-
Notifications
You must be signed in to change notification settings - Fork 62
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Baseline: 2+ major versions isn't enough #174
Comments
The same is true, but worse for Chrome. If a new version of Chrome lands "today" and if that release would trigger a feature to be included in Baseline, then there would only have been a one month window for users of that browser to adopt I bet some of us have even gone longer than one month between restarts of Chrome and have delayed updates that way. This is only relevant if developers adopt Baseline as a support target and if people use those features as foundational building blocks. Which is why I don't think Baseline should be communicated as anything more than a learning/teaching aid : #132 (comment) |
It's great to see discussion about the criteria for what's considered baseline. The current "last two major versions" criteria is a best effort at "will work for most developers most of the time" but it's not set in stone, and there are tradeoffs. A few approaches that have been up for discussion so far:
We haven't considered it an option to use global user share in the style of "supported for X% of users regardless of browsers." Above all, the criteria needs to make sense for developers, needs to work for most developers most of the time. There won't be a perfect and objectively right criteria, but I hope that feedback like this issue will move it in the right direction. |
I also asked this in the other issue, but this keeps coming up :) What does this mean? It is perfectly possible to stay in that description and still have an extremely biased baseline that only works for wealthy people from western countries that are aged 20-60 and do not have any accessibility needs. I know this is not the intention here, but I do fear it will be the outcome. The current baseline is this extremely biased baseline. I urge you too look at browser versions from the perspective of non western regions, children on second hand devices, pensioners on old devices, people who can't afford to upgrade, ... "Works for most developers, most of the time" is even against the spirit of the web which is intended to be an open and accessible space. Can't stress this enough, but this is irrelevant when Baseline is purely an indicator for "ready to learn more about X". But that is not how it is currently being framed here, in the announcement and on MDN. |
The guideline "works for most developers most of the time" doesn't just need clarification. I argue that it has no meaning at all. When you say "works for most developers most of the time," it sounds quantitative, but the only actual quantities that we can measure are end-user usage-data numbers like "X% of users/pageviews are in browsers that support There's no such thing as measuring whether a feature "works for a developer." Features work (or don't work) for end users and whatever creaky old browser they're using right now. And "most of the time" seems to imply that a feature would work for a given developer some of the time and sometimes not work for that same developer. But this is unquantifiable twice over. If a feature works for only 85% of my users, did the feature work for me, the developer, 85% of the time? 0% of the time? (Of course, it works for me on my machine 100% of the time. 😎) Since the guideline is unquantifiable, it's impossible to present data to support or refute it. I can show that
Ironically, it's the wealthy western web that's mostly on iOS, which is usually the laggard browser for modern web standards. Import maps just released in iOS 16.4 in March 2023, but it released in Chrome 89 in Feb 2021. Globally, 76% of users can use import maps, but in the US, only 58% of users can support them. Probably that'll be a bit higher by the time iOS 17 releases in September, but if I use import maps on my site in 2023, and so my site stops working for 30% of Americans, that's not OK. Since I'm running a business here, I want to pick a baseline that will serve the rich and poor alike. 2+ major versions doesn't serve either group of users well. |
It's about answering the question "can I rely on this feature working for my users?" It's the same question that you come to MDN's compat tables or caniuse.com for, but condensed into something you can understand at a glance. For the answer to be work (be useful) it has to match developer expectations and it has to match the reality of what will actually work for their users. This isn't a precisely defined and quantifiable guideline and I don't think anyone expects data alone to guide us to the best definition. IMHO a compelling argument against the current definition is what @dfabulich noted in this issue: |
I think thats the problem. Baseline is not really better than the current matrix at the bottom. Because as a developer, I know which browsers are supported by my application. But I don't know which browsers are supported by Baseline (its a moving target). I think step one is to ensure that Baseline is never mentioned without also declaring the year. In addition to that, perhaps the year could represent the date that all browsers supported the feature. So it would be more like a lowest common denominator. Then I could think about Baseline 2022 the same way I think about ECMAScript 2022. |
