#8539 Config preferences backend restructure #22923

peterjoel · 2019-02-21T22:04:22Z

A procedural macro for generating static structures for use as the backend for config preferences, as well a mapping from string names to accessors.

Preferences can be accessed and updated via a map-like interface with String keys, and now also via a convenience macro: get_pref!(path.to.pref). Various serde-compatible field attributes are also supported, including renaming the string keys. This could be useful when changing the backend structure without breaking existing usages.

I have added the choice to use i64 as well as f64 for storing numbers. As it turns out, none of the existing preferences used non-integer values. Setting a floating point value from a command-line argument requires a decimal point in order to parse correctly.

./mach build -d does not report any errors
./mach test-tidy does not report any errors
These changes fix #8539

There are tests for these changes OR
These changes do not require tests because ___

I have a few outstanding problems or questions:

I am unable to get rid of this warