.Net Foundation Design Reviews, 4 Jun 2020

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: .NET Design Review

Description

00:00:00 - Approved: Add overloads to string trimming https://github.com/dotnet/runtime/issues/14386#issuecomment-639002448
00:37:42 - Approved: Non value returning TaskCompletionSource https://github.com/dotnet/runtime/issues/17393#issuecomment-639004797
00:42:08 - Approved: Add methods to convert between hexadecimal strings and bytes https://github.com/dotnet/runtime/issues/17837#issuecomment-639018813
01:09:00 - Approved: TCP Fast Open implementation? https://github.com/dotnet/runtime/issues/1476#issuecomment-639025242
01:20:41 - Approved: Add Task.WhenAny(Task,Task) overload to avoid Task[] allocation https://github.com/dotnet/runtime/issues/23021#issuecomment-639034041
01:30:09 - Approved: Add System.Numerics.Half 16 bit floating point number conforming to IEEE 754:2008 binary16 https://github.com/dotnet/runtime/issues/936#issuecomment-639050110
01:47:29 - Needs Work: Add "Create" method to EqualityComparer() class https://github.com/dotnet/runtime/issues/24460#issuecomment-639054509

A

And it seems like mike is working. That's great. Can somebody under the call say something is.

B

This, the actual bog, oh yeah,.

A

That is the actual red cloak, so no that's the first 10% of it or something. So we have I, think 57 issue open issues or something so the oldest issue we can only have is from 2015 there's only 5 years old.

A

So I guess it could be worse.

A

Does anybody have contacts on this one I.

A

Guess we marked it API ready for review tarik.

A

I guess Levi will probably be the person I need to comment on this.

B

C

Of adding these instance methods directly on the string either, but.

B

Yeah I, don't like them, as instance, methods and I- think rather than starting and I would rather have them named prefix and suffix.

A

So, instead of instance, method, what would you suggest static.

B

Methods like most of the other api's I.

A

Thought most methods on strings are in distance methods. No.

C

Are they I mean.

A

I think trim and others are all instance. Methods are not in there. So.

C

Yeah I guess we also have like starts with and ends with so prefix and suffix I.

A

Mean I will say: I go back and forth, I guess, generally speaking, when you take two arguments, nobody seems weird that they are statics I. Have there sorry there are instance methods like especially comparisons, because you usually have issues like not references when you do that. You.

D

A

Kind of things seem fine, as instance, methods I mean the way you would use them seems, whereas instance methods.

B

Yeah I guess: if thrust the API, sorry I'm, probably just crazy.

C

Okay, so that this is not a trend, it's essentially removed this prefix or suffix yeah.

A

I think trim has the semantics of anyway at the char way that it takes us like any charge.

B

A

Array, so if you, whatever, we move for particular sequence of chars- and this would be the API for that yeah.

E

Well, it's not yourself, but if you, if you say, trim X, for instance, and the string begins with xx both of them get removed. Presumably this one would not remove both prefixes right.

C

E

A

C

Guess this could end up, remove some combining characters or some basic characters without removing the combining characters.

E

If you pass in a linguistic string comparison, that should not be the case.

C

What is a linguistic string compose.

E

A current culture or variant culture, but.

F

You could pass in a different comparison and end up causing the string to effectively have corrupted. You see two didn't hug.

E

That's that's not likely, though, like in in order to get corrupted data like you would have to pass in a literal, prefix or suffix that already had corrupted data inside of it. So it's kind of a garbage in garbage out scenario right.

C

Is that true I mean you could have like eat acute in your string and then you just say remove start e and then you would just have an acute or something right.

E

Yes, but that's not corrupted data, it's it's meaningless, but it's not corrupted. Okay,.

D

I mean e acute is actually one code point.

D

Yeah I mean you: could you could normalize it to E and then the overlay acute, but so the the overlay characters are always going to be able to be in an inconsistent state. That's just true with strings always yeah.

E

But that, like yeah you, you already have that today, with trim index of substr and everything else like it's. It requires the string to already have that characteristic going into it, which I don't think is a huge problem, because if, if the common case for this is that you're going to be doing ordinal non linguistic comparisons on the string anyway, if you were already doing non linguistic comparisons, that combining acute accent, character? Who is already going to be treated by your application as a standalone thing, not something that combines with what came before.

C

Are these common enough operations that they should be first class method, so I, don't know how often they've had to actually do this I.

B

Actually, do it all the time and I generally do it by, if starts with substring yeah.

A

I haven't I, do the same thing, I mean my other argument would be even though they're not common. Where else would you put them? We don't really have the second tier of string methods. I mean ignoring link because they hold it show up, but.

E

Yeah we've we've had quite a few requests for similar api's throughout the years. I I.

B

Know, for example, in my peon bookbinding generator I frequently have to trim, underscores and the prefix tag, because Windows loves to type def using tag X as X.

C

Should the naming be removed from start.

E

Or remove prefix remove suffix but yeah.

C

I mean I, think prefix and suffix works I'm. Just thinking we also have starts with and ends with so starting and kind of have an argument as well. They.

E

The problem, what starts with them ends with right now is that they use current culture, I believe when you call them, someone should double-check me on that, and these ostensibly would use ordinal. So if you give them a similar name but use a different default, compare that might confuse people I want.

F

But I kind of wondered whether what no matter what we feel about the choice of current called tripper starts with and ends with I kind of wonder. If these should match, because I imagine, a common pattern will be if string dot starts with then remove start and you'd want those using the same comparison method.

C

But I think the purpose of this is that you can just call a remove start and it checks for you. But I do agree that the behavior should match yeah.

B

So it looks like starts with defaults to current culture, but there's an overload that allows you to pass in the culture. I would imagine we'd actually want overloads for remove start to take the culture as well then takes.

F

The culture or the.

B

F

B

C

Comparison does it oh there's one that takes a culture as well. Does.

E

It take a culture and a bool is is upper case.

G

E

Yeah I I'm not really a huge fan of those api's on string, because we already have the text info and compare info class, which is a more powerful version of that. It's just less discoverable, fair.

B

Enough as long as there's some way to do the intended comparison, you want yeah.

E

There there's an existing API, comparing photo is prefix. Compare info, dot is suffix, which is the equivalent it starts with and ends with.

E

However, there is no API in the framework today that allows you to do the equivalent of remove start and because it gets very, very weird once you start having a culture and go involved.

F

So when you say you mean api's, like the ones on the screen.

E

Correct there there's actually no correct way right now in the framework using the existing public API is to implement any of the methods on the screen.

F

Why is that? Because.

E

When you start having non ordinal comparisons, a string of length, a can compare, is equal to a string of length, not a which means you have to start accounting for those length. Differences when you have starts with ends with, which means that if you say that you don't have a remove start, API say that you only have starts with and substr. If my string starts with foo, how far into it do I substr in order to remove foo, we actually don't expose that information today.

F

You mean the food out length may be incorrect, correct.

E

This API presumably would wrap all of that logic for you, so that you don't have to worry about length differences.

F

It sounds like if this is approved. Levi has to review the pa5 while.

D

Doing like substring of the string would be wrong. Presumably if you did, if starts with and then remove, remove, will remove the right thing and I don't know if we have removed last, but you would get the same thing for ends with.

E

Maybe I don't actually have the API service in front of me I'm on a mobile device. I know.

F

We have removed I, can't remove but doesn't remove every occurrence. I thought that was remove. All I, don't think there is remove all then.

D

Maybe I'm thinking of JavaScript regular expressions again.

B

Remove that takes a string just one that takes an index wow.

F

There's there's replace so you call with the string in like empty string.

A

Yeah, that's what I do and that's actually annoying, because.

E

Oddly enough, a few months ago, I.

B

Would use the this API so I think it would be useful to add it would simplify a bunch of code. I agree.

E

Sorry by you, you got really quiet. Could you? Oh sorry, umm if it's, if it's called remove, start I believe you were arguing, it should have the same default as starts with I feel.

F

Like it should, I mean like someone can make an argument otherwise, but that would be my inclination sure that.

E

That's fine, we would just have an analyzer that says, don't call it I.

B

Would think it should either.

C

Be the default then.

E

B

Was just saying I think it should either be called removed from start or removes prefix I think remove start just. It sounds weird I agree.

F

I'm also fine, with Cori suggestion of we require you to pass in a string comparison.

C

Do we need another overload that lets you pass in and move? Is it a compare info? No.

E

So you would instead, you would have instance, api's on comparing foe itself.

C

A

So sorry I missed the one on the start uh-huh. So you guys saying don't have to start briefly as it starts suffix.

C

We either call it removed from start or remove prefix I.

B

Like the term prefix, better I.

E

Mean on compare, and so we would absolutely call it prefix, because the existing API is used, the name prefix, but on string we don't use that term anywhere. I mean.

A

To me, it seems weird to use something else, because we have already trimmed starting tremenda, feel kind of similar issue.

D

But trim is the verb like word like cousin.

E

D

Isn't it's not like removed the first six characters from started its remove if this is found at the start, and so I think the from conveys enough semantic difference there like we're, not necessarily actually removing anything, was.

A

Same as trim right, you may not true many thing if there's no occurrences of the trim characters, yeah.

F

A

I mean I, don't mind prefix and suffix, but as Levi said, we don't use it anywhere else. I don't want to be you start anywhere else, or was it just trim start to amend? We.

E

You starts with ends with I.

A

Guess yeah yeah.

E

On on string, we tend to use the terminology start and end in on compare info. We use the terminology prefix suffix. So when we add the api's to both classes, we would just follow existing nomenclature. I guess is.

B

The proposal for adding it to compare info missing from this proposal. It.

E

Is but we would just do it as part of the same work.

C

um Can I argue for having an overload here that takes a culture info I think that very few people are aware of compare info, but passing a culture info around is something that people are familiar with. They know how to do. It would make string class a lot easier to work with we had that here. We've.

E

We've been trying to avoid that with other string overloads we've generally. What we've been saying is uh with string. You really should be using either a well-known comparer or the ambient current culture one, and if you need something that is for a specific culture, you really should be using the compare info class anyway. So there's no reason to pollute string with existing overloads for that.

C

Okay and and trying to think of how this fits into current code, that is already passing culture and foes around it's I.

E

I wish, honestly, we could obsolete all of those it's there.

E

The way that we wanted to mutate that class going forward.

E

You've muffled yourself again leaving sorry, we can tell people if you really wanted this call this other API instead like you're, probably in the one percent case, if you really wanted to pass in a specific culture info, okay, so.

A

Levi on comparing for what the api's be called on, there would be instance, methods, I suppose it.

E

Would be instance, methods we probably called remove prefix and remove suffix, and they would take a they would have similar overloads. They would take a source spring, they would take a value fring or spam and they would take a compare options.

E

E

C

Pasted what is prefix looks like on the & Perrins. Oh.

A

So you said they take a source string.

A

And then you said, I think Vaidya string right, yeah.

C

Source prefix and options yep.

E

And options would be defaulted on most likely so.

A

It's prefixed rest calls its source and prefix I guess it would go with that.

A

And then we would say, compare options, options, okay and presumably.

B

A

Optional, similar to its prefix may have an overload with an overload of an.

E

Overload with the options and overload without yeah.

H

E

Why not just default the options? I, don't.

A

Know what other options are there yeah there's a class or an enum yeah.

C

It's just 11 is prefix method for span of char that defaults, the options to none yeah.

E

That's a that's a fairly new method. We just added it a few months ago. So.

C

Where's, that guy compared info dot is prefix.

A

Where do you see this again? That's what see this it's it's.

E

C

It might not be a I'm looking on source dotnet right now. So what would you be? What they'd be defaulted to? Lana compare options, none.

A

But presume what happens to be 0 I am yeah. 10 works for me. The.

E

The compare options allows you to specify things like ignore case, ignore symbols and so on.

E

So it's a flags, do you can or any of them together, I see most of them together.

A

All right, so what I don't have is on string. We have remove starred, remove start with comparison, every move and and remove end of comparison, and then we have on compare until we have a move prefix and remove suffix with each string source string, prefix I guess you have one will be called string, suffix and compare options which is defaulted to none.

E

And whatever span overloads need to exist to yes,.

A

It is what I'm looking at there is no span. So then, what is this ban? One called.

E

Its so on, compare info is brief or on compare info. Remove prefix and remove suffix would also have span based overloads with the parameter, names and default arguments, exactly matching the string based overloads and there.

A

Would be span of char, then read.

E

Only Spanish arias football.

A

E

Knew would return to read. Only Spanish are.

E

Presumably that.

A

Wouldn't make sense to me.

E

Actually, we should check to see if compare info has any API, so return read only span of char.

A

E

A

Know what I'm looking at this zero span api's, but that's because believe, slowly, indexed in.

E

So the one one thing that could potentially be a problem is what we discovered with read: only Spanish hard split. Where do you have to have multiple different overloads now, if the source is a span of char instead of a read only span of char, because do you have to return a span of char and so on.

A

It's almost like you want to say T where T is spend I yeah.

E

I leave I leave that discussion up to all of you. Well, we what we ended up doing is. We ended up having a span, dot split return ranges instead of read-only spans of teeth, yeah.

A

I mean sure that might be extremely annoying to use, though yeah I guess maybe focal painful. It's fine, because it's not really a first class API anyway, it's kind of like a plumbing API but I feel like for mainland scenarios, I'm, not sure, that's the best generic fix for these things. Okay,.

E

A

E

The sports world, with the Sun I.

A

Think I mean the argument for using against ranges is that we don't have to duplicate all the API is all the time, but.

E

Yeah, that's why it was done for force band split, because otherwise you'd need span of char read-only, span of char memory of char and read-only memory of char. So we have one overload instead of four.

A

All right, then, all right, then let me how do I show this again. So what about memory? Extensions I'll be happy with the proposed overloads here.

C

So these are essentially slicing.

A

That's I would understand it. Yeah yeah.

E

They're slicing and if the prefix or suffix doesn't exist, they just return the input I.

D

Don't know that we like did, we do trim start of T right, if not I, don't think we should add the remove yeah.

C

Nothing could be, did I, don't think we did so.

D

This the specialized to chart- or we don't Lee span of char, seems reasonable, but I would remove the generic ones.

A

Yeah, so we we have done some things like contains this generic copy tools, generic ends with his generic binary searches, generic info office, generic lost in the next office, generic.

A

Ok is whitespace is not, but that's good, yeah.

D

Cuz cuz for for string, for you know, string likes the there's, the complexity that you're the thing that you're removing in the amount that it removes aren't the same, but the generic one. If we have like this, is just like two lines of removing two lines of code so to.

A

Answer your interconnection little trim trim starting tremenda. All generic.

D

We did, we have trim start which takes a read-only span of T.

A

That is correct. We have a trim star that is specialized for force for memory of char and we'd only span off char as well. What do we actually know? We don't have the way to do it. I think those are just generic, but for the memory one, we have a specialized one for memory of char.

A

Actually, a stain correct, it is the trim star that takes you to only span off chars, well, yeah, so be specialized in amiable generics. So it's kind of interesting choice.

D

Right because, because if we're adding remove start and revenge generic and we're having it be the it does the trimming for you, then we need all four of them.

D

A

Then for them, because.

D

Of its generic, if you have a writable span, you'll want a writable span. If you have a writable memory, you want to writable memory, and if you have a read-only memory, you want to read only memory just because you can turn them all down into span doesn't mean that it's sensible to turn them back into the read only spend doesn't mean it's sensible to turn them back into the type that you had right.

A

D

Other alternative would be that it returns the range, but that's if you're calling her move start, that was just secrets, equals slice and remove IND is ends with and slice. So these are.

D

These are just like two line or convenience methods that we have to add aid. So four, so eight methods to deal with like I would say that they're not valuable enough, but that's just me.

A

G

A

Doing that all I'm saying is that if so, when you say they're not available enough, do you mean the the methods themselves, so you wouldn't have any any ones that take generic. So would you only take the ones that take literally the instantiated versions of chars I.

D

Would do the one that is a extension method on read-only span of char I would not do the one that's a generic on. We don't we span of T.

A

So, but why do you think you're saving any other lots of that? Doesn't the same argument hold for memory as well, then I.

D

Wouldn't do it from memory either.

A

So you, okay, so maybe I'm missing it but like. Why do you think we'd only spend off charmix enter? The only span of T doesn't make sense, because I mean it seems to me that there would be the same. I mean we could just read. Only men of Khalid.

D

Only span of char has the string semantics where the thing that you're removing can be or the amount that you remove, can be different than valued at length. 440. It's not it's! It's literally just two lines of code that the method is removing.

D

So I don't see adding eight to line or methods as valuable, but if people wanted to add them, I'm like I'm speaking of I, don't I think that there's reason to not add them in terms of there's not enough bang for the buck, but I'm not like fundamentally opposed to them. Yeah.

A

I mean I I mean to me the question that we need them at all. Right like that. That argument, I think makes sense to me the other one I'm, not sure, because you like my concern, is when we do the span one si. Let me do the Chawan now, then they don't only add the span, one as well, sorry, the t1 as well, and then again we have like twelve methods, so that seems like I mean unless there's value and specializing them for char, which I don't think the general we.

D

Have we have the char when a needs, the string, comparison and B has to do the or I guess it needs the string comparison? So it can do the thing where what it removes is not just valued at length. Okay,.

A

That's fair yeah. We need the string comparison which wouldn't apply to the T. Yes yeah, so.

D

We'll always need the specialized ones. I think that we don't need the non-specialized ones, and we can add them later. If.

A

D

A

H

A

Mean to me it's thank you. That's kind of a wash at this one I.

C

Agree: Jeremy I'd rather add the great only span of T later yeah I think.

A

They make sense, so if you only have ten remove start remove and we'd only span off char, the only spend off char string comparison and then we move end we'd only spend off char. We only spend off Chaucerian compared and.

C

Would compare info also have these spin methods as well? Yes,.

A

Except this compare options. Yes,.

C

They return range there so that you could operate on other things.

A

They would probably.

C

A

We don't spend off job with you. My guess, I have.

E

I would be fine with them. Returning range I mean it is a somewhat more advanced API yeah.

D

I mean using or having the the complicated one return range means that we don't need to to define all four variants like we can more comfortably define just the one.

E

That that seems to find by me I mean it's. It's kind of a new pattern being introduced to compare and foe, but.

D

E

A pattern that is more generalizable, yeah I.

D

Think we've used it in a couple places already not on compare info, but so.

A

I guess I can see you so comparing for what does trim do right now.

E

Compare sorry say that again.

A

So we don't have any sort of true methods right, so that's we don't have anything I'm campaign for right now that shines.

E

Correct we, we don't have anything on, compare info that returns a span, we only have API. Is it return string or that inspect spans.

E

We don't have any API is that taken a span, input, slice it and then return the slice value back to you right.

D

So returning a range would be much more would be matching what it's already doing, where we have methods like index of where it just gives you answers of. In this case, what it's returning is. Here's the range that doesn't include the thing that you wanted yeah, but yes, but given that nothing on it is currently returning a string. I think the range pattern definitely fits here.

A

So then hold on yeah, so it's a proposal than the one to remove prefix remove suffix on comparing, for they both take with only span of chars and compare options that is defaulting they both both return, arrange works for me.

D

Oh right, both is start an.

A

Interface that works for me, okay, so let me.

A

That is not annoying at all.

A

Jesus all I want to do is pay something in chess.

A

So it's probably horrendous sweetie, but we have to copy and paste it somewhere or not, as you read it, but that's the API. Shame.

C

What's wrong with dragging a notepad window over because.

A

I'm not sharing my screen, I'm sharing a window and I share a window because I have a vertical screen. That's like she had the whole window, I mean I, can't.

B

A

Super awkward to use so yeah I need to like my setup, is less than idea right now, so.

E

Feedback on what you pasted um I think we had said pong string, the remove start and remove and overloads that don't take a string comparison do not exist is in the only overloads that exists are those that take string comparison because we plan on introducing an fx, corporal or analyze or whatever it's called now that Flags calls to these api's. So it would be weird that we would introduce an API that we would immediately have an analyzer rule, saying don't call it.

A

E

A

I must have missed it, then. So, basically, the proposal to move started, move and like we don't have one. It doesn't take a string. Compare right.

E

For Steve string comparison to get Houston and on compare info have only stand overloads instead of string overloads or did I miss here. The recommendation earlier.

A

What so a general rule, and otherwise it's not usable than listening, which supports and physical inversions are.

E

Our general rule is you're right, have both but I think either Jeremy or Corey said. These would actually be the first string returning api's on compare info.

A

Let me take everything on string everything I'll compare to what he take strings right, I mean that see this yeah.

E

But they don't, they don't return them, because it's mainly inspection rather than generation right.

D

I mean remove, yeah, removes start on preparing folk and take both the read, only span of char and a string and both return range. They don't have to return the string. Oh.

A

You coupe made them range returning. Yeah I, don't mind that yeah.

D

I wouldn't have the span when returned range in the string would return string it just. It feels a little too asymmetrical for an overload just.

E

Wait till you see a read-only Spanish hard on split.

D

Since it has to return to an iterator yeah.

A

All right so then anything else.

E

Think we're good.

A

A

This is why I hate tabs.

A

Here we go: let's try this one more time. Yeah, that's readable, all right! One down.

A

Non-Native returning task completion source.

F

So this is me so when we first introduced tasks in.net four, we added this task: completion source of T type, which allows you to create a task of T that you then manually complete, rather than it's worth.

F

Mainly for simplicity, because it was already because tasks of t drives from tabs can so you could just use task completion source with object or something completed with a dummy value and then just use the resulting to ask 50 as the task. So what's actually.

F

On top of that time, even if me to have one implementation, why it's task completion source of non-generic? We have needed the commerce, the generic task, simply because the debilitation details we didn't have the means of creating the Manatee completed non-generic task. However, that's I hate that latter issue.

F

It's just confusing folks, don't know what to do. They want to create one of these. That's non-generic end up creating they don't realize they can create the generic one and then they create the generic one.

F

Don't know what type to use, because it doesn't matter, but they don't realize that, and then we end up with a lot of variation even in our own code, some of it uses bullets and some of it uses object, and so the proposal is here to just add the symmetry and basically just duplicate the generic API as non-generic there's a small perf benefit, and that everywhere we're currently doing this to create the generic will create the non-generic instead and we'll save basically on 64-bit, we'll say they bytes per object because we won't be carrying around and on an unused, unnecessary T.

C

Yeah I think it's good. It's.

F

Been needed for a while, so actually I already understand, like the implications of it, I already created a commit that basically replaces it across that never run time. So I can just go and do it.

E

I have a quick question: Steve um on the set exception methods we take exception, ienumerable of exception, should it also have an overload that takes an exception dispatch info I, don't recall if we've had this conversation before we've.

F

Had the conversation off and on I think there's actually a separate issue for it. I wouldn't add it as part of this issue. If we add it, we should add it to both the generic and the non-generic.

F

In fact, internally, I believe that that API exists, because we basically capture a dispatch info for each and then store those that's what's actually stored in the tasks. But to my knowledge, we've never really used exception dispatch info as a sort of an exchange type, so I'm I honestly have no problem doing so, but we should talk about it separately and think through. If there's any other places, we want to kind of propagate that idea. Yeah.

E

Sounds good and you're right, I don't want to randomize this. It just came upon this decision question.

F

Okay, then I guess that's it. I did have one one question actually on the API, so there's one place where we may want to entertain a difference purely in naming or leave.

B

F

As is for symmetry, and that is set result in try, set results now.

B

The one hand it's nice to.

F

Maintain symmetry, on the other hand, and also in the not really meant for public visibility. Api is on the async method builder. We still use set results, even though for a void, ASIC method. There is no result, so it's all set resolve. It doesn't take anything. The flip side is it's a little strange to see a method called set result or try set result that parameterless so.

A

D

Be set completed and try set completed, yeah.

A

D

Yeah, let's go with that.

F

D

F

Is a minor the naming on tasks in general has been a whirlwind task. Dot is completed, is inclusive of, however, the task ended. So it's he's completed it's true whether.

E

It's successful or not yeah so set completed successfully. Yeah successful.

D

Completion or whichever one matches the yeah.

F

I think it's it's completed successfully, so it would be tried, completed successfully or try complete successfully or something like that. I don't know or we leave it as set result in call it a day. I, probably.

A

Lean to one's dad, I mean I mean this is not an API that people who use if they don't know tasks right so.

C

I think result is slightly worse than tri-set completed, I think, like both kind of neither of them and ply the the state of completion, but, like result like what is the result in this case, it's.

E

C

E

Get a waiter get result.

F

Yeah, that's a good point about that that symmetry, which is weird as well the Oh, a waiter on task it. It follows the patterns. I would return, you do the methods called get results even though returns void.

A

B

A

I mean I would just probably leave it as as the symmetric one unless there's a strong reason to deviate: hey old ship it. Alright, then, let's leave it like this and then one more down hexadecimal strings and bytes. Oh I mean I, don't know these actually API yet, but I yeah I agree with the play.

E

I need to figure out where the actual proposal is in those who.

A

E

A

E

Marked it as ready, I think so,.

A

Deal fault then it's my.

E

Fault I think this menu right.

E

Yes, yes, Alan.

D

Amusingly, converting to and from hexadecimal characters is the thing that walks everyone through what it looks like for span and operation status, etc in FDG v3. So this definitely needs operation status methods because the work yeah, if.

C

We thought of a more general like to and from base I.

A

Mean realistically, what would you pass in besides 10 and 16.

D

Sixty to sixty four the problem, sixty-four is one of the last two characters.

B

D

E

Alpine area, so for for this, what we were really going for here was something that just kind of fit the existing patterns that convert to base64 string, convert from basic 64 string, already have and just kind of punting, the more advanced API that returns operation status, blah blah blah blah. Maybe.

A

Have them I'd only have a basics before time? What was it.

E

It might being it's either in core FX lab or it's in the system dot buffers somewhere, one of the two yeah.

A

I thought I thought in this system about house news busy already have a basically for time, I think yeah.

E

But it's it's really meant for like dreaming low allocation scenarios, and this is meant to be just like a dirt simple. It sits next to convert from base64 string like I I have something I'm reading from a file. It's a hex string call this API boom. Now it's a byte array.

E

Like this, this isn't intended to be the low allocation variant. It's intended to be the easy variant.

A

Yes, a system buffers the text, we have a base64 type. So if you were to do this, it would probably have a base 16 type in there. I didn't would follow whatever by 64 is done.

E

Yeah, if we wanted to add it, there I think that would probably be for a different issue, though not for this one yeah.

A

I mean I reviewed I, get all I'm saying that if you were to do what I wouldn't do it on convert I would I would I would put that next.

B

To basics, tier 4 and.

A

Then I would say these ones basically should matter whatever we did for you know from by 64 on convert. Yes,.

E

A

I think they should have very simple: API is, like you know, take string, returns, might erase or whatever.

E

A

E

What is proposed here, what's proposed here, is the easy stuff that already matches the pattern already exposed by convert I just.

A

Seems like a lot of API, it said I.

E

Convert to base64 string has a lot of overloads holy.

A

Do we need them all I mean that gave.

E

The only the only real difference here is that we're on we're not taking and options enum like base64 takes because base xt4 has concerns like do you want to split after 72 characters, blah blah blah hex generally doesn't have that same concern.

D

Yeah I mean there's a bunch of. Do you want I mean the problem with texturing? Is it's it's a big doghouse right like do you want to put colons between every pair or double colons or hyphens or spaces or new lines, or nothing or do you want to wrap after you know so many characters on the line or somebody bytes of input and like a super flexible one would have you know a trillion options.

D

E

And we and I think the the proposal for having like a hex convert class in the system. Buffers namespace is a fine place to put all those options, but I wouldn't want to simply if you have to worry about them, yeah.

D

And that's reasonable I'm I'm just pointing out yeah.

B

D

Complexities that come from two extreme cuz like right now, people usually do this I think with bit converter to string and that one puts hyphens I think because it was the style used by MAC addresses in Windows. You know.

B

D

It was designed for yeah that followed by remove to get the hyphens out, and we.

E

Actually discuss this a little bit in the issue here, where there are all kinds of different scenarios for formatting like that, like if I'm a hex editor, for instance, I want a space between every other byte, except for at the end of the line like I want no more than 64 bytes per line, except for the beginning of the line I wanted to have, instead of a space like random stuff like that, so it's it gets very complex, but the simple scenario of I just have like 8 or 16 bytes and I just want to turn them into a hex string.

E

Most people don't want that to have any type of special formatting I.

G

A

E

A

Your hex editor, it's okay! If you implement that method yourself, it's like whatever, whatever you need right, I mean it's. This method doesn't do that much to begin with right. It's just. It's meant to do the simple thing so that you don't have to write your code yourself, but so.

B

The existing API is on convert that do conversion they take a base either to or from, and they that base is allowed to be two eight ten or sixteen. Would it be better to mirror that here and just make this a general convert array of to target base and support the ones we already support, I think.

E

That would be confusing for people to use, because I I think they wouldn't understand what parameters they would have to pass in in order to just do simple, like hex device conversion.

D

I

D

I

D

Do that we want an enum.

I

Just so, it tells people what things we actually support. Why would that be confusing, though, because presumably.

B

It's it's the same API you take a byte array in or even a t array supporting the current parameter times your already support and the base. So it's just like the current API it just instead of taking a single value, it takes multiple, but.

E

That's the thing right now now we're increasing the concept count just to do a simple operation of convert text to bytes for that right.

B

But at the same time, we don't currently have things named text today. We just have ones that take base and if we ever expand this in the future, because we have hex we're going to have to add binary decimal.

B

D

B

Names as they come up, I.

A

Mean I'm fine, we're just calling the method differently. I mean we could just call it base 16 right. That seems fine I.

E

Think it would not be discoverable, though yeah.

D

I mean the real question is like I know that we all know that base 16 and hexadecimal may the same thing but like if you're in, like I, just wonder like cs1. If it occurs to you at all that hexadecimal in base 16 or synonymous.

H

D

B

And they should update their teaching.

E

Like I, do we we need a more powerful method to do all the different conversions that you mentioned, tenor and.

B

At the very least, we do have base to base 64 today so to base 16 with parity that yeah.

E

But hex doesn't really call base. 16 I mean it, it is base 16, but it's you don't really refer to it as such colloquially. No.

B

But at least it fits with the other API names. I mean.

A

He's another way to think about this, so let's say eventually we will add, like a base. Api that allows you to specify the base number right will be actually implemented as such or what we actually just say if base equals. This call this method if base equals that call that method, because if we do the ladder, then quite frankly, I don't care, because then you could just say yeah. We have an octal one.

A

We have a hex one and we have a whatever fancy things you want to have and then maybe that's the one we dispatch to, in which case it doesn't really matter. I.

E

Don't quite know what overloads a generic base one would have, because when you're talk about page 16 base, 64 octal, whatever they all have their different use cases, meaning people often expect them to be formatted differently right. Oh.

D

Yeah yeah cuz, we've the you know a byte array to hexadecimal or I guess a byte. We don't. We span a byte to hexadecimal you. You want the leading zeros because you're looking for around triple number. What is it like? I mean you can figure out the same sort of always make a fixed-width printable value with you know, 2 to base 10, but like that they don't have a good alignment. So it's it's hard to actually say what such a policy would be, but I I can agree with that.

D

We just take that all the methods we have that say to base64 and we add a version that says to base 16, perhaps leaving off the do we wrap after 64 characters and yeah I I do believe it will be harder to discover.

D

But it's like that if people are looking for hex they're looking for the word hacks and not the word or not, the number 16 is my gut, but it does allow for. We could decide at the same time that we want to do, convert to base 8 and because now really what we're doing is we have convert to base and then all the numbers are just like there.

D

So there's semi overloads at that point, instead of once hex and once octal and ones base64 and so I I think it puts us in a better place to have them all be to base and then a number the.

E

The only concern I really have what that is with the overloads would be different, but maybe that's acceptable. Well,.

D

They would just lose.

D

Like convert to base 16 string would not take a base, 64 formatting options, because that doesn't make sense for base 16 all.

E

D

It says 64 in the name and so.

E

Would we have a base, 16, formatting options and really all it would have as a lower case or not I mean symmetry.

D

I mean that's, I mean.

A

If you wanted lower case, you can just call our kids afterwards right now. If you really insist on that.

I

E

But why am I typing two lines when the API could very easily just have it be a one-liner by taking the correct parameter in the overload.

A

Simplicity I mean like the I'm already not gonna, know about having you know nine methods here to do a from hex conversion, so I don't know having a bunch of all those. If it's funny that.

E

We're talking about simplicity at the same time, making it less discoverable well.

A

Yeah, like I, mean well.

B

And and honestly like this covers the very simple case of I could have just written for Kimber dot to convert, to string value and then sixteen because that's the existing API and you would write it as a for loop and and and if you want to add, spacing or anything else or any other custom formatting. It's just a tweak of that existing algorithm. So.

E

The issue is that people get that wrong today, because what they'll do is they'll call bike, dot, they'll. Basically, a they'll call to string passing an X modifier but forget to write x two, and that means that if the value is less than sixteen, it only gets written as one character which corrupts the end result. Yeah I mean well yeah it really. It really is like a very frequent thing to get wrong. Yeah I think I see that.

D

On Stack Overflow, like every other month, I.

A

Mean to me, like I, mean if you just look at usage data for the existing basic C for API as there's two ApS is going to be used from hex string, taking a string and two hex string taking the by the way. These are really the two API is ever used by far the most frequent and the rest is just all in the 1% in less so I. Don't even think that we need to add them or, quite frankly, like.

A

Even then I mean it's just. This is the usage right because, as you as Tanner said right, if you really care about buffers or other stuff, where you really do it up to my stuff, I'm, not even sure you would use this API have to begin with no.

E

But this isn't meant for those scenarios. This is meant for people. This is meant for what I call the nine-to-five developer. They're, like you literally just want the right go home and enjoy the rest of you know.

A

I get that. Why would you not just do from X string to hex string and you was the bytearray and let's do whatever you need to do with you by the railing? Why.

B

Would you do if.

A

We could spend also.

B

Add an analyzer that Flags you did X without specifying a count.

A

But it's not always wrong with. It depends on what you're doing I'd.

B

Say more often than not when you're formatting is hex, you want it as X, 2, 4 or 8 I.

A

Mean sure, let's say for a second: we do that right, I mean I. Think you guys point still stands right. If, if I have a hex string like, why would I have to go to stack overflow copy and paste it in and you know not in line but dip 8 light, snippet or whatever it is I mean come.

B

On I, don't you know how to write a for loop.

E

Sure yeah, it's not it's not necessarily that, but it's all of the it's putting all of the concepts together. If you get any one of those concepts wrong, then the entire thing kind of falls apart, perhaps and undiscoverable fashion yeah, the example I gave earlier of using X instead of X 2, is one such example. So.

B

My view is I think it would be useful, but I don't think this is like necessary, Larry necessarily fleshed out enough. It seems like we're trying to cover the super. Simple scenario of I just want bytes, converted to hex all mashed together, yeah, but realistically, what we probably want is a more general I want to convert to base and maybe I want to insert a separator character or control casing.

B

A

In my I've not seen any evidence for this, like I mean like the evidence that we see and I see this on Twitter at least you know once every three months that people complain about the lack of an hex, API I have not heard somebody saying, oh, my god, I wish I could do octal or you know binary so I think like it's. It's clear that people won't pay 16, but the question is: could we generalize this in the future?

A

Yes, we could I'm not opposed doing that, but I also, don't think we need to do it now. I will also say be very often over engineer just because we think we you know, should do it, but then the question is also like. Is it worth it right and I?

A

In this case it seems we couldn't need an API that does in a hex conversion, just because everybody's asking for this for like 10 years right, and we could do that with just two methods and then see what we are missing in terms of features and if you have to grow the method to support more scenarios. We know how to do it. I mean basics. Before he has a bunch of overload. Yeah I think.

D

We get three methods but yeah sure.

E

Yeah and I I think we I think we should actually look into what header was suggesting. Honestly I I would probably have it as a separate proposal, because I think is more generalizable useful right.

B

E

If you, if you wanted to investigate that, Tanner like I, would absolutely be a proponent of that.

J

Maybe simple things.

B

J

If they are optional right and maybe empty string by default or empty.

B

Maybe it's something that we could start a prototype in core FX lab there. There's probably some people on the C sharp discord that would be interested in prayer typing something yeah.

E

Just keep keep in mind that if it's not one line to call it, people are going to get confused.

D

E

D

We want to keep this one simple and say that we want to that. We want to do sort of what we did for base64 when we were adding span of that. We make. You know a hexadecimal utility type that has the operation status, things and- and you know everything else under the Sun and that it takes a bunch of options.

D

Then I think that here we could get away with from either from hex string or from base 16 that takes string and overload that with raw char and then the to base 16/2 hexadecimal, which takes byte array with no offset in length and one that takes Ross, byte and I think that those four are gonna, be probably the 99% use case.

E

Do do you think we should have try methods or not.

D

Given that the current that the tri from base64 doesn't follow good, try rules, I'd be slightly encouraged, they're slightly inclined to say no, because it's a it's already a bad one. For.

E

The two methods would you have as proposed here: boo lower case equals false, or are you proposing new options class or nothing at all? Did you always get up if.

D

We want to be super simple, then it's just upper case because I I believe that's what people will want more and whoever proposed this did too because they made it the default and then, if we wanted it to to be fancier than I, think we would add base-16 formatting options which would have lower cases of flag and then arbitrary cutoff points as a flag and maybe insert Heights like I, would go from nothing to the enum. I wouldn't go or nothing to the flags.

D

Enum I wouldn't go I, wouldn't do a bool, because we certainly see that there's room for expansion, yeah.

C

In particular, though,.

D

From 64 we know that there's hey I want this chopped and from this proposal we already know that there's lower case, and so that's that's already enough. That says we should just squish it to a Flags.

E

So, does that mean that this is needs more work? No.

D

I think we just massively Samara I, think we're massively simplifying this one and saying for the fancy stuff. We would need another proposal for as I mean your suggestion was we needed another proposal for the hexadecimal class to mirror the base64 class or base 16 class if we prefer sure.

A

Do the so I haven't looked at, but I think the basic C for class probably only has operation status. Ap is right, so they're, probably a pain in the ass to use yeah.

E

They're they're really meant for, like buffering streaming scenarios there they're about something that you can use in one line of code right.

D

I mean we can certainly, we can add the simple things on top it on the same class, because once you have the operation status, API that that's the basis of every simpler form of span buffer or of span API. Also.

A

Fun fact: the basics, if I want only hand with UTA.

E

Utf-8 has no concept in base64.

D

Well, I think it's he's saying: we only took bytes and that chars well.

A

The API is actually called decode from utf-8 and encode to utf-8, so we don't even have any sixteen one. So I have I have tried antic heartache with those names, so yeah.

E

You don't do what they say they do, but.

A

Yeah yeah all I'm saying is that I think if you bear with this, then we make the they would follow us of the world happy. But we don't necessarily make somebody who want slightly more features: happy yeah, I,.

E

Would suggest stayin away from those names because and honestly on those api's utf-8 is a buzz word, but they don't actually implement proper utf-8. Well, it's it's literally. There was a buzz word, but.

D

I think that they're, throwing or otherwise going into their fail mode as soon as they see something that wasn't a base64 character or white space. So yes,.

E

They're not gonna handle utf-8 whitespace, so they don't do white space detection correctly for utf-8. That's a big problem.

A

Fair enough, alright, so then let me let me rephrase this so for this one I think probably given that this is like probably also five years old, I'll, be inclined to say: let's approve the subset that we they we all agree with, and then that doesn't mean that we can't add more API. So it just means that we're not ready to do this quite yet.

A

Did someone just fall out of the window or that.

B

H

A

Let me paste this guy in.

B

A

I guess I'm tree ash doesn't make a little sense right now.

E

Also feel free to unassigned me and mark it up for grabs I.

A

Will do that too? So make sure does this might not before forget, so you said: I'd assign you and mark it up for grabs, yeah.

E

And unless someone here wants to take it I think someone in the community actually already did the bulk of the work I mean.

D

I have these in all of the crypto test projects.

A

So shape so what did I miss something.

E

I think Jeremy you had said: there's no need for offset in length, but I mean.

D

I'm it's if we wanted to do minimum total minimum, then we would say if you wanted offset in length use this band, but because it certainly is there all. Our implementations are going to be that they all null checks, then defer. So yeah.

A

I don't mind them because it's a it's like 20 a pattern so we're gonna have a ton of those. So that seems right.

D

So the only question was: do we want hex string or base 16 right.

H

D

I agree with both of them for different reasons, so I I, don't know how to I, don't know which one to land on so I'll agree with the majority yeah.

B

I think that hex strings, probably more discoverable but base 16, will have parity with the existing API, so we expose and with ones we might expose in the future.

E

If we expose octal in the future, you think it would be called base 8 instead of Aqil I.

B

Think that depends if we exposed base 60, would we call it what what is it second said? Second,.

E

B

E

Base 60 like honestly in wide enough use that we would expose it as a first class citizen here. Definitely.

B

D

160 I know 62.

I

D

I

And numbers 60 sometimes used for trig.

I

A

Mean I can tell you that she had pretty much universally prefer s. Hex I mean.

B

A

Don't have strong opinions. I can see the argument for either one I will say if we only have hex and we never add the other ones. That hex seems fine, if we add all the other ones soon, then that might sound weird, but.

E

We're not into them for the.net 5 time frame. Why not.

B

Both and we can just have them forward to each well.

A

That's the microscope approach like I when I was an intern 11 years ago. We had this joke because Apple at that point announced it. The new hard drives are being to be base 10 indicators rather than base 2 indicators, and we had the joke that Microsoft would never do that. We would. We would basically have a wizard the first time you asked for the disk capacity and then it would also give you reg key so that you can't configure any other base. You want, because we can't decide right and that's another example like know.

A

Well, let's make a decision one way or the other. Let's not do both options. That seems worse.

F

G

D

But then the person who wrote the feature would disagree with the fact that I T could control their machine, so they would have the override to group policy. That would be a they would try and bury the documentation somewhere so that it wasn't fun documented yeah.

E

This makes me want a group policy configure default parameters now.

D

So I mean I, do think yeah if we added binary and octal, and we felt that we were doing name spread and then we went. uh Why didn't we just name it base 16? We could add the alias at that time. I! Don't think that we have enough reason to add it like if we were gonna, add a base 16 alias now, I, don't think we would add text so I think we should do hex now and if we later wish me to added it as base 16, then we can.

D

We can have the duplicate names, yeah.

E

And I, to be honest, I seriously doubt we would ever out binary or octal in the future. There's just not enough momentum behind it, but I'm happy to be proven wrong with evidence. Yeah.

D

I mean the only thing I know of that uses. Octal is mode bits, no.

E

uh Maybe maybe you want to pass compatible Jason's on net payloads to your application.

D

So yeah I'm fine, with using the abbreviation hex instead of hexadecimal and etc. This is it's what people want?

D

A

An API for the people, all right, let's go with this.

G

A

G

D

It if you didn't.

G

A

You prove the disease right. We.

D

Approved it as proposed, but oh you did ender get perinatal, okay, yeah, so that shape seems 99% use case.

A

All right, CCP, fast, open implementation. That seems like a big item. Oh no, it's a in a member, never mind!

A

Corey! Are you still with us I.

E

J

This basically the biggest issue here was our crossbar portability and I. Think there was some research trying to figure out if this could be done on all supported platforms.

J

So it's basically an option for the set socket options where the operating systems have some capabilities, which we do not exposed via the socket API I.

D

Mean if we had done this seven years ago, then we would just be asking ourselves the question of what should we do in those pals and and the answer is probably throw I feel like in sockets, we've been trying or we've been getting more and more we're just exposing stuff, and you may get a symmetry than the rest of the framework where we're trying to avoid things that we'd have to PN se I mean I could be wrong, but that's the feeling that I have so. This seems fine to me. Yeah.

A

I mean I have zero opinion on the implementation side, but, like the Eno member, it says trivial other changes. It can imagine.

D

Whether whether Mac throws PN SC like if you can't, if it's because I don't know enough about what fast open means, but if it's something that someone can't really detect that, then it can just ignore it until Mac supports it and if it's, if it completely and fundamentally changes the nature of the connection, then PN SE and says I'm. Sorry, you have to you, have to write a platform check now. Yeah.

J

In most cases, the platform D text, what's right and I guess there are rare cases when somebody wants to override it. I think the caveat right now will be the phone's I think we tend to do research on the normal API, so normal open systems and better. This would work on. Let's say: I watch, I I, don't think we know. Hopefully that would flow the mac OS and again it seems generally.

D

I'm saying like, if, if this is, if the fast open thing is, if that the, if you don't use fast open, when you call connect, then the server is going to fail and saying hey, I, don't actually support. This option would be good if it's. This is a I think that this is a perfume. If you can do it, then it's a suggestion and the OS didn't respect it. Yeah.

J

It's more like that, it's it's a optimization yeah the allows you to send more data faster, essentially yeah.

F

So so I have a question on that, so the example above basically shows doing setting a buffer and then initiating the connect a sink so that the assumption there is that the connect is going.

H

F

In opening the connection and sending the buffer correct.

J

That's what Cory was looking at years so.

F

That's not just an optimization, that's a functional behavior that if you have fast open, set and you've set up buffer and then do a connect that it's going to send the buffer so we'd need to. We would need to either ensure we simulate that behavior of fast open itself wasn't supported by doing the connect and then doing the send or we should throw P.

B

Mse because otherwise it.

F

Is a difference of behavior and when the buffer is sent or not yeah.

J

That makes sense, yeah yeah.

D

So and I think that that'll, presumably if that really is just you know a wait, connect a wait. Send then that's easy enough to have a software fallback for and if the OS supports it, and now you can do the fast.

A

What are we doing another you know, members like ipv6-only, presumably would throw pn is e. If you can't do I PB 6 right.

A

So, in other words, it wouldn't be the first thing a member very might to know for yeah.

F

No, we got very definitely so Jeremy's points. The only difference is most of those we inherited from a pre cross-platform world and we're now she's now being more thoughtful about it, as we add them in a world where we're you know, targeting multiple platforms from the get-go should.

H

D

Yeah, so it sounds like there's an easy software fallback. If there's a behavioral difference, we're the software fall back or PN SC and probably the fall back is nicer.

D

Where then it becomes a do this thing, if you can, if, if we think that there's a semantic difference between them, then it would be pianist E, but that's now just an implementation detail again: I didn't look.

J

Myself, but from a conversation we had with the networking team, it seems it should be doable for all the platforms we support. Currently so but I agree, we again, we can put it to the notes or it could be our discussion on a PR me and.

D

I mean just because we approve it here doesn't mean they actually have to implement it.

A

All that works here, in fact, that's another scan. We should probably do how many API is heavy him approved versus how many have you actually done.

F

Some notion of while we approve that API three years ago, like we withdraw our approval.

D

We've learned things since then: you haven't implemented it. We want to change it now or change. Our mind is.

A

It a breaking change.

A

Nobody can observe in.

F

E

So, just to confirm this was not milestone to fiber. This was milestone to future I.

A

Don't know yeah.

J

I, probably should be future I mean we could do it if it's small or if somebody picks it up and I. Believe Corey was working on prototyping it, but I don't think we would hold 500 for 8 right. So I.

F

Imagine Tomas Tomas would only pick it up himself. I would imagine it was going to impact one of the other things he's working on like if it was gonna have a notable impact on SSL streams. Authentication. You know, then maybe you.

J

Think you'd be HTTP, probably very key, like blast the data, while you still got a connecting right, then.

J

A

B

A

Good then, let's go to the next one.

A

Stephen you Marcus is ready for view 15 days ago.

F

Because Shai had encouraged the contributor from the community to submit.

B

F

They did and yeah.

A

So, what's the API that we proposed here.

D

So this is API ready for review, because it's PR blocking.

F

Well, actually, I closed the PR, but I do the PR to get two scales. They did a lot of work on it all right, Shia, basically,.

F

Suggested that they open a PR and as a result, they figured out all the api's. They thought were valuable and I think I agreed and then they nicely added that to the issue.

D

Looks like two interfaces are getting modified.

D

Which see every other discussion we've had about that we're scared to use them and that we can't modify interfaces.

A

Yeah, that seems like a larger API. Then we probably should have somebody from the data team. There.

A

That just don't strike me as ready without context yeah.

A

Also seems to be a decently sized API Edition right.

A

Okay, so let me just update the the comment where the person kind of place to the API is because I don't find them super readable but other than that seems like shy, or somebody else has to represent this.

D

Yeah- and it looks like what the PR did was just add clothes, a sink and open a sink to ID be connection, so it that was the breaking change. Instead of the we don't know what it means to use. Dims change.

H

D

I also, don't know that you could add this with dims. Like you can't I mean you. Could you could say that the clothes they sink just as the synchronous, clothes and returns tests that completed tasks, but that's probably not good so so I I, don't know that there is a sensible path forward on this, but at least for the close and open or execute reader or essentially it's they used interfaces. So I, don't I think that they I mean the concrete types can do it, but I don't know that the interfaces can I mean.

A

We already kind of said the interfaces are dead right. It's just people keep using them.

D

Right so maybe the answer is we just trim the the interface changes off, but BAM data should we'll probably have questions, and this is big.

A

Yeah, it seems like this to me.

D

And looking at the like, it's adding three or four virtual fil-a sinks, the public one doesn't take a cancellation token, so yeah I, don't I. Don't think this is ready.

A

Well, I mean it's I mean it's ready to review, probably both the author right but like provide them providing context. I'm, not sure, there's much of a point right.

E

There we can, we can just bring them in in a future date and have them go over yeah.

H

All right, I think I got this one right.

A

Only put spell the word interface.

G

A

A

Think Stephen, it's a public comment on this one task when any task, when all I guess it just thoughts with any, do.

E

It sorry, let me look. Oh.

F

This one sorry meeting myself you that we currently have an overload of task one any that takes uh params.

C

F

If you look at all uses of tasks that, when any in save dot, Nick or or even our internal index, like almost literally every single usage, takes just two tasks and so every use of just paying not only for the array allocation at the call site, but then we make a defensive copy, but we worry used to make a defensive copy internally. Now we, as a month or two ago, we have an internal overload that takes two tasks, so we special cape it. You just call that one.

F

This is just saying: let's make a public and avoid the params reality.

D

Seems legit it's the same as we have on the seven overloads of string, dot, concat.

J

Cases like your network and you either send or receive me you try to do either on. So that's really a lot and I would echo what keep said. I think it'd be nice.

A

Seems good to me in them.

F

So I think I think we should do this. There is a larger sort of question to ask as well not not to be reviewed today, but I just want to raise it that a lot of the call sites, not all of them, but let's say half the second task is from task delay, where it's being used to do a time out of the first tasks.

F

So there's another issue: that's open about sort of making that first class thing where he basically can wait a task with a timeout, and so it's probably worth looking at at some point is how we want to actually expose that functionality.

A

Make sense, but you would still approve this one as it is then I suppose yeah.

F

Yeah I think we should just make it public.

H

All right so then refresh this one here.

A

So then, this is what it is. Maybe should be pretty straightforward.

D

And I think that matches the recommended naming for this pattern seems good.

A

Tanner I think you probably the best person to comment on this. This.

B

Is one of the ones where we need a dedicated review and I?

B

Think it's on your list of things that needs a dedicated review, I! Think.

A

That's probably true: what would the reason I haven't looked at? It is because it's not mark looking. So let me fix this as well. It's.

B

Not blocking it's a nice-to-have thing so.

A

You wouldn't do it for 500, then.

B

If it was approved, I've got people who are willing to implement it and I would be able to review it, but it is not a strictly required thing for 500. You.

D

Do currently have it scope this 500.

B

That should be milestone future. Let's do that if we can get it on the backlog. I don't think it's that complex to approve it's essentially the existing API that we already have, except on new types that are generic. That way, we can expand it in the future without having to come back to API review for half and int, and you went et cetera.

A

Yeah this seems pretty much like a pattern based kind of thing where we look with one of them and then, if you say that's fine, then we just we.

B

Still probably want a dedicated review session, though just because it of how big it is, and there were some design decisions that we may want to tweak from the original to this new API, such as public members being exposed in the original right.

D

So do you think it's a half slot.

I

Or a full two hours, I.

B

Would hope we could get it through in a half slot.

A

Yeah- let's let's shoot for this for next week, then because I mean this. One was also open for three years now so I mean given that we just walked the backlog, I'm I'm, okay, with marking. This is blocking at this point and he was getting it done.

A

Because otherwise they will always stop all four parties in probation right right all right. So let me do this. I will invite yarn the JIT folks and you obviously, and then that should be it right.

B

Yeah I think so.

B

And you said we would get to it next week then yeah.

A

I would I mean I would just set up because right now we don't have any other big items. So it's all backlogs but I like given. If you have 60 open issues, I just want to drive this down. Okay,.

B

Sounds good would that be Tuesday or Thursday, just so I know to be ready. I, probably.

A

Would do it Tuesday unless you know it doesn't really matter like whatever, but then, whenever you're ready, yeah.

B

No I Tuesday should be fine.

A

All right, then, let's do that and then next one.

B

This one is actually ready for review as.

D

In like right now review yes,.

B

It's actually it's this one. Unless someone complains about something should be a five minute review. Uh-Huh.

H

G

E

D

B

D

Copy and replace half basically.

B

Because it's a user defined type, there are some operators for dotnet five. The operators exposed are limited specifically to converting to and from double and float and and doing the relational comparisons. Other operators, we don't think, are needed for dotnet five. They may be added in the future.

A

H

A

D

Família has so you don't have.

B

An addition and subtraction correct the people who currently need this are people like a tensor, tensorflow, Sharpe, ml, dotnet, etc and they're going to be using it as an interchange type. That is, the data is stored on disk as half and they need to convert it to single where they will do their operations and then they will convert it back before. Storing the entire purpose of the type as an interchange format is to save space, because it's half the size of float. Okay,.

D

And are we doing it as net 5? Only are we shoving this in a compact pack somewhere.

B

The as currently proposed would be only I'm not aware of anyone that, once it is compact pack, ah I.

A

Mean there will be almost orthogonal right, I mean my mother question, though, is like: will he ever have language support for that, so.

B

We we had the discussion with Mads and everyone else. There may be language support in some future version. There won't be today there would be a risk of a small breaking change if any operators are exposed, just because implicit operators or language defined operators work slightly differently from user defined operators, not that the language would be defining them, it would work like decimal and it would just call the framework implementation, but there would be language support, and so you end up with weird implicit conversion cases.

B

The breaking change would essentially be the same as what we just took for in int, an int pointer, though, which wouldn't be big at all. Yeah.

A

I think I I'm not opposed to those. It's just that in the past that a language team is always pushed back when we did these things. So as.

B

A

As they're happy with it, I am happy with that, because it would be very bad if a ship an API know of operators, and that makes it impossible for us to do language support later so long buying in with that seems fine. No.

D

I think that's why he left the operators out well.

A

Right, there's a writer's with like less than double equals right and so those right, but.

B

We had the discussion with with Neil Chuck and Mads, and we got by off on this proposed surface, and we we have notes on if we're going to expose the others in the future. What things we need to coordinate with the language team and have exposed at the same time that.

A

Makes sense to me then so.

D

Can we scroll down and look at the operators just to make sure that we're implicit and explicit in the right places they're.

B

All explicit implicit would be a huge breaking change for the language to implement in the future. Yeah.

A

Because now you have something in the middle that is implicitly convertible right, yeah.

D

A

D

This implementation of half will require a visible casts to go to float, but in the future the language could add an implicit conversion that would, as part of the implicit numeric conversions list, correct. Okay, so, okay, so.

B

So this is, this is the same same surface, a system, single system, double system, single and double, don't define any operators because they're all language and runtime implemented these ones expose just the relational and conversion to and from floating double and.

D

I assume they're going to expose them in the same way that the language would if the language had support. Yes, because I, you have made me slightly painfully aware that these types are actually way more complicated than one would imagine less than is not an obvious answer.

B

This will implement the I Triple E, compliant implementation based on the berkeley, soft load implementation, which we, which is bsd license, and we already have usages of elsewhere.

B

D

A

Do any other conversions.

B

A

I mean usually, we have like convert to somewhere right, because right now you only have the type. So how does it.

B

Convertible and I convertible doesn't necessarily support anything other than a strict number of types today. Well.

A

I mean I convertible. Has the generic fallback right away? You have convert to type kind of thing, but, like you can still have static methods on the convert class right. That would prevent us from doing that. It would just mean you would be able to do it necessarily yri convertible, but that seems alright. Fine.

D

B

I think those are out of scope for the usages we have for dotnet 5 yeah.

A

I mean yeah I mean to me the only reason there is like if we, if we would have them, then moving them is harder. I mean the type is obviously easier. I'm just wondering like what's the tail of this, because usually when you add a type you want to actually accept the somebody usually and then the.

B

The one usage which which we have a separate proposal for would be with intrinsic api's.

D

Yeah, if this is in then I think it'll come up when we talk about sea bore because seaboard does have a specific I'm carrying a and I Triple E half.

A

How does one construct one right now only through parsing through.

B

Parsing or converting from floater double.

D

And since this is mixing a user-defined operator and language operators can I say, can I do an explicit from the number five or do I have to write 5.0 I believe you can do.

I

Explicit from five okay.

D

So we don't need, we don't need, intend long and short and okay.

A

Yeah I think sous-chef has made. Rules are on that, but yes.

B

Right, which is where you get into the weird breaking change edge cases if they were implicit right.

D

Because literals are evaluated very, very late in the game and what about? If it's, a typed int, will the typed in to get Auto converted to double if it sees the half operator or will it make someone do the cast I believe.

B

It will also work there: okay, I.

D

Mean if we're getting the type in mainly so that it can be used for you know, store and retrieve /, just getting the surface here so that we're prepared for the future that then, if it requires some extra casts right now, that's probably still fine I'm, just making sure we've thought about the scenarios. Yeah.

A

And so Nikki I mean I trust you the scope, it's double, but are there any other special methods? You headed.

B

No, this is this currently just mirrors what we have for floating, double okay,.

I

The structure should probably.

B

Be declared read-only, which it wasn't yeah it's going to be read-only and it will contain a single field of type you short or short, depending on the implementation, and then.

D

I assumed that this was anywhere that said, double D, it now says half H and all the string s is are as they are in the existing types and that we shouldn't care too much about the parameter names here. Right.

B

They should match what we've already exposed: okay.

B

And if they don't, then we should ensure they do during implementation.

A

Cool I mean like unless there's any weather anyway, any other concerns around this Jesus, then I think I'm, fine. That.

D

Yeah I'm pretty sure the answer is yes, but I'm gonna ask it just because I've been looking at too much lately, we we do call the type system, single and system double right, correct.

B

D

B

That was one question: should this be system numerix, half or system, half I.

A

Mean I can personally go either way, quite frankly, like I, don't I know that people hate dumping into system namespace, but numeric.

I

A

Primitive types seem somewhat fine, I would.

D

Say if, if the language team has said no way, will this ever get language support, then numerix is better and if they've said it's not a priority, but we're not opposed to it, then being in system with its sibling type seems better. Yeah.

B

It its they've, they have not said no way, there's just not enough data today and it's not currently a business critical priority. Hence why it's on the backlog. So my vote is for system because it lives with its siblings. It is an I Triple, E defined data type. There is hardware support for this on multiple platforms, etc. I.

D

Yeah, if the language team has not said no, never, which I'm aware that even though never can get pulled back on, but if it's not no, never then system is the better namespace for this I.

A

Mean it is more of a fringe etai, but so are a lot of a number of the other types you have in the system. Namespace right I mean yeah.

D

But emails think of the poor book author, who one day is going to forget to write numerix in the word half.

G

A

Your parrot eyes.

G

Think so don't.

A

Make me update the book: is the next place you're going to use all right, then? Maybe just copy and paste. This whole thing change eight characters, so I added value today, I.

D

Think you're deleting nine so.

H

D

G

D

Of the dot as well smarty.

G

A

D

It's the comedic value that I bring to this group.

A

D

A

As your video is not purely kinetic, it's good.

A

All right system.

D

And so, while emos typing things Tanner I'm simply curious, is there a loss of there's no loss of precision if you convert half into double, do math and go back right, but you might get different rounding errors than you would have used the processor instruction directly.

B

Actually so the way it works is and there's a formal proof on this. If you use, if you do floating-point operations that start in precision X, you do the you convert to precision Y where Y is at least twice the number of bits of X do the operation and then convert back. There is no double rounding error, so doing floats as doubles and converting back is fine. Doing doubles as quads and converting back is fine. Doing doubles as a tea bit.

B

Precision and back is not fine and you have to run back to the original precision after each operation. If you do two sets of operations and then round, then you risk double rounding error. Okay,.

D

So anybody who wants to do math with these right now they can't use until we get to your other proposal, where you're, adding the intrinsics and will this work with vector automatically no.

B

It will not work with vector automatically that support is not slated for five. Oh it's a nice to have it sort, not block anyone wanting to use these types etc, but.

D

Right, okay, so so, essentially, people are gonna pay, the the widening cost of copying it from memory into a fat register, but otherwise they're not blocked from doing math, correct.

B

They can still do I, Triple, E compliant bath with half.

G

A

Public should have headed c-sharp there. We don't need any methods on and math either for this no.

B

That would be a huge amount of work that couldn't be completed for every.

D

B

Yeah, it would be a huge amount of work that couldn't be completed for five Oh bleep, namely because the B current implementations of math called at the C runtime, and there is no half implementation in most C runtimes for D math methods, plus.

D

I think, based on the fact that we have math and math F now it would be adding.

I

A new type called math, H or.

B

We'd have to further discuss the separate proposal, which is move all the math types onto our math methods on to the types themselves. So we don't have this problem moving forward. Yeah.

F

A

Right so then this is it all right, then we can do a few more things in ten methods and ten minutes.

E

Yeah this was a this was for symmetry with um comparator T, as Yun pointed out well or is the pasted in yes.

D

Yan just did the before you could move issues. Yes,.

E

The person who copies and pastes is the one who gets the credit, I guess yeah.

A

I mean I am fine. With that, I mean the only reason I did the comb, the other one, a semi we ship. The comparison type item you have API is to take comparison. Take comparisons.

A

We don't really have that many API is to take a func of t4 equals or func of t forget hash code right, so yeah I mean yeah.

D

But I think that this is the man I wish. I could use generic inference, which you can't use on a constructor.

A

Well, keep in mind that these are different things, so you basically compare of T and equality of compare of TR types you have to inherit from right and comparison of T's and it's a delegate. So it's just converting basically from the delegate some text to the to the abstract base type syntax. If you will and basically quality compare, doesn't have that notion right because there is no such thing as a delegate based quality. Compare this.

D

A

That's much a bit weird so.

D

This would be making a new non public equality, compare value that calls the delegates yeah.

A

I think that's what would happen yeah and.

G

A

Me it's kind of like mother post to it, but yeah I am also non I mean once you have two different methods. You know deriving from the type seems fine, the problem of the first one really is you have an instance of comparison right in the constructor right, because you have an overload that took a comparison. I have to call something else that took a compare and all you.

B

A

Get the compare for that.

A

So that's why I think I'm not opposed, but I also don't see it as like. Oh my god finally be ever.

D

Mm-Hmm, could it take because we have an interface right, I? Guess it's a quality. Compare is the IE quality? Compare so it's it. So we don't have comparison so never mind.

A

Yeah, we only have interfaces for the types themselves right like when you compare yourself to somebody else. The compares are always to take the other guys and they dare not interfaces they're, always.

A

I mean people seem to like it funny enough, but like I I don't see, have it I never had the need.

A

But, given that some people said they had this issue multiple times, I'm fine with giving it to them, I mean it's because I didn't have, the meat doesn't meets. Other people didn't right.

E

E

I'm, looking through the issue, description- and it looks like um it looks like the scenario is: I have a type like a person and I just want to create a one-off equality. Compare that compares like parts of the person like maybe I, only care about name, as opposed to like name agent address that a normal person, equality, compare, would care about.

D

Right but then you just say: like you know, you have your your internal or nested type somewhere, that is a person named comparer, which is person compared of T, and so what's because it they have Oh cuz funk is the bull is. Was it what it's returning never mind? I'm like they have a three parameter funk here like never mind, I.

E

Mean yeah the sample code that they have posted, wouldn't even compile with the API, so they proposed, which is confusing me.

E

We may be an appropriate answer, would be kick it back and say like what is the actual scenario for this, like update the sample code to compile.

H

E

Oh I see never mind that the sample code was actually for a completely different API proposal.

D

Yeah I mean it.

D

The the last message here of like I, just want to use distinct with a custom filter, seems closer to the right answer. I mean the biggest. The big problem is you. You do have to write both the get hash code and thing is we're making you pass both, but it does feel like it's way easier to it's easy to be wrong versus. If you're, looking at a class that you it feels like you'd, think about it more than a lambda, but that's yeah. It sure I mean.

A

Get hash code now is no longer hard to implement, but you can just say hash code combined XYZ right, so it seems that I can see people being successful. In doing that, I mean before I would have said. There's no way you can implement, get hash code in a lambda and not have bugs, but now it seems pretty straightforward, so I mean I believe you can actually write a verbal thing in the one-liner.

E

And yeah yeah instead, Stack Overflow, does have quite a few people talking about like how do I? How do I make distinct work better so it maybe we could just scope it to that. Yeah, like Jeremy son,.

A

Yeah I mean I have no obligation to get it of taking it, but I also have no pushback on this per se, so I mean if, if anybody feels like, we should just take it as is I'm fine with that, if anybody thinks we shouldn't take it I'm. Also fine with that, like I'm I'm in the.

E

A

E

Did we actually help that yuko system by adding it I guess, that's the big question I mean.

D

A

D

Use case is distinct, then maybe we should work on distinct if it's distinct plus like three other things, then then this is the provide the one funnel that they'll all go through once as we piecemeal update them.

E

We need to know what those other things are go first right.

D

D

Feel like it's not really needed and that maybe their complaint is just with distinct.

D

There's also a discoverability right like you, if you're looking at distinct and you need the Equality compare like, is it going to have even occurred to you that you wanted to say the quality compare of person dot create because yeah, because you still can't even do generic inference because it's on a generic type, static methods on a generic type have to be fully qualified? Oh.

E

You're right, yes, so.

D

Yeah, this is way better on distinct, because then imprints imply or inference applies.

A

Right but then it means you've.

D

Already been selected by the first argument, but but yeah it removes the redundant type word so I.

A

Mean I would say linkers once you have a lot but of a lot of lambdas. They also don't get better necessarily, so you might be better.

B

A

Actually assigning this out to a variable, in which case having the API somewhere else. That gives you a comparatively so.

D

It I think it's if, if we were going to add it, we should add, but a new static class equality, compare. That is not generic. That has the create that allows you to do inference. It would be creative T instead of a quality compared to create the same as we have with tuple dot, create which right.

A

D

A non-generic with a bunch of generic overloads, if you maybe.

A

Discussed this for the comparison one, and why did we not do them.

D

But cuz I think in general, we we like letting people get the t's out of the way. You know as long as it's not introducing ambiguity. So, even so, even if we took it I think we would. We would want a new equality, compare er or equality equality, compare that was non generic type to hold the create two etcetera.

D

A

Think I know we didn't do it for the compare one, because it is no comparable.

E

Yeah there might be an argument to be made for supporting anonymous types as well, but if we really wanted to make the ecosystem work better with anonymous types, we probably need language support as well.

A

Why are you saying it because I mean like you would like anonymous types, would basically be types that you value only add properties, and you can provide the compare literally like that right. Well,.

E

Right right now, for instance, it's really hard to instantiate a list of T where D is an anonymous type like you, you have to bounce through a generic method to do it like your own helper method. If you had a if you had a I'm, this is where you all you all come throw rocks at me.

E

If we had something more akin to like the C++ templating system, you could actually like properly get the type object that represents the tea and then instantiate the list of tea with that and everything would just work and it would all be statically compiled yeah.

A

I mean I'm not much time to me is more like a link thing right away. You effectively flow them in and then again.

B

A

Just said to list, for example, right, in which case you don't actually need to say the type ever right.

E

I mean now that we have records in the language. I think the use cases are going to be a little bit more common now, yeah.

A

A

Yeah, we need to see our records also play with all the other features. So that's the other thing where samples would help, because it kind of depends also which types you're composing, whether these things actually hello nicely but I mean the same of tuples. I haven't seen use tuples in link I mean give it a shot. But like your code, it's hard to see the braces.

A

There's like a lot of them. No so I end up very often have local variables, which the link is not easy to do all right. So then we leave that, as is for now, and then mm-hmm call it a holiday all right, then I'd say we did yeah I mean I can probably yeah. Well, there's five issues of whatever we did. So only ten more meeting in central the backlog is empty. Wait all.

E

Right what I mean up next by the way, just out of curiosity, so.

A

Which, which are the next ones yeah.

A

Okay, I, don't know what that means, but I.

D

Mean if it's its backlog, it's just ready for review by day.

G

D

E

Just curious as to what's coming up, okay I mean.

A

Formal issues them in the year 2019, which is actually pretty good, yeah.

F

E

Man, that's how one's gonna be fun.

A

Look at his pictures, I'm scared now. Well, it's.

E

uh It's it's just very contentious. That's all! Hopefully,.

A

It has the line those.

B

A

Is trivial or something well.

B

It's the name should it be tower to pie.

G

B

Name is the controversial thing: okay,.

G

E

We can get to that when we get to it all.

A

Right then I'll leave you all to it and then I see you guys next week.

E

Did you think bye-bye.

I