.Net Foundation Design Reviews, 2 Jul 2019

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: .NET Design Review: GitHub Quick Reviews

Description

00:00:00 - Approved: Add WriteTo convenience APIs on JsonDocument and JsonProperty https://github.com/dotnet/corefx/issues/39037#issuecomment-507769350
00:09:37 - Approved: PipeOptions, StreamPipeReaderOptions, and StreamPipeWriterOptions should not hardcode default sizes. https://github.com/dotnet/corefx/issues/37984#issuecomment-507774253
00:22:51 - Approved: Deprecate PipeWriter.OnReaderCompleted https://github.com/dotnet/corefx/issues/38362#issuecomment-507770664
00:23:40 - Approved: Change how the Arm intrinsics are exposed https://github.com/dotnet/corefx/issues/37199#issuecomment-507775790

A

All right so I think we are no live, which no means, of course, I'm live about. The second screen, though,.

A

You know we actually dive with the second screen um all right. So basically, this was a result of usability study where we asked people to write a JSON document or the content there off a pretty printed form to give in Jason Ryder, and so people tried various ways, so they can call the right APRs and it was very clear that they were somewhat confused by what we offered and where we offered it, and so I also made a proposal on how to simplify this. So one of the things we stumbled upon may be looked at.

A

Xml, for example, was.

A

Yeah, basically, how do we name those the RS, because in XML and basically everything is called right, which is generally the way the framework does things because you'd call things right on the writer and you call it right on something that writes to something else which in this case would mean me with their seven by two method. Then the other thing is. We have various ways. We can write things. We have individual values, we have individual properties, and so there were some inconsistency where things are so he basically has two proposals.

A

So first one is, we would add, additional write methods to property and documents, so we can actually write them individually as well, and then the the B part is, they say, get updated the element on jld api as an element to follow this right terminology and then, basically, within that proposal he has one separate balls of which is: can we just get rid of all of those and just have just write two methods and all three elements?

A

And then you can all write them individually to the writer, and so then the only scenario that you're effectively losing by having a single API call it's the one where you write a property under a new name. The way you would do that is, you would place it just call the writer you would write. You would call write my to go with the name is what write name or something or my property name, and then you would call the right to API on the on the jason element. That represents the value of the property.

B

A

B

Calls, instead of one correct.

A

And it's not just it might be less discoverable so with that something maybe have invalidated with the you know in these ability study. But what I find appealing about the you know the option two and the B thing is that everything is just a single right to method that looks very consistent to me and actually I have to say I like that. Quite a bit he's.

B

Crawling an email to show you what we currently have that's.

A

What we currently have, so this is the only API as we have so we have nothing on property and we have nothing on jason document.

B

B

So today, you're already making two calls to do what you were just describing no.

A

You would call it Jason element right property, then you would just pass in the name you want to. You want to write it up. Oh.

B

You you substitute a different name via the first argument. Yes, I say.

A

But what is also clear is that people didn't figure out how they can write adjacent property because they didn't figure out that they can. Basically, you know, call right right, property on the Jason element and then pass and stuff there right. So is.

B

This, the some toll of all on the whole sort of DOM, is this all of the api's that take a utf-8, jason writer I, believe that is correct. Yes, there's something very nice about getting rid of the methods that take something else before the writer and.

C

Things that pass on and I discussed, there's you know aging. If we keep the right property should so Part B one is make it right as property two and then another part of it is flip. The writer in the property name. So the writer, because right to our zero, should always be the thing that gets modified via side effect and but with the proposal of Jason document. Only having right to jason document is just jason element at the root.

C

So if, if we don't want to add all the right property overloads to jason document, we should just remove them from j cell and then the model is everything right. Since natural type document is, is a value element of the value. Property is rights as a property with its name at its value, and if you want to write document or element as a property, you call right property them on the right over makes.

B

A lot of sense can you scroll down to the sample before and after.

C

B

It certainly looks cleaner and more discoverable yeah.

A

I think the the the second one is basic. What you would intuitively is assume you could do, and that's also what you can do. An XML document I mean the only thing that that document that XML has that we don't have. Is this notion of inner and outer XML, but the basic right to and I have right to inner I believe but, like the mechanics, are pretty much the same.

B

I mean I think you know the first thing I might try. If I were a user, would be writer dot right doc, but from a layering perspective, I totally understand why we don't want that, and so.

A

B

Thing I would try would be reversing it and then I. Would you doc dot right? Oh right, writer, okay done maybe.

A

I'm, biased I just know from XML that the writer doesn't know about the dome, but the dome knows about the writer. So that's that's. That's I mean that's how most things are layered, so I think that's kind of already also what people would intuitively expect.

A

Right so that means nobody seems to have a dissenting view. Then the question more like is: what of the of the buffet of options? Are we going to do with happy? Are we going like.

B

A

B

A

I agree with that, so that means I'm actually open this guy. In my own brought a few, so I can actually type this like about. Looking at the other screen like.

C

A

It's also small group which help.

A

Yes, I mean, as.

C

Far as I know, a son's last opinion Lizzie, based on things he doodled on my way forward, passing up right.

C

A

A

A

You so somebody's asking him. Instead ad right methods on you, TF Jason, murder, Henry said we don't want this folio.

A

The moments go.

A

A

A

All right, so that makes the API approved.

A

The knowledge as the function of like who's, going to do a bit.

C

A

That one will be me a bit like that today, I think I talked with Dan yesterday, we'll figure out what the processes I suppose. It's just an email to tactics.

C

A

I mean it's pretty straightforward. Coding was.

A

All right, so this is the rider on reader completed so I pink-collar, because I wasn't clear what he specifically wanted. Do you know what I.

D

Remember so we implemented these adapters right, the stream hypership typewriter and the stream adapters, and what we realize is that, if you are trying to implement on reader completed, it's really hard to do two things. One is know whether or not the method is actually gonna. Do anything, for example like if you don't have a default pipe implementation. They call on reader completed with some funk like that thing may never be invoked at all.

D

Okay, actually, I think that's the primary reason, and so like one idea he had is maybe I can pass in a pool like weather or like I'm sort of keeping abilities. It's like whether or not this function wide would be called or not, and then I think you just gonna, realize this. This one's just completely unnecessary in that part I'm not as certain.

D

This is like why we're moving at vs., I'm kind of seen he's actually here or not well,.

A

It's also abstract, so that that would mean in order for us to deprecated it. You would have to tell you, but what else would be? Would you.

C

A

Would have to make it at least virtual and make it do nothing. Yes, yeah.

A

Yeah Mikey a lever for those kind of things is mildly low, because we haven't really figured out what the what the best story for this is I mean it's my the question of what about what deprecation his mind means because it, the simplest, be just basically do an obsolete plus make it virtual yeah.

E

A

Then the question is: is that what he wants and then the question is: what's the message one is saying like just don't do it or do something else or like what's the guidance exactly yeah and in general, like 300, is I think the best time for him to do things like that, because I wouldn't do it at 3 1. Thank you fee. If he really wants to do it, then the next opportunity is 500. Yeah.

D

D

D

And we're okay with doing this change, but like a lot more justification, maybe.

D

D

A

You all right so that one is this one, and then we are there for this one yeah.

D

A

So I am pretty sure we talked about this before, in some other context, I think in general. I think that makes sense to me because buffa sizes are definitely things that we know we want to tweak yeah using minus one makes sense, I. Think the last proposal was somebody said well, we could make it 0, but 0 would have meaning I mean to me what makes the most sense, because it's very clearly not the buffer size.

D

We don't make inaudible, that's the best option, I would say: if that's the pattern is.

D

In implementation system value button, when zero is not as a zero here would be valid. um You mind looking at the screen pipe reader insert like writer options, see if there's a buffer size of that, because.

C

If 0 has no sensible, meaning.

C

A

So this is this dream. That's okay! It.

D

Has a buffer size on it, so maybe I think with that it's by default. So what's the the constructor for the people, I think it's probably 4096 I'm, like that's so I think we should be the same with these ones.

A

Do we have these guys as well yeah.

D

Centered inside still.

A

Only half my video people type video options. I guess we don't want to have my video options, but so yeah.

C

We have a negative one constant for like people like do. We have a default bucket size, constant anywhere, like we have time span about infinity.

E

Like the one yeah, don't worry, do we view offer constants in a few places to be neighbor for I? Think, like the stream API is use it as well I mean? Is it public.

C

The fact that it's negative one so do we have a public bond, store public static, read-only that represents default exercise. You.

E

Know we've just used negative one in line.

D

So make sure I understand these options are specifically for the stream pipe reader and writer. The pipe one is not. This case is there's pipe options right.

A

um Look like it but I mean I asked this one.

D

It looks like we have: what do we do here.

D

A

Leave constants there.

D

Too so I mean if the idea is just to make it so it's negative one everywhere, instead of how many bug even buffer size. These are these.

A

Are effectively.

D

Just like thresholds right or the positives in thresholds, so I guess she's negative one again there and then in like a remark to say like what did you surprise with.

A

It because I thought we've reviewed this Anitha made a decision. Okay, but I mean I might be one with anybody else. Remember something here or is it just me and my or mine, Jeremy Mary I can't beat that guy senior team out before and I. Don't know it might have been the pipe option thing but I'm positive. We talked about this I think.

B

We talked about it and we said tanner. Can you go up in a separate issue, as we were discussing? Another issue that arises from me is Tanner opens an issue we're now looking at it.

A

E

Because I opened it day out and.

D

My guess is that this applies to both types of options: the pipe options, yeah.

A

So so we're gonna reader.

D

And writer options.

E

Have you seen- and this is basically just applying to whatever we've changed the API to from what the original was. So when you guys specify negative one.

D

Do you pretend, like a remark saying like what the current default is.

E

I, don't think we've done that I got before so like, because we don't want people to become dependent on whatever particular value. If they want something, then they should explicitly set it based on their own profiling data and otherwise we'll do whatever we think is the correct default, all right contract. You.

A

Should we document the default views and.

E

A

Might be wrong way, it.

E

Might be some more in the remarks.

A

Section yeah I mean you can also generally ask so, for example, you can do up a buffer stream and ask what its buffer sizes right.

E

A

You can discover at one time at least what the value is and which case would be the same here. I guess, if you pass the pipe options, new I would not expect the these radius to be minus one, and the options I would expect them where their actual default. That's right.

E

A

Which case you, if you really want to blog it, it's very log it or do something with that or otherwise it's very impossible for you to I mean how do you would even profile if you can't, if you cannot plot on a graph right, yeah.

C

Well, this is a proposal should we do it at the same time, should we add a constant buffer dot default buffer size negative one? So we don't have to discuss you to be negative one or zero ever a hit in the future.

A

To me, honestly, there.

C

Would that live them yeah.

A

C

We had a class full power, but buffer, not default buffer size system, buffers buffer, yeah. Well, that was system buffer buffer, not block copy that button.

E

I don't know some of the buffers namespace no.

C

Buffers is just memory anything.

A

And I wouldn't be a poster, but at the same time it seems a little bit over the top, because I would say that you know negative. One is kind of like you know, I, don't think anybody moves really surprising. What minus one means in this context? Yes, I.

E

B

E

B

Also shouldn't we definitely shouldn't do that now, because that has very broad implications. There are places where you pass negative one and it will blow up like file streams, buffer size, right right.

E

And and I think there's some places where we've said it's: okay to use zero and it's preferred to use zero when zero doesn't have explicit meaning because zero is cheaper to use right. But here we were saying: zero doesn't have explicit, meaning, but we're saying use negative one right, I think I think zero would be fine if we say it doesn't have any explicit meaning and won't ever have explicit meaning for pipe options. You.

C

Know what the exhibit mean is don't bother.

E

Right, which could be something that you want in some cases so.

E

Offer sites here.

D

E

D

E

Always that constraint memory scenarios where someone might want to say don't buffer yeah, but do it immediately? Is this number an actual number? Is this number a.

D

Hint because if it's used for every pool.

E

It's a hit so so, when you sometimes.

D

It's used for when you called pipe a pipe rider bracket memory, any passing zero there, that's just a hint, so it will get some bufferless larger, no matter what I.

E

Think in general is your negative one tends to be safer because you can't have a negative array, size and so yeah, and it gives the option for if, for some reason, a customer in the future says. Oh, we need. We need no buffering or anything like that. You can use your to specify that, rather than like negative two.

C

All right, I get my question: I mean the idea behind having a named constant for this is going forward. We have an easy answer. If you're defining anything that has the pixel buffer size, you are saying we now have platform guys mistake, negative 1 to mean either call or do not care. You know, whatever you think, is best I.

E

Think that probably be good, but maybe we should have a separate API proposal for that review. It yeah.

D

I, don't know why we known anyone.

D

A

Cry all right, so that means to me this is basically what we just discussed.

A

Anybody has dissenting opinions. Now is your time, otherwise,.

C

A

Appearance was just asking going to make stream default copy buffer signs public thing, yeah he's also suggesting doing overloads, but this is one of those things where I think overloads actually not working very well, because you end up with this commendatory explosions of what you know when you giving it their all lungs are. All ins is like. Well, you only gonna specify one of the three like. How do you would? How would you even design overloads for that yeah.

E

I think it I think in general. It's way better to just use negative one, because it's cheap to construct you just not a zero yep um and it's cheap to compare because it's just less than zero. So it's basically the cheapest way that isn't making up zero to say, let's use the default value and we and unlike a constant value, that's in line. We won't end up hurting downstream consumers. If we decide to change it later and they've already compiled the constant yeah I mean.

A

In practice, we have seen breaks with a change buffer sizes, but.

E

A

Encore I would say: that's in the realm of what you would have to absorb I mean right.

E

And we need to be able to target what's best, for what we've determined to be the common use case, and if someone else has something more specialized, they can always customize it themselves right.

A

Then that's pretty much all I had for today. This Emily hurts anything else. Otherwise, that's it and I will probably set up another meeting for sure. Let me push this API back to API needs work here, so it disappears from my rare.

A

And then I think the one thing we need to use the WinForms request for, and there was at the toss, dialog thing, but that that is actually quite a lot of API surface, so I haven't done it yet because we probably need full to our star chick over there plus. You also want to have the contributor on the call so I will set this up for as soon as we can is anything else. That's coming still for 300 Jeremy. Do you know anything in your area? I think you have.

A

You had a proposal, but it means that C and let's approve yeah. That's.

E

Basically approved I was planning on marking it approved today, but that's also for FIFO and it's too fun about.

A

E

It's on it's to unblock the arm, folks, from being able to work on on it, starting after the snap.

C

If a proposal that I've brought up months ago, for they do once a way of getting a memory back from a fix where we have spam read only spam, then we read only memory in the lower thing and Tanner I think you were gonna, see if they Mel Dutton. That was having to do a lot of work to work throughout that.

E

Would we ever heard back.

C

E

They aren't now because of the limited use of span on full frame, work and they're still targeting full frame workers their primary. So it may be more of a concern when they start using span and memory more and they back.

C

To the conclusion we had before, they may be useful, but we don't have enough demand for it also before right. Moving right, we.

A

Have about 40 open issues in the future milestone, so unless somebody pushes back I would just suggest like the next thing, whatever the next review cycle, is we just going over those I've got.

E

A few that people are keep asking me: can we start working on these yet.

C

And instead of a poor, West, yeah.

E

And then they want to hold you master that way they could start consuming them. Alright,.

A

Then this is for today, then. Let me end the call in a meaningful way and then see you guys next week, imodium.