, which again is totally fine, because, if you haven't oh say say that we add utf8 methods to the utf-16 class.

F

What we would do is we would just say if you haven't over in the utf-8 methods, we use the utf-16 ones as the source of truth, and we just always transcode each way under the covers, like it works totally fine and that's actually how some.

D

F

The existing time, having an existing type name, is text encoder in the namespace systematic encoding is web.

C

And that one is utf-16, probably because it takes a string, takes Ryder and then it's fine for this character playing cool which takes the Chop winder and then yeah from there on I, said yeah I see.

F

So the that's one of the other interesting things about these types, you'll notice that there are two men. There are three methods which you have two over, which remark is abstract. So if you want two subclasses, you absolutely must override these methods. Two of them are on the safe, and all three of them are marked is out of a browsable, never know it's late, yeah. So in practice, no one ever subclasses these types.

D

Aj, yes, both.

D

C

In Kingston older so like how does but you only proposing this one, you don't expose you don't proposal utf-16 texting order, because the original texting cover is utf-16. So, are you basically saying that if you will utf-16 you texting code of youth, utf-8 used to theoretically yeah.

F

And we basically don't touch the original types they stay asses, so there are also subclass types: HTML encoder, javascript, encoder, URI encoder, your encoder, it's similar when.

A

C

F

C

That narrowing the other yes.

D

Mirrors it exactly yes,.

F

Just the alternative is editing existing class, adding virtuals that are I'll call the utf-16 when I fell, we encode this yell.

E

So that we can work.

F

Yes is, is anybody so you mentioned that, probably in practice very few people are deriving for me right. So theoretically, if we just fixed before of these we've solved, 1990 versus fine, yes and the new virtuals would not ever be.

D

F

If for the seams, I guess the the problem is when you're deriving it we're creating the same problem every when creating the same problems that we've done the screen every time we've upgraded it, but there's another thing that you're supposed to override, but the language can't help you and you're supposed to, but if no one's ever already an out of them, yeah I looked and I looked in api's I've done that and the usage of this was I, didn't see any usage whatsoever of anyone over any less.

F

And it's always said if your core extension API is basically ends its code.

C

F

Yeah well I think what I did is I looked at the constructor of each of these types and saw zero usage, the protected constructors, no.

C

I think that's I think that's the way that I proposals in the original markets, originally you proposed to like break the type entirely it just stick, do something else with it right, delete some methods, add new methods or something right, and then I think this one is the close as we get to about and complete duplication, and we were breaking people we've seen that occur so.

F

You still have you still have the original proposal on the screen, like the one that actually adds whole new types if you scroll down halfway through the payment, that's the proposal that uh that their alternative. This is the one that that adds api's to the existing type and doesn't break anything as far as any where it does obsolete some properties in some methods.

C

The benefit of this is that people don't have to know what the encoding is correct.

F

Well, you never have to know per se well.

C

What do you mean? Well, the agreement is so the parole that you currently have basically splits this hierarchy into two hierarchies, where this utf-16 one and there's YouTube 8:1 yeah. So when you it, when you exchange the tribe, you have to make a decision. What the underlying encoding is. So basically, your API is manifestly including that you're writing to my address is the other one.

C

You have an actual abstraction in text encoder and you don't care what the whether you write to you to fat16 on utf-8 there's only one there's only one type to choose right in terms of choosing an encoder in terms of like UFC, you, you except an encoder your right to that encoder, and then whether the underlying thing is ends up being utf-8 in JavaScript or utf-16, and you RL, you don't care right. So you leave agnostic.

F

With it, yes- and no I mean yes, it prevents you from having to worry about utf-8 versus utf-16. You just write to the encoder and both api's. Were there the intent. The intent of these api's was not to abstract away I'm doing HTML URL JavaScript, like the caller, absolutely has to know which of the three is right into, and in fact the original implementations of all these encoders did not have a shared abstract class, but it means that the original that that the caller has diffused I'm doing HTML Javascript. Well, yes,.

E

F

Need to care that the output is utf-8, that's why, if the the only usually only one person could use the semantic notion and another person can choose the binary encoding? Yes, yes, so I have a general preference of amending this context. Instead of creating an entire parallel hierarchy, that has you yeah.

B

The only question that.

F

I have is how many places are we gonna add the word by only mental and Chari, or do you not that's the other thing? If we start adding byte overloads like we absolutely have to start putting utf-8 somewhere in these method, names.

D

So lazy thing like.

F

F

I, don't think that's intuitive, though you know things like intellisense. There are some things that change here. For instance, uh the max output characters per input character, it's the very first property shown here. It goes from abstract to virtual return, zero and becomes obsolete.

B

D

C

A

The what is the customer doing besides, they just passed them in tonight. They don't actually call any methods of the.

F

Customer if you're configuring asp.net what you're doing is you're calling one of the existing static factories on these types. That code does not change the the line of code that you write does not change from what it is today. So what is differently? There are create methods on HTML, encoder, URL, encoder and JavaScript string. Encoder they're not shown here because they're already existing methods on this one post types, but you would call if you wanted to customize how asp.net works.

F

You call one of the static factories today, just as you did yesterday, just as you did last year and I know changing what is changing is if you wanted to subclass the type which I have no evidence.

F

Anyone on the planet does right the way that you sub classic changes, for instance, this property, which you were previously required to subclass or which we were previously required to override because it's abstract, even though it's marked editor browse whenever it goes from abstract to virtual and becomes obsolete stuff like that, because the property just returns a lie to you.

F

Because I can't give you I, can't give you an answer like.

F

Fun, yes, technically, that might help but character or character. Yes, it's talking about something I've, never surrogate or a basic multilingual plane. It's talking about a system car! It's talking about a utf-16 code. You know, but no one uses the property internal.

F

And because it was editor browsable, never no one uses a property externally. I know no point to even having that like it. Honestly I would just delete the property, but that's a breaking change, even though again api's I've done now shows that nobody on earth uses it.

C

Yeah I think I said this last time, like I I, think it'd be motivated with properly. It seems like something. Let me make it away in three all right. If I mean.

F

C

We basically say, look, there's no usage, the thing is just plain wrong should have never been shipped. You shipped it. Unfortunately,.

F

It was originally intended to be an implementation detail of the original encoders, which is why never returned the correct value anyway publicly, so the the one problem with the leading would be. If somebody has extended it. They'll now have an override to a method with them type under campfire yeah.

C

Right and so they know, I made it to breaking change, but all I'm saying is that it's on that blue surface, as we know, nobody devised something nobody uses them. Yeah like at that point is like well how much more data do you want to motivate? A breaking change right, like I, mean sorry to take mitigate the impact of a breaking change, but if so all we know nobody does that yeah. Then the question is okay: what's the benefit of not having the thing well, the same, API again.

F

Right, it's the way to do it, which is why this is I, thought them the least impactful thing where we leave it editor browse. Whenever we go from abstract a virtual, we return zero and we're just off Selina Sam. Don't call me, don't override me pretend I'm not here, which anything is also.

C

Many message might need some tweaking about this.

B

Maybe we should add in a pie emoji in there somewhere, the pie is a lie: yeah I mean I'm I'm, fine with me. Okay,.

C

I, don't think yeah I think the only thing is we need to go through the motions. I. Don't think we have they did we have files for any yeah I. Don't think we do foil for anymore for any behavior breaking changes, for example, for three: oh I, don't even know what the what the process is, but it wouldn't be involving I would be interested in like doing it to just see that whether we can actually get away with it. um So I do.

F

Actually, I take that back. I do have a proposed breaking change coming up for the ASCII encoding class, but the feature area owners have said that they don't anticipate. Anybody relies on that behavior and they were willing to say yes, this is an acceptable change. I, don't know what the process is aside from just getting the future homers. Do you agree that it's exciting.

C

So, okay, look for! Let's just bail goes back then to this to discussions here. I don't mean that minimum we just chipped the AP and I, hide it and tell people not to use it. Your sanity.

F

Versus we change these words that effectively.

A

C

There's a cleanliness argument, for we could go the extra mile and just make it clean. So assuming we do this, is there anything else you want to get rid of like what about this? Guy all the same same.

F

Same story right so that method, because it takes a pointer I, want to get rid of it like none of these methods should take pointers, especially the ones that are abstract, because it means some class types have to fall back to unsafe code by just in order to subclass us, which is not a good way to begin. I would love to rewrite these in terms of spam. So then we also Eve which related apparently abstract with body yeah yeah. So he got it right on the first.

A

F

Have the rest, yeah sorry.

C

Yes, that makes sense, and basically you would say, hey so now the question. So if we were going to say.

F

Just discussing if we were going to make a breaking change by removing these abstract, but we also make it be the same class breaking change of adding new things as abstract. The new utf-8 stuff, even yeah, I I wouldn't make it abstract because it's not strictly necessary. We could always under the covers.

F

Do the work to transcode back and forth I mean just like if we think that it would always be a bad idea, then you could say that somebody has to if they want to just write the trans code and do that themselves, but that really what they should be doing is writing the directing coding, while they're doing it.

C

So so I would say if the argument is, we are a team of removing abstract members, which means it definitely will break everybody, whoever derive from this time. It might as well add new abstract members at that point, because you basically have to touch that Corrigan.

F

This is like it's essentially a part guy who's on these is there is a base implementation. You are highly recommended to always overwrite it yeah. Then, let's just say: there's not a base implementation. You are required to provide Europe because then we're gonna do it in all our types sure. So if we think that no one will derive this other than us, then we should, if we're going to delete the abstracts I, think we should introduce the new phase. Okay to indicate these are things you're supposed to overwrite, there's an easy and easy implementation.

F

You can even make it a protected static of like you're transient. So if we had to redo the entire tie from scratch, there are only two methods total which we racked up, which are absolutely required to be abstract.

F

The first is: do I need to encode this particular room, and the second is: how do you encode it like everything else in the world can be built on top of those two notes, everything right now, the existing type before any of these proposals has for abstract methods, which means the existing type is already more complicated than it has to be.

C

And to be clearly, he originally wrote at the time that was most bad look. There was no room corrupt, and so now we would say they've expressed it as what span of one I suppose it would be a of.

F

The end of whatever he protected virtuals would be exposed in terms of room or int or whatever it doesn't really matter right, and the public API surface is exposed in terms of span yeah. So that would make sense and.

C

Then I think yeah. This is an area where I think we should probably go with that like I think originally, when you proposed that I said, can we see it impose that it's non-breaking for us before we yeah this.

E

Is what it looks like.

C

To be numbering, but I think that now I think that we understand what the thing does so like do. We all agree at least that having one tie will be better than having two types or is there any benefit of having two types? It seems like you're, basically losing the ability to be a bit independent right. There.

F

Are benefits to having to having one time like I, did kind of reverse my position on this over the past few months? It's it's just easier. It's easier to configure it's easier to consume, it's easier to subclass, easier to reason about that. Also no I dislike the notion that we take everything in the framework that exists that and it deals with the concept of text, and then we had a utf-8 in front of every single one of them. Yeah oh yeah,.

E

Everything and.

F

So I think that we need that starting to get out of that pattern, I would be fixing fixing this so that it can work for utf-8 and utf-16 and all the various other encoding stuffs. Clearly, your abstract, how do I I have this room? How do I write it down unless you're doing room ruin into a series of rooms it would need to somewhere convey if someone.

E

F

Kind of encoding is someone asking do this. Now we have a. How do you talk about a particular coding type.

F

But yeah I would definitely like to see the proliferation of utf-8 specific types go away. Yeah we especially given that you have to use unsafe, derive from this at all, and the our telemetry says: we've not seen anybody do it so I would I would say that this type can just be rewritten and maybe we'll find out. That was a bad idea yeah, but it isn't. It is a major version.

F

Therefore, we are allowed to introduce a breaking change and therefore I really want if we were to introduce breaking changes just even in an experimental fashion. This might not be the worst candidate to do so, just because of how many people do suspense unless it I have seen I shouldn't, say no one in the world subclasses us I've seen one one place in all of my searching where I've seen someone subclasses and that's an asp.net unit host.

F

In the unittest could obviously be changed about your parents.

E

It also sounds like for this breaking change. If we took it, there wouldn't be any implicit breaks that show up to the user correct.

F

From the only breaks are two people who subclass it, assuming that we we make all the abstract method change the entire public API surface exposed to external callers include the identical unchanged right.

E

And, and for the people that do subclass it if they exist, it's not a case of. We are silently introducing a breaking change to them. It will be up and in their face. Yes, yeah.

F

Well, I mean these abstracts are all public but they're yeah, but they're all editor browsable, never.

F

That even I'll show up on ESPN. Oh, they probably do.

E

I think my biggest concern with breaking change is always are we doing a breaking change that is silence to the user? Oh they'll end up like recompiling their code and never see that the breaking.

F

Yeah they show up on understand.

F

That's I've never stood it still. I must see. Unto me.

C

Yeah Ben is asking whether we should just have utf-16 utf-8, but I think we answered that we don't want you force the consumer or consumer. Sorry yeah, either consumer or producer against of api's. That accepts one of those guys to already choose me in Kuwait. So the.

F

When you were talking about accepting a text encoder like as an abstract base class, if I am the Jason serializer, for instance, the the idea behind a lot of this- is uh in the Jason serializer settings like we would put as a property that you can sent a JavaScript encoder like you can actually set that property move on, but we would take it explicitly as JavaScript encoder, not text encoder, because we want to signal to the user like no, no, no, you were doing JavaScript string encoding, write the text encoder base class.

F

The only reason it exists is that a lot of these types have a similar API shape. But beyond that, nobody should ever be using the base class. Nobody should be referencing them. Nobody should be subclass on it. It always.

C

Be doing it to one of the private sphere, I think I think they yeah. This is the. This is the the thing you said earlier about that you know the the semantics are important, but I think the question is still like. Is the encoding yeah to the business, and that gives you tips? 16 was duty.

C

I made the important it'll be basically one avoid that you have two dimensions of the types to choose by the the semantics on one axis and then the physical representation on the other one yeah, and it seems we want to not do that. So then, let's get to the following. That's like give me a right above the motivation that you have everything getting a little bit more internally as well, that we can just take the ship room and see what what what the thought is it public pain.

C

They will probably ask me what the Obama library can changes. This and I would say. Well, in this case, the motivation is there, the usage is basically non-existent, so it seems fine, thankfully like if you ever want to do breaking change them, and that's that is clear, determined.

F

So I'll follow up with you offline there. Then we can just figure out what needs to go into that email right. Okay,.

C

Something the images mark. This guy's API needs work, but it's basically said their goods. Do we.

F

Know what work needs yeah, which puts us in a better position. The last time.

F

But then, when it comes back, it would be what's the DLC.

D

C

All right so know, we've been through my show, that's raising yeah, he said that's time left I can't believe it. Let me talk half an hour, um so let me just bring this on the screen and then we can collectively slightly so.

E

I actually have a issue that we previously reviewed, that I ended up having a couple questions on after we approved it, and it might be good to just clarify on there's also this breaking change to unsafe unbox. This.

C

E

You know this is a three four. Eight eight one I've not updated it to be. If.

C

It's learning the list. I can literally open it because.

F

I only feel really unsafe, unbox.

C

In the middle and.

F

Down down this worker.

D

Okay, I am torn on this.

E

I'll update the view that way. That's.

C

You want a box of tea.

F

No I don't want to introduce this class I'm, just saying if you introduce it in your own code, it's a way to work around it with.

C

You, if you then, what are you performing? What's.

D

On this we turn left. We.

C

F

So the unbox instruction in il returns, what's called a control unit, ility reference, which we don't have a way of representing in c-sharp the closest way to represent. It is a refereed only t-not rft, but it's still not an exact way to represent it.

F

The reason for this is the original motivation of the unbox. A funbox api was to unbox an INT mutate it and then send the newly mutated box 10th somewhere else or processing by like a 2 string or something api. The thing is, like the runtime says: thou shalt not do that, because box dance really are supposed to be immutable. The there are alternative proposals or alternative scenarios where I have a boxed.

F

My struct and my struct is absolutely intended to be mutable, so you should be able to mutate the thing that's within the box object, and that could be a reason why you might want unbox to return a ref T instead of refereed only T, but that wasn't the original motivation behind the API beam form unit and.

E

It's probably worth pointing out further down that it's actually, although the codes not very common, it's entirely valid to have a mutable box which makes this more confusing. Yes,.

F

Basically, what you want is if T is marked read-only then you'll want the unboxed to return refereed. Only to you, if T's not mark read-only you'll want the bought, the unboxed or the term ref like that's the closest thing to what we actually want.

C

Oh to me feather like you, you don't even have to define its track, but you could just use the strongbox when we already have right. If you really want to you sure, but I think the question is, but that will be a break.

C

I just got out of the car, apparently because why.

F

B

F

Believe this is when they did the lights, the fight. You know five minutes remaining intermission.

C

This is like you know, you're halfway to the moon, and then you get this update poem. That says you have five minutes and then they will remove your machine. Yes,.

B

C

If you say the unsaved or am box is already shipped no, not into one it's new for you, the reason I'm saying it shift is, and that could be wrong like it shows up for 300, but it also shows up as don't make corporate uh from extensions in system not run team. That comparison is that answered the bishop in nearly a package Oh unsafe, no strippers under your catheter I.

E

Notice this week.

D

F

E

Think we also just moved in box. Yes Sweden. It.

C

Was available auto band before so, it will be a breaking.

F

Change to the outer band version: no, this I don't think this ever shipped in any in any supported version.

C

Yes, this was a very recently on.

F

If the nougat package was still only available as preview package with you.

C

Well, it shows them is looking a couple stuff, my extensions, which is always manual so maybe.

F

If click on 307 308 can see when it was introduced, I.

E

Don't think the API, yet it is, did you say classes now so.

F

The the issue to add this API was opened end of June 2018.

C

So it's not on you get I can tell you that right. So the package is on you there, but the package that is on you. It doesn't have the API, in which case it's an on-chip davia.

F

No yeah so I'm out of like different mines here, because on one hand it's on the unsafe class like it's unsafe, but on the other hand, the unsafe class isn't supposed to be a place where we put things that allow you to that. Encourage you to like destabilize a long time, for instance like we'll, never be.

H

Unsafe ass exposes il operations, which is all this is doing, but.

F

Yes, but it's returning the wrong type, though that's the problem like what we really want is to return a control, mutability reference, which we can't really. We can't really do in c-sharp. Unfortunately, it's.

H

Not returning the wrong type, though it's returning exactly what the ILO operation exposes, the the I/o operation.

F

It so you're right, um but if you read through the echo spec, it says that the only safe way to do this and by safe I mean like will not destabilize. The runtime is to expose it as a control. Mutability reference like we can't just take a boxed, int, mutate it and then expect things will just work like that could destabilize at one time.

F

It will not destabilize core sealer in the current implementation, of course, ELR correct, but for 3/1 we can't make that same guarantee before. Oh, we can't make that same guarantee.

H

All right, I hear you you're saying we might make a change that would break that at the same time, there are, if we don't do this, if we don't expose it, people will just produce their own nuget package that already exists out there, which does exactly this. Yes,.

F

But that's okay, because it's not us shipping it as a blessed API. We.

H

Already shipped, the il is a bus, an API not for this did. We know.

F

B il is there anyone who can write I out? Yes? Yes, yes, but if you, if you, if you write that il directly like, if, if you write that il, then you have read the ACMA specification, which said: thou shalt not do this unless you're willing to take these things as consequences right.

C

Well, it's even let me ask a different question, so are you opposed to marking it as video only or are you just playing that with a fricative saying? Why do we care to enforce something that may not be relevant for everybody.

H

I-I think we're jumping through hoops to shield people from something.

G

That we don't need to shield them from, but it's unsafe technical.

C

Downside come back, something as refereed only mean the worst. That can happen. Is you have to pass the way that we don't mean this.

E

The other thing that happens is it is very easy to introduce an accident copy based on the c-sharp semantics, which is, if you end up invoking any method etc on that type, and that type itself is not read-only and it creates a hidden copy and does the operation throws away the copy and then gives you back the original back.

F

I, do also want to stress it like the original code sample we were given when we approved the original API is such that it could destabilize the runtime again, not one three, oh I stopped time. It can no like there's nothing. Stopping the runtime from saying a boxed instance of a read-only T is itself a singleton. The runtime thing absolutely do, though. You cannot do that for arbitrary keys for read-only to you absolutely own, like Indian food.

H

That she could absolutely mutate itself, not.

F

If it's mercury only sure it can using unsafe code, yes, no.

E

You've been using safe code, Souma gravel gate in 307 30 gave an example of just using console dot, write line on a struct foo that has two string incrementing the value in how you can cause it to mutate and.

F

Then the struct itself is more freed. Only no.

E

Okay, so yeah, not in the case, winds, drunk itself with.

F

Lakeview I when I say when the struct is read-only I'm talking about like int or as in the original sample usual stuff like that like. If the truck itself is not marked read-only yeah, it's mutable have fun, but you can mutate a box in a box truck.

C

Sure I think so I mean the argument would be.

C

If you were to ship, the API I think we want to expose effectively what the thing does right. I mean you know, I, don't know whether read only ref or refereed only is that is the reference. Referee is the closer thing for my l, but the thing is that the general spirit of the unsafe thing has been that.

C

Well, if you call any of these api's, you better know what you're doing right like none of these api's are designed to be well, you just call the API and if it doesn't require you to put unsafe in c-sharp you're fine right, it still means you can model as well like arbitrary constraints that somebody might have, and so you need to be mindful of that.

C

I think John's position in general has always been that he doesn't like this thing in the middle of like well you're kind of guarded, but at some point you fall off the cliff. They can either guard if you're not guarded. I find this on one inside it's appealing, because it's very simple messaging, on the other hand, I think it's a bit poor love the baby with the bathwater at times where it's like.

C

Well, you can avoid easy mistakes, but I think if what Turner says it's true, then I would also say: well, you, you trade, one class over a easy mistakes of another class of easy mistakes. Right, it's like well, you know better off with this thing being mark meet only or not very nice. The.

E

Read-Only issue is going to be slightly better in c-sharp aid, because so in c-sharp 7 you can only mark an entire structures read-only, and so you have the issue with partially mutable structs. Even when the struct is read-only, you could have a field which is itself not read-only and therefore can cause copies. But T Charmaine is adding read-only method, so you can explicitly say a method no longer mutates, which will help in some of the cases to avoid those hidden, copies. Well,.

C

If people put this all over the place right, that's the right.

B

Which I am doubtful later.

F

So I guess my opinion on it would be if, if something in making it be ref T, if the way that we're layering something is we are we're violating what's supposed to be invariant, then we should fix that if we're not violating the variant thing.

E

So what the wrong comes back says here is it's a controlled mutability managed pointer, which means the type decides whether the value can be mutated and there's nothing at the io level. That allows you to specify whether or not the type allows the thing to be. It's.

A

E

Knowledge of this type is neutral, even if it's marked immutable or etc. Right.

F

Look, the closest thing we have is a declarative, read-only keyword, but it's still not an exact match right and.

E

So I agree, the safer thing would be to say it returns, ref read-only and if you do decide it's beautiful. Yes, just.

F

E

F

And this is why I am of two minds on this right, like a it's on the unsafe class like you're you're, doing something that is clearly unsafe, but, on the other hand, like I, don't want an API that encourages people to do things that that other parts of the runtime might be relying on you not doing.

F

Yeah, if I understand correctly you the read me this means, if you have a box, if you box of five yeah, you can without making a new box, replace the box as being a box wrapping at seven. Yes,.

E

F

You reference equals again to the same box, but the value that comes out wasn't evaluated with it correct the reference equal being present and you which.

H

Is no different than having your class, it's like a strongbox in it. Okay,.

F

That is correct, it I mean and yeah. So if this is literally just the same as writing that il, which is probably actually packing its implementation, then I don't have, then my official answer is I, don't have it it would it okay? If, if we wanted to keep this as we're up to you, should we just like really strongly document like hey there? Is you have to be aware of all these crazy things with this very particular API yeah.

C

Somebody the chat, said you're invoking a method on the unsafe time you have been warned I've been at the same time. Yeah I will probably help to say like what is the expectation for this material and that's it because otherwise it's like well, she would have to be careful, but what does that translate to right? So I think linking to the artifact is also reasonable to say this: does this aisle instruction, so you know there's already a spec written that said the tense of what what?

C

If what you are supposed to do, act upon me when you say something like you know, it makes people they might destabilize. The runtime I think what ends up happening in multiple titles like well I'll find out, if that's the case, and unless it's not the case, I assume everything's, fine and I. Think here like, if you see they care, if you, if you try to scare people too much, they just start to ignore the warning.

F

And I do I do want to. You know, hold that out, because I'm, not I'm, not trying to you, know, spread. Thought and Steve is correct, like in three.

F

Oh, it's not going to destabilize anything, but remember we agreed earlier this meeting to obsolete string talk coffee because a future runtime reserves the right to do things like string, basically automatic string entering via juice during tea, see like a future runtime reserves, the right to pool boxed instances of read only tease, there's no reason to run time couldn't do that and if we get people relying on this API we've now broken a long code.

G

Or make the API grow burden cooking, I think.

F

But then why would we introduce an API now doesn't light throw in the future.

H

Well, I hear what you're saying and, at the end of the day, like I'd rather make it refereed only then remove the API, but there are a myriad of ways that people can and already are doing the same thing already, whether it's using the aisle directly or whether it's doing the example that Yan has in one of these threads, where he just uses unsafe. As with a simple class that is a single field of tea, and you know unsafe at value yeah like whether or not this API is there.

H

People are going to be doing this, and if we change the runtime, we will be breaking people. So I. Don't believe that exposing this API is ref versus refereed. Only in any way shields us from breaking people.

F

I, that seems almost like a defeatist argument, though, in the sense that well, if other people are already doing things that are going to that are that are going to you know, violate these invariants. We should expose an API that just allows them to do it in a blessed fashion. No.

H

It's not defeatist. It's realistic and saying. If we're gonna make what's actually a breaking change based on behavior exhibited today versus behavior exhibited tomorrow, then we need to be aware that we are going to be breaking people and we need to do the notion of well okay, its side by side. People won't upgrade and they'll change their code if they need to so, if someone's using unsafe thought unbox and there you know assigning a value into it and that ends up breaking in the future.

H

Well, then, they'll have to not do that, regardless of what mechanism they're using to achieve this.

F

Or no I thought that argument makes sense.

H

Son said so I just dug into the game like honestly, hey, if you want to add, read only I, think it sort of doesn't matter with.

F

The easiest thing just be to make this a dark. Everyone I think.

C

That's probably fear I think, as I said, I think it's really good on unbox or so on, unsafe. The document, what the expectation of the API is I just think, like anything, would be permissible. I think we'll be okay. To quote that section from the arrow. It would be okay to link to it, I mean I, don't think you have to go out of your way to make it like idiot to be there buddy that this code.

E

Is also for perf oriented scenarios and in the cases where you do want to mutate it, which is probably the common case for this API having to also insert and unsafe as to split off the read-only Ness is going to end up impacting jet throughput, because it's got more oil to process. More calls has to cetera or.

F

Local studio, some of the features perspective string copy. The problem is, if you want to use, string, dot copy within, pin and modify the string. There is a. There is a gap between the time that it is returned and the time that you've pinned it where it could have been D duped and the reference ends up obliterating the data. In this case, if we start changing in the runtime to do collapse boxing the runtime understands it just gave a ref to inside the box, and it could not collapse this box.

F

It can move that to another DC phase. So like one of them, the runtime understands that you have the ref in the other one. It doesn't because you have a reference type instead of a value type. So okay, but if the only thing that we're concerned with this pooled boxing later I think that we would know that we have this API and we would yeah. That is, that is almost similar concern and and if we think that we can't do the ref tracking.

F

With that feature, we would make this API start throwing and that's better than having people who are hand rolling it and then they're now violating invariants.

H

Okay and the terminus, but Jeremy you mean this- would if we did start cooling boxes and you used this on a poverty box, we would throw that's.

F

One possibility it's I mean it's all the question of how much we value that feature versus the safety of this or unsafety of this API etc, like once, you think that you're replacing what's in the box. Presumably you think that you're, the only one who understands the box or that you know that you think that you understand everyone who understands the box. If it's pooled box five, then how many places in the runtime are you replacing their notion of five? So that's when it would get dangerous and that's now just a cost of that feature.

F

I would say the easiest answer. Is we just never pull boxes?

F

H

Really easy to not be featured.

F

H

And I mean honestly, you know the the going back to the string copy thing. I I would not be surprised at what we find is we obsolete string copy and someone says oh I'm, getting an obsolete warning. What I better do is call string to char a new string, and now the D do feature will still be dupe them and they'll still have the exact same problem. It'll just be less efficient.

H

Well, it will always come up with. You know. Ways of.

F

H

F

Wonderful and you're right and we we did try to to to work around that earlier by saying. Maybe the the obsolete message can and give you like explicit steps on how to do what you're trying to do and again all. Alternatively, we can you know we own the runtime that would be d duping your strings. We could make a conscious decision to taint in objects that comes back from string that copy to say this.

F

One shall not be duped because someone's being stupid- and you know we we can do that- we can also make it throw again like, but hopefully, if somebody gets an obsolete warning, they at least look at why it says it's obsolete before just writing code to do what they want.

F

I can't promise that, but that's what I hope to do.

F

Okay, okay, so doc you shoot up so resolve this as I guess not resolve this as well fix that just turn it into a dog Fisher yeah yeah. If it's already dark contents already on the docks.

C

Oh cool yeah, I think the ones he picked that somebody gave I think Tom. Who was saying that we should probably include the span overlords spring Korean and obviously message whether the other two things yeah.

A

C

Better than they already covered that um yeah so that okay, so we basically concluded that we don't want to do that right.

F

I think the conclusion is that wrath, the current state better matches what the il semantics are and any future problems either is a cost of the future future that introduces the problems or is on the class. That says.

H

What say yeah and we should be very explicit in the docs- and you know, be very clear about what issues. What this does is, what it when unsafety it's exposing and that things future could change and.

F

If we really wanted to, we could also on a separate, not unsafe type in the future, expose refereed only T. If we wanted to.

F

If you didn't want to mutate, you just wanted to like cheaply almost no.

E

I mean, then you can just do referee. Only T equals unsafe god, no.

F

Again, without using the unsafe I mean it's I, don't really have a good scenario for it. It's just something we could do in the future. Thursday.

E

Night, oh, is this on.

F

On this drunk the boxed object is large. He would want the ref to the inside instead of yeah his for an end, it's like where he offered it up cheaper than a pointer. So, whatever.

F

C

Sure yeah just respond to this.

H

Box box about inspector gave.

F

Me back here, we don't, we just get.

E

The language, tad control, mutability hunters.

F

I'd be fine with us.

D

E

They technically have the concert today with refereed only they just there's no way to express that my l yeah.

F

Alright, so what's our current three to account remaining email, all the three yokan remaining? Yes, there we go.

A

F

Of them today, I just want to understand our bomb one burned down.

A

Like pretty family I know, Oh.

C

I told Jeremy yesterday I'm less concerned about this number I'm much more concerned about this number here.

F

Yeah we're gonna say that's a good number.

D

F

Nothing is ready. We.

C

Have API approved 31 all mark for three or, like somebody needs to decide what that we do there or not.

C

By the way, who is the customer of this unsafe, unbox, API Steve.

F

C

F

Or it was in the left, bug which is fine like right. We have customers for clearly like that's not under this Steve are used to Milano when he jumped up I'd.

E

Also see him like three or four people in the getter channel asked there's been people in the discord channel that have asked.

F

And the original, the original bug that was filed said it could be used to optimize IP address on Steve gave a code sample for help you use for them. I see.

C

Cool alright, so then, and I think we're done thanks guys online.