Everything
php
Learn Community Releases The PHP Project
Home » Project » Newsgroups » php.internals » Feature preview (formely was: experimental features) »

Feature preview (formely was: experimental features)

php.internals

David Rodrigues

3 years ago
In order to try to organize a little better the original purpose of the discussion I started, I decided to "rename" this subject to something closer to the purpose. From "experimental features" to "feature preview". Original thread: https://marc.info/?l=php-internals&m=166491585711553 These are my suggestions, as per the discussion that took place during the original thread. (A) What are preview features: (A1) There are features ALREADY 100% APPROVED through the voting system among the PHP team and that can already be implemented in the language in a preview format (via PR), available to the general public through minor/patch releases through "preview" markers. (A2) These are features that are understood to be effective for the language and that will be available in a future major (eg. 9.0) or minor release (eg. 8.2). That is, it is not intended to be removed "at any time". (B) What are NOT preview features: (B1) Suggested features, even if well accepted during discussions, but not passed the voting process as approved, even if a PR is already available. (B2) Experimental features (literal), whose idea would be to test "in practice" its functionality before deciding if it will be part of the language or not. (B3) Features that may have a high risk of rejection due to performance, security or usability issues, even if they were approved during voting. That is: when in doubt, even if approved, keep it as exclusive to the next release. (B4) New features approved but with pending discussion should not be previewable. For example: json_validate() is approved, but the function name is under discussion whether is_json_valid() or json_validate() is better. In this case, it prefers to keep it out of the preview until that is decided. (C) What changes in the voting process: (C1) An additional voting topic to decide whether a new feature can be previewed or not (i.e. basically deciding whether the feature is safe to test ahead of time, taking security primarily into account). (C2) Additionally, decide what the target version of PHP will be (eg. PHP 8.2? 8.3? 9.0?). This helps more complex features get more preview and feedback time. (D) Advantages: (D1) New features will be able to be tested before an official release (eg. 8.2), and all issues can be resolved before this release. (D2) During the voting process, a feature can be developed for the next version (eg. 8.2) or postponed to the next-next version (eg. 8.3), if it encounters difficulties, without freezing/delaying the release of version 8.2, for example. (D3) Regular users will be able to test features in preview, without installing nothing (eg. PECL), since they will be available natively in PHP. Of course, with the notion of the impacts such features can have (like needing to use markers, minimal performance issues). It is important that the documentation makes it clear that a particular feature is under development in preview mode. (D4) The development team will have early feedback on these features, and will be able to work on fixes with more "peace of mind" and security, since there will be no obligation to release a patch that quickly (except for security issues). (E) Disadvantages: (E1) The development team will have to pay attention to early feedback for a feature still in preview. However, this is something that would happen after the final version anyway. In practice, I imagine the time used would be equivalent. (E2) Once the final version is released, the language would have to accept some markers temporarily and deprecate-notice them until it is sure that users have migrated to the final version. Probably these markers should be permanently removed at the next major or major+1. (E3) Users using preview features should be responsible for updating the code as soon as the final version becomes available. (F) Final considerations: (F1) Preview features need markers, either in the function (eg. preview_json_validate() or maybe Preview\json_validate()), in new classes (eg. Preview\XYZ::method()), in new methods (eg. XYZ::preview_method()) or, if affect the language itself, in declarations (eg. declare(preview_xyz = 1)). (F2) Preview features can only be removed when there is a serious security risk to the language that is irreparable. By way of comparison, removing a preview feature is like, after releasing PHP 8.2, dropping support for "readonly" as class modifier. That is, it is understood that it must be thought of as something unfeasible and impossible to happen. (F3) Preview features can be started in PHP 8.1, for example, and made final in PHP 8.2, 8.3, 8.4 or 9.0, and so on, as long as this is decided during the vote. (F4) We can try a single preview feature. Instead of thinking that "features in preview" should become a default for PHP now, in a definitive way, we can opt for a new feature approved in PHP 8.3 (currently I don't think there is any approved, but eventually we will), for example, and make it available in PHP 8.2 as preview. So we test in practice whether the idea itself will work or not (and even user adherence, with open source projects and feedback). And if all goes well, we can make it default for PHP to offer this kind of support. If something goes wrong, we can let this idea go. (F5) Preview features should generally be released in patch versions (eg. PHP 8.1.12 to preview a PHP 8.2 feature). Atenciosamente, David Rodrigues

Larry Garfield

3 years ago
On Tue, Oct 11, 2022, at 3:25 PM, David Rodrigues wrote:
> In order to try to organize a little better the original purpose of the > discussion I started, I decided to "rename" this subject to something > closer to the purpose. From "experimental features" to "feature preview". > > Original thread: https://marc.info/?l=php-internals&m=166491585711553 > > These are my suggestions, as per the discussion that took place during the > original thread. > > (A) What are preview features: > > (A1) There are features ALREADY 100% APPROVED through the voting system > among the PHP team and that can already be implemented in the language in a > preview format (via PR), available to the general public through > minor/patch releases through "preview" markers. > > (A2) These are features that are understood to be effective for the > language and that will be available in a future major (eg. 9.0) or minor > release (eg. 8.2). That is, it is not intended to be removed "at any time". > > (B) What are NOT preview features: > > (B1) Suggested features, even if well accepted during discussions, but not > passed the voting process as approved, even if a PR is already available. > > (B2) Experimental features (literal), whose idea would be to test "in > practice" its functionality before deciding if it will be part of the > language or not. > > (B3) Features that may have a high risk of rejection due to performance, > security or usability issues, even if they were approved during > voting. That is: when in doubt, even if approved, keep it as exclusive to > the next release. > > (B4) New features approved but with pending discussion should not be > previewable. For example: json_validate() is approved, but the function > name is under discussion whether is_json_valid() or json_validate() is > better. In this case, it prefers to keep it out of the preview until that > is decided. > > (C) What changes in the voting process: > > (C1) An additional voting topic to decide whether a new feature can be > previewed or not (i.e. basically deciding whether the feature is safe to > test ahead of time, taking security primarily into account). > > (C2) Additionally, decide what the target version of PHP will be (eg. PHP > 8.2? 8.3? 9.0?). This helps more complex features get more preview and > feedback time. > > (D) Advantages: > > (D1) New features will be able to be tested before an official release (eg. > 8.2), and all issues can be resolved before this release. > > (D2) During the voting process, a feature can be developed for the next > version (eg. 8.2) or postponed to the next-next version (eg. 8.3), if it > encounters difficulties, without freezing/delaying the release of version > 8.2, for example. > > (D3) Regular users will be able to test features in preview, without > installing nothing (eg. PECL), since they will be available natively in > PHP. Of course, with the notion of the impacts such features can have (like > needing to use markers, minimal performance issues). It is important that > the documentation makes it clear that a particular feature is under > development in preview mode. > > (D4) The development team will have early feedback on these features, and > will be able to work on fixes with more "peace of mind" and security, since > there will be no obligation to release a patch that quickly (except for > security issues). > > (E) Disadvantages: > > (E1) The development team will have to pay attention to early feedback for > a feature still in preview. However, this is something that would happen > after the final version anyway. In practice, I imagine the time used would > be equivalent. > > (E2) Once the final version is released, the language would have to accept > some markers temporarily and deprecate-notice them until it is sure that > users have migrated to the final version. Probably these markers should be > permanently removed at the next major or major+1. > > (E3) Users using preview features should be responsible for updating the > code as soon as the final version becomes available. > > (F) Final considerations: > > (F1) Preview features need markers, either in the function (eg. > preview_json_validate() or maybe Preview\json_validate()), in new classes > (eg. Preview\XYZ::method()), in new methods (eg. XYZ::preview_method()) or, > if affect the language itself, in declarations (eg. declare(preview_xyz = > 1)). > > (F2) Preview features can only be removed when there is a serious security > risk to the language that is irreparable. By way of comparison, removing a > preview feature is like, after releasing PHP 8.2, dropping support for > "readonly" as class modifier. That is, it is understood that it must be > thought of as something unfeasible and impossible to happen. > > (F3) Preview features can be started in PHP 8.1, for example, and made > final in PHP 8.2, 8.3, 8.4 or 9.0, and so on, as long as this is decided > during the vote. > > (F4) We can try a single preview feature. Instead of thinking that > "features in preview" should become a default for PHP now, in a definitive > way, we can opt for a new feature approved in PHP 8.3 (currently I don't > think there is any approved, but eventually we will), for example, and make > it available in PHP 8.2 as preview. So we test in practice whether the idea > itself will work or not (and even user adherence, with open source projects > and feedback). And if all goes well, we can make it default for PHP to > offer this kind of support. If something goes wrong, we can let this idea > go. > > (F5) Preview features should generally be released in patch versions (eg. > PHP 8.1.12 to preview a PHP 8.2 feature). > > > > Atenciosamente, > David Rodrigues
So let me use a concrete example as I'm still not completely clear here. Let's suppose, hypothetically, that the asymmetric visibility RFC passes on 1 December this year as a preview feature, with the current syntax proposed. It's a syntax change, so polyfills are not possible and namespaces are irrelevant. You're saying that "public private(set)" would become available in 8.2.1 (the late-December release of 8.2), but only in files that have "declare(asymmetric_visibility=1);" at the top. However, we could still change the syntax from "public private(set)" to "public private:set" in 8.2.6 if we decided to; that wouldn't be considered a BC break. But once we get to 8.3.0, whatever the syntax is at that point is frozen and no longer changeable, and available in all files. The declare() is now irrelevant. Is that how you envision it working? --Larry Garfield

David Rodrigues

3 years ago
Nice example!
> You're saying that "public private(set)" would become available in 8.2.1
(the late-December release of 8.2), but only in files that have "declare(asymmetric_visibility=1);" at the top. Yes. In that case, I should suggest declare(preview_asymmetric_visibility = 1), just to make clear that it is using a preview feature.
> However, we could still change the syntax from "public private(set)" to
"public private:set" in 8.2.6 if we decided to; that wouldn't be considered a BC break. But once we get to 8.3.0, whatever the syntax is at that point is frozen and no longer changeable, and available in all files. The declare() is now irrelevant. Partially. As the usage definition has not yet been decided between private(set) vs. private:set, so this feature is not ready for preview. Unless the idea of allowing the two syntaxes to co-exist in this feature is acceptable (which I personally think is a bad idea). Atenciosamente, David Rodrigues Em ter., 11 de out. de 2022 às 22:23, Larry Garfield <larry@garfieldtech.com> escreveu:

Larry Garfield

3 years ago
On Tue, Oct 11, 2022, at 8:43 PM, David Rodrigues wrote:
> Nice example! > >> You're saying that "public private(set)" would become available in 8.2.1 > (the late-December release of 8.2), but only in files that have > "declare(asymmetric_visibility=1);" at the top. > > Yes. In that case, I should suggest declare(preview_asymmetric_visibility = > 1), just to make clear that it is using a preview feature. > >> However, we could still change the syntax from "public private(set)" to > "public private:set" in 8.2.6 if we decided to; that wouldn't be considered > a BC break. But once we get to 8.3.0, whatever the syntax is at that point > is frozen and no longer changeable, and available in all files. The > declare() is now irrelevant. > > Partially. As the usage definition has not yet been decided between > private(set) vs. private:set, so this feature is not ready for preview. > Unless the idea of allowing the two syntaxes to co-exist in this feature is > acceptable (which I personally think is a bad idea). > > > Atenciosamente, > David Rodrigues
(Please don't top post.) Well, the syntax has been decided at this point, by the poll conducted earlier. But it was the only reasonable "otherwise BC breaking" change example I could think of. :-) So if your proposal would not allow that hypothetical change, what would it allow? --Larry Garfield

David Rodrigues

3 years ago
From what I understand, your concern is "what if we need to change the direction of something already decided"? For example, initially it was decided and very well accepted private(set), but after a while the idea was revised and it was decided that private:set would be better instead. In this case, I still believe that the ideal is that this does not change. But we have a options to decide: (1) Do not allow changes of this type, as private(set) was initially well accepted and, unfortunately, nothing can be done. Perhaps, with the exception of code conflicts that might be discovered later. (2) Allow BC if the change is reasonable (in this example, if some sort of code conflict is later discovered). AND: (2.a) Keep the old syntax (private(set)) at least until the official version is released, which will use private:set if possible. (2.b) Keep the old syntax for at least a few releases (eg. patch+3), or a few months (eg. 6 months) if possible. (2.c) Consider BC permanent and users will need to change existing codes immediately. In particular, I believe that option (1) is ideal, or in the last case (2.a). While option (2.c) would be easier to maintain and also acceptable, as users using preview features will be using it at the risk of things like this happening. In general, the proposal is to allow new features to be tested more easily and earlier, as long as they are ready for use and in their final version (or very close to it). Something like a "Release Candidate" version of this new feature. (By top posting, you mean what gmail does by keeping the previous message at the end of the email, right? I'll be more careful, thanks!)

Jordan LeDoux

3 years ago
On Tue, Oct 11, 2022 at 6:44 PM David Rodrigues <david.proweb@gmail.com> wrote:
> > Partially. As the usage definition has not yet been decided between > private(set) vs. private:set, so this feature is not ready for preview. > Unless the idea of allowing the two syntaxes to co-exist in this feature is > acceptable (which I personally think is a bad idea). > >
If a "preview" doesn't allow us to make breaking changes, then what exactly is the point? I don't see any benefit at all to this without that. If the "preview" is *actually* just "put out an RFC in the next patch release as soon as it's merged to master", which is what it seems you're saying (as that seems like all that's left with all the things you said we can't do in a preview), then that seems dubious, prone to instability in the engine outside of the preview features, and a total breakage of the release cycle and RM process that is currently in place. Jordan

Alex Wells

3 years ago
> On 12 Oct 2022, at 11:06, Jordan LeDoux <jordan.ledoux@gmail.com> wrote: > > If a "preview" doesn't allow us to make breaking changes, then what exactly > is the point? I don't see any benefit at all to this without that. > > If the "preview" is *actually* just "put out an RFC in the next patch > release as soon as it's merged to master", which is what it seems you're > saying (as that seems like all that's left with all the things you said we > can't do in a preview), then that seems dubious, prone to instability in > the engine outside of the preview features, and a total breakage of the > release cycle and RM process that is currently in place.
100% agree. Kotlin's experimental features are drastically different from what was described by David. I don't think "feature previews" deliver any value to PHP's development process here; mainly, it doesn't allow breaking changes, thus not allowing developers to be at ease when accepting features due to knowing they could make changes as they go, which I see as the main benefit. It would have helped deliver features which were otherwise declined due to someone not liking the syntax, not seeing the usefulness or just not convinced. If a feature is proven to not be a good fit for PHP then it's removed. Feature previews as I see it solve exactly one problem: feature's runtime stability - but I don't think this is a problem of PHP in the first place.

juan carlos morales

3 years ago
As I previously wrote in the old Thread, ... I would like to say that we should keep the release process as it is, until a true, real, use case shows up. Whoever participated in the threads is already aware about the so many possibilities, and adjustments to the release/work process , that have to be done; because of that, such work should be at least justified with a real case. The actual release process is consistent, secure, predictable, not perfect ... sure, but to justify such a radical change, I believe we should have a real use case.