I mean that it's an obvious option, but not seriously considered because it would involve browser market share, which would be very hard for browser vendors to come together around as a criteria. We'd have to tackle at least two problems. First, what data can be trusted here, and I suspect there would be no agreement on that. Second, how to deal with the fact that market share can go up and down, so at least theoretically features could go in and out of Baseline as that happens. But none of this was discussed at any length. |
@dfabulich thanks for sharing that Android Studio screenshot, I agree that something like that based on yearly Baseline cuts has a lot of promise. A great feature of this proposal is that it doesn't bake market share into the definition of Baseline, but uses it to give a recommendation about which Baseline year to target. @styfle your suggestion is along similar lines, and I think that may very well make sense once Baseline 23 is done and we're into 2024. Note that a yearly release of Baseline is already part of the plan, see https://web.dev/baseline/. |
#173 is also obviously relevant here. As it stands, “last 2 versions” means 1–2 years for Safari (which seems a fairly reasonable sort of period in general, though adding a month or two at the low end would probably improve it), and 4–10 weeks in other browsers (which is very unreasonably little, and also means that something like Firefox ESR is never considered, since it’s more like ten months back). This metric is currently probably ignoring around 15–20% of actual users most of the time. (The amount will vary through each browser’s release cycle, being worst immediately after a release.) The only reason it’s not a total disaster is the different versioning scheme used by Safari, because Safari from a year ago is in most areas close enough to other browsers from a year ago. But if #173 gets fixed (so that “last two versions” of Safari would currently mean 16.3 rather than 15), then this would be a hopelessly bad and terribly misleading metric of a baseline. “Major versions” is clearly an unsuitable measure. You need something either time-based or usage-based. Time is the easy and practical one to use. And it should definitely be more than “as little as four weeks ago”. Fourteen months feels a fairly practical sort of a figure, not overly conservative but giving some time to update that usage figures show is needed. |
I'm really grateful for the discussion here and I'm learning a lot (it's especially great to see the Android Studio example—it's both wonderful and painfully intimidating at the same time). We relied on survey data in which web developers said they developed for browser version numbers per se, but it didn't capture the whole story, I think. I do want to note one thing about scope, prompted by @dfabulich's earlier comment:
I don't think that's exactly right: we can survey and interview web developers and learn about their experiences with features and browsers over time and what constraints they work within. "What are web developers under pressure to do?" is a distinct question from "What browsers are people using this week?" That is to say, a feature not working for a given end user may or may not be a problem, if a dev (or their bosses or clients) made an intentional choice to not support that user's creaky old browser. Or maybe a specific creaky old browser is actually fundamentally important and widespread usage is totally immaterial (let's show some sympathy to the poor developer who has to maintain compatibility with some ancient iPad kiosks that haven't had an update in years). We know that web developers routinely make support decisions that are not rooted in global usage share numbers or time since release or number of releases. We know that no one summary indicator of support would be definitive for every developer. But I think we can still help developers understand what the web can do with respect to their own constraints, if we choose a definition of Baseline that is understandable and predictable enough that a web developer can compare their own work against Baseline. But plainly we have to do some work to help developers know where they are relative to Baseline (e.g., with dates, with usage reports, and so on) no matter where we end up with a definition. |
De facto the way "2+ major versions" has been interpreted is "2 years for Safari, but 8 weeks for Chrome or Firefox" (or maybe it's 1 year, and 4 weeks?). For consistency, Safari major version should go to one digit after the period; 16.4 is logically as much a major release as Chrome 112.0, and includes substantially more web platform features and bug fixes, so it's hard for me to see why one counts as a major release and the other doesn't. We potentially ship web platform features in all of .0-.6, for a total of 7 releases a year that should plausibly count. Considering only 1 a year seems to be an overly literal interpretation of the version string. This is somewhat separate from how many major versions to go back for each browser. Perhaps a lot more Safari users are on older versions. However, at a glance at least, this does not seem supported by the available evidence. Below I'm including data from Statcounter "Browser Version" share for April 2023 (I transposed the CSV download). Note that mobile versions. of Chrome and Safari are collapsed to single catchall versions like "Chrome for Android" and "Safari iPhone", so the stats below are meaningful only for desktop. I note that if you go back to Safari 15.4, what was then the year-ago Safari release, there's 9 different versions of Chrome with more use share, with the oldest of these being Chrome 79.0. If you go back to Safari 15.0 (the last "major version" by the scheme that only counts .0 releases), there are 37 different versions of Chrome with more use share, of which the oldest is 49.0. Perhaps Statcounter is not reliable enough. But ideally the "how many versions back" policy should be informed by actual data, not by counting the dots in version numbers, or by anecdotes about user complaints. Source: https://gs.statcounter.com/browser-version-market-share
|
There's been a lot of good feedback on the definition of Baseline, but I'm very unclear on what's happening next. Is this going to be a discussion topic for the next WebDX Community Group meeting? If so, when is that meeting scheduled? Per web-platform-dx/admin#2 there's no publicly available information on when the next meeting is scheduled, or any information on past meetings (or even whether there have been any past meetings). What happens now? |
Since people are commonly not aware of this: it’s based on trackers that are blocked by extremely common configurations (probably most ad/content blockers: certainly uBlock Origin with its default lists, and Firefox’s built-in Enhanced Tracking Protection blocks it in Strict mode, and in Private Browsing windows in its (default) Standard mode). It will almost certainly significantly undercount Firefox especially. The way it pretends that most mobile UAs are instantly up to date (as you noted) is also rather perplexing and unrealistic. |
This is a fair point, but it probably doesn't significantly change the point about relative version spread of different browsers. Perhaps Firefox is even more weighted towards recent versions than the data suggests (since ETP would not be in very old versions). Maybe all browsers are more weighted towards recent versions in reality.
I don't think they are claiming they're always up to date, they are just failing to make version distinctions. That seems like a string limitation. I believe mobile versions of Chrome and Safari both have identifiable version in the UA string and/or via UA Client Hints. Anyway, I presented StatCounter not as the best available usage share data, but because it's an easily accessible public data source. Better data or other sources welcome! Even taking account of all the methodology limitations in StatCounter, I think the data still contradicts the position that a month of Firefox or Chrome updates gets the same user saturation as a year of Safari updates. And the current definition of Baseline assumes that position. |
But... Shouldn't there be? Even if such a dataset (or maybe even more specifically: way to create such a dataset) doesn't exist, shouldn't it? It definitely feels worth a valiant effort at least to see if we can find a way to agree on facts that ultimately shape so much/so many decisions. |
Also - I don't think browser vendors were consulted about the current criteria, which are 8 weeks for Chrome, Firefox, and Edge; and 2 years for Safari. At least, no one from Apple was consulted. No one even asked us which Safari versions are major versions. We would not have said it was just the one .0 release a year. The current policy is neither neutral nor non-controversial. It wrongly implies that Safari updates much more rarely than other browsers and/or that Safari users are much less likely to have a recent version. |
Thanks a lot for all the comments here. Definitely very useful in iterating towards a more useful definition of Baseline. I want to address a couple of specific things: @dfabulich Yes, we’re going to discuss this in the next WebDX CG meeting, and I’m collecting incoming feedback for that purpose. We’d love for people to join the CG and attend the next call. Once you join, you should be able to subscribe the calendar. The next meeting will be on May 25th @othermaciej we did bring this up in the WebDX CG multiple times and have asked for feedback, but didn’t hear about any concerns. We based this initial browser set on research we did where we asked about browsers and versions that devs expect to see a feature in. That said research on this is not easy, this is just a starting point and it looks like 2 major versions back is not the one to go with for now, and we’ll focus on a better definition. I’ll ask to put this on the agenda for the next WebDX CG call. |
@othermaciej thanks for the feedback! It's a long thread so I want to bring your attention to some possible approaches in #174 (comment). I'm fairly confident we can arrive at something reasonable given the input of browser vendors and web developers. I'd also like to highlight that the rollout of this on MDN isn't automatic, rather it shows up when features are designated as Baseline in this repo, which is manual with human review. So we can and will steer clear of cases that are in this gray zone of probably too new to call Baseline. Of the features we've already done, I think A good test feature for any definition might be Container Queries, which we (@ddbeck) left commented out. It would be Baseline by the "2+ releases" rule, but not the "2+ major releases" rule. |
@bkardell it's an excellent point. Do you want to file a new issue for that? |
Yes, I did see it. I find the options open-ended enough in how they are stated that I don't know if they solve this problem?
Interesting. Is the policy for manual human review (who decides, what criteria they should apply) written down?
I'm not claiming that "releases" is a better rule than "major releases", necessarily. My claim is that your definition of what is a "major release" is incorrect for Safari. Safari 15.4 is a major release. Safari 15.0 is a major release. Safari 15.4.1 would not be. We increment either the first or second number for what we consider a major release of the browser, instead of always the first, like Chrome, Edge or Firefox do. We do it this way so the version number (on both iOS and Mac) stays mostly aligned with iOS version numbers as we find this is less confusing to users and developers. Any Engagement with the Safari team would have clarified this. I think for now, it would be best to update the definition of what counts as a major version for Safari, even while continuing the broader discussion of what the Baseline criteria should be. (Perhaps this would be more on-topic for #173?) |
None of them rely on a semver-like concept of major release, so any of them should solve the problem.
No, but @ddbeck was going to document some things around governance and it would make sense to document review policy as well. Also let us know if Apple wants to have reviewers, similar to BCD.
I agree. I think we should use BCD's concept of release, which doesn't depend on the structure of the version string. This shouldn't change any Baseline badge shown on MDN today. As @atopal said this will be on the agenda for next week. |
Maybe it's better to use a different wording in the context of Baseline? Something like For Chrome a significant release is an increment in the first number, for Safari an increment in the second number. |
Taking yearly snapshots of the interoperable web platform surface makes a lot of sense for many different uses, and the current definition of Baseline (2+ featureful versions) works reasonably well for including or excluding features from these snapshots. However, my concern is that reusing the criteria to continuously update what’s ’safe to use right now’ results in imprudent advice to developers, as others have pointed in previous comments. What’s the argument against offering more conservative guidance? One idea that hasn’t been discussed, as far as I could find, is placing the feature in relation with these snapshots. So a feature could be one of:
This essentially makes the current Baseline a semi-recommendation, while raising the bar for what's safe to use. |
I should clarify that the intent with using browsers versions was to have an easy to understand proxy for how much uptake a feature has between versions of the same browser. Moving forward, we'll make sure to not to refer to Safari Y.x releases as major versions, to ensure there is no unnecessary confusion about their significance. Independently of that, we need to ensure that the shorthand (x browser versions back) actually represents the coverage that developers expect, which seems like is somewhere above 90% and below 100%, and over the next few weeks we'll try to get identify better data on that as well as research on what expectations are. @danburzo snapshots and the ongoing classifications solve different problems. One is a "push message", and gives us the ability to talk about what has changed between releases, the other is a "pull message", giving developers information when they need it and are looking for it. There is definitely an argument to be made for being more conservative, and 2+ versions might not be enough, but that would also be true for the snapshot (at least shortly after announcing it). With all of this, it's probably a good reminder that we expect more than 90% of the platform to safely be in the "Baseline" bucket. The vast majority of features of the platform have been around for a long time. |
Platform features, sure, but that’s not the point of Baseline, and completely irrelevant—if it were relevant, there wouldn’t be much purpose to Baseline. The point is how ready new features are, which is about actual availability in the browsers people use. And the most generous stats on the current “two major versions” definition, including similar browsers by engine (even though they don’t always have exactly the same feature coverage, for various reasons), fall way short of 90% of users. If the current definition goes back to Safari 15.0, then depending on the point in the release cycle, it’s only covering about 80–85% of actual users. However, by sheer accident (as it appears to me), most of the remainder featurewise is caught due to the longer Safari coverage, since most features that Safari had 1–2 years ago that Chromium and Firefox had 1–2 months ago, Chromium and Firefox also had 1–2 years ago. This is the sole saving grace of the current definition. If the current definition were changed to match the realistic definition of “two major versions” for Safari (as declared by their developers and observed by anyone who can look past a stubbornly fixed notion that X.Y.Z must be major.minor.patch to the clear reality), to what would currently be Safari 16.4, then at some points in release cycles (being worst, I think, immediately after the release of the next major version of Chromium after a Safari X.1 release, which is probably around early November) it could be covering less than two thirds of all users (and that’s globally—the figure would be worse in some areas). The current definition is already dangerously broken and misleading. Yes, the difference is only visible in a few cases, but those few cases are most of what Baseline’s purpose should be. |
What's the definition of the "last 2 version" in the case of Safari. |
@romainmenke are the numbers you're sharing from the server logs of a specific website, and if so which one? I'm under no illusion that the data behind browserslist is authoritative, but is there any publicly available data source that is less bad? |
@mrleblanc101 we decided in the CQ meeting yesterday that we're going to use the last 2 versions of Safari listed in https://github.com/mdn/browser-compat-data/blob/main/browsers/safari.json, which currently means 16.4 and 16.5. Initially, the plan was 15.x + 16.x, but there was a lot of good feedback in this issue and #173 that this doesn't make sense, it's too literal an interpretation of "major version". |
But still only two versions? Ugh. So now it will literally be covering less than two thirds of users at some times of year. It should urgently be left as it is until that aspect is fixed, because otherwise it just makes this a mockery of a concept of a “baseline”, rather than a bad definition redeemed in most cases by an accident. |
@chris-morgan we're very aware that the current definition is too close to the bleeding edge in some cases, and while we're working on an updated definition, we'll steer clear of the gray zone and not mark things as Baseline on MDN if they're too new. As I mentioned in #174 (comment), It's possible those won't be Baseline by a new definition, we'll see. |
(thank you all for the great input so far - I would like to ask we all refrain from using derogatory or inflammatory language when expressing our opinions to ensure we keep the conversation constructive and professional) |
Woah that's crazy ! I guess that problem is more that Safari is tied to OS release, but people don't update OS as much as browser (since OS don't auto update in the background like browser do). I also raise HUGE concern about supporting only the last 2 minor version of iOS as Baseline. That will leave ten's of MILLIONS, maybe hundred of millions of people at risk of using a broken website. At this point Baseline is just useless and people will need to keep using tools like Can I Use to know the real support and juge depending on their Analitycs data. |
the decision on Safari is only to align it with other browsers; it doesn't impact features currently marked as baseline, and it doesn't imply this will stay as we revisit the definition. And again, let's avoid inflammatory and derogatory (or ableist for that matter) language. |
My point was more that this data source can be valuable, even when everyone agrees it is not a good source. But if the bad parts are not ignored, the assumptions and conclusions based on this data will be off. 40% is just massive and skews the data heavily. Our data covers ±40 sites, mainly regional to Belgium.
I want to make it clear that this data is not authoritative in any way :) I think it's easier to test things that are not true at this point, than the other way around. example :
This is untrue because That it is untrue can be easily checked by anyone keeping there own stats.
No idea |
@romainmenke cool, so IIUC you work for or run a web development agency in Belgium with around 40 sites, that you can get stats for? I agree that's useful to contrast with other public data sources. Do you happen to have numbers for mobile browsers to give us an idea of what the version breakdown looks like? |
@mrleblanc101 I understand your concern. "Last 2 versions" indicates that something might be labeled Baseline when it in fact is not widely supported yet. The goal is to come up with a definition that is both easy to understand and reflect developer expectations when it comes to availability. It seems like coverage of somewhere between 90% and 98% will be needed for that. We're currently exploring ways to better understand that expectation and better understand the data available to make a determination of availability. Thinking out loud, I'm wondering if it would be better to declare something much more conservative as Baseline, while we explore options. Maybe something like "available in all browsers for at least 2 years", which of course might taint expectations the other way as "This is obviously too long to wait". @jgraham That's an interesting idea. So, each vendor declares a version threshold they want developers to target, and keeps that updated regularly. That seems very similar to the Android Studio example in the Dan's comment above. I do agree though that vendors would have to also publish the relative percentage of reach to make it transparent for devs what they get when they do that. We could then test whether those vendor recommendations are in line with developer expectations, and if yes, we could declare a feature Baseline when it has support in all Baseline versions of all browsers. |
I have some data to share! https://github.com/dfabulich/baseline-calculator I gathered data from tl;dr: It takes about four years for 80% of features to achieve 95% market share.
Details on methodology, steps to reproduce, and interpretations of the data are available in the README of the repository. Based on this data, I argue for the following conclusions: Baseline can and should be time-based
|
That's fantastic Dan! Thanks for sharing. I particularly like that you looked at which features would be excluded at different levels of availability. One major caveat when it comes to historical comparisons: IE11 was released in 2013 and never updated. Other than UC browser, I'm not aware of any browser that shares that issue. I'm going to guess that with IE discounted it takes much less than 4 years after availability across all browsers before developers feel it's widely available. One way to test that would be to look at usage metrics for features. Either way, fantastic tool to have, thanks for making that available, Dan! I'm sure it will be very useful in testing assumptions we are making. |
I have a section about this in the repo readme, which you might want to review:
But that means that we shouldn't pick and choose individual browsers with low market share and ignore them. We should say "we target 95% market share." The fact that IE isn't getting update isn't what matters; what matters is that IE has 0.37% market share. But watch out, because Opera Mini has 0.99% market share all by itself, and Android 6 and below has 2% share, and iPhone 7 and below have another 2%. That's why it takes four years to get to 95% share, and that's why it really, truly, isn't safe to adopt features after just two years from the day all major browsers start supporting it. |
Thank you Dan! I did see that section, but I don't think it applies to IE11. IE11 was an actively supported browser with much more than 2% marketshare that was installed on new computers for almost a decade. In the table you provided, there is a massive gap in time between 90% share and 95% share, from 22 to 48 months. Reasonable developers can decide that 5 points are not worth waiting 26 months for. Also, I haven't run the data, and I'm not sure you can, but my hunch is that getting to 90% would be much faster today than say 5 years ago, specifically because IE11 and because I can see many positive developments to reduce fragmentation (eg. Safari on Mac now being OS independent) and none in the other direction.
But it's not "really, truly safe" to ever adopt a "new" feature. There will always be people excluded, assuming that's what you mean by not safe. Most of them never ever get to 97%. Why is that not the safe threshold, but 95% is, why not 93%? In the end, it's a judgement call, because developers and businesses have to balance limited resources and consider diminishing returns. I think one thing we have seen so far is that we neither have enough/reliable data about browser uptake, nor enough research on developer expectations to make definitive statements. That's worth spending time on, I don't think we need to be in a hurry here. That said, I'm quite excited for all the gaps we have already identified and really appreciate everyone's effort to contribute to a better understanding of the situation. It's not a simple or easy problem to solve, but this is clearly a big web developer pain point and something we can have a positive impact on. |
I did run that data for 95%, but it's too sparse. Out of 110 features released in the last five years, only 74 of them have reached 95% share, because most of them just haven't had four years to mature!
Sadly, that's false. There is a recent major change in the other direction: Google Chrome withdrew support for Android 5 in 2021 and Android 6 in 2022. The plan is to withdraw support for Android 7 later this year, and Android 8 next year, in perpetuity. This policy will create persistent fragmentation in the Android ecosystem forever. 😭 More broadly, I'm afraid that a common view among the team here is that the purpose of Baseline is to signal to developers in a clear, convincing way that features are now mature in less than two years. But, if your target market share is 95%, that's just not true; it probably never will be.
I agree with half of that: we need to survey developers to find out their desired market share thresholds and how long they're willing to wait for it. But I don't agree that we don't have enough/reliable data about browser uptake. We have enough data to tell us that 80% of features take four years to reach 95% share and two years to reach 90% share. That lines up with the data from the RUM Archive https://github.com/rum-archive/rum-archive that we've discussed in #190. I see no reason to think that additional data sources will give us a fundamentally different answer. Now, we just need to pick a market share threshold, which we can do with a survey like the one I've drafted in #208. With that survey data, we'll have all the information we need to make a decision. |
I don’t want to use the word “hurry” but I do think time is of the essence if you want developers to use Baseline, because its already visible on MDN docs. You can’t wait years and then change the definition because developers are going to be upset or assume the original definition. (For example, look at what happened to twitter’s verified checkmark when it changed meaning) |
@dfabulich https://github.com/dfabulich/baseline-calculator is pretty cool, thanks for sharing! I'm wondering if instead of computing a time to availability for each feature, if it's possible to compute the time to 90/95/98% availability at different points in time for a hypothetical feature that is enabled at the same time in all browsers? This is the worst case for a new feature, because any time a feature lands at different times, there's a "head start" for the early browsers when it lands in the final browser and we start counting. In other words, for every month of the past 5 years, assuming a feature is available in every subsequent browser release, how many months does it take before that hypothetical features is available to 90/95/98% of users? Both a per-browser number of months and an overall number would be interesting. My expectation is that per-browser 95% answers will roughly match #190 (comment), and that the overall 95% availability delay in months will be very sensitive to the precise inputs for the browsers with the slowest upgrade rates, both what its market share is and how fast users upgrade. |
@foolip I'm not 100% sure what you're asking here. I've filed this as dfabulich/baseline-calculator#3 and maybe we can follow up there. |
Good news! I've published a year-by-year cohort analysis on https://github.com/dfabulich/baseline-calculator, and it does show improvement over time. I'm now convinced that we can define Baseline as "release + 30 months" rather than "release + 48" months. Focusing just on the "80% of features" column, here are the results for all cohort years:
Based on this, it seems plausible that when defining Baseline by a number N months, we should consider only the latest cohort year for determining "N". 80% of 2015's Baselineable features took 54 months to reach 95% market share. 80% of 2020's Baselineable features took just 30 months to get there. Based on this data, we should define Baseline based on the latest cohort year in which 80% of features have achieved 95% market share. Currently, that's 30 months for 2020, but we can update this in future years based on new data. (We don't yet have 30 months of data for features released in 2021.) |
Closing this issue as generally addressed: The group rolled back on the "2+ major versions" approach and adopted a new baseline definition, thanks to the research conducted here (and elsewhere). Additional issues may be created to track more specific points. |
Baseline sets a target of 2+ major versions. Today, that means that the
<dialog>
element, released in Safari 15.4, is "baseline," because the latest version of Safari is 16.4.I tried launching a feature with
<dialog>
a couple of months ago in March 2023, without a polyfill, and we got a bunch of customer support tickets from users of iOS Safari 14.Furthermore, under the 2+ major version policy, assuming Safari 17 releases on schedule in Fall 2023, on the day Safari 17 drops, every feature from Safari 16 will be officially "baseline." For the rest of that month, no more than 50% of users will actually have Safari 17 yet; the "2+ major versions" policy will effectively drop down to 1 major version.
"Baseline" should be 3+ or 4+ major versions.
The text was updated successfully, but these errors were encountered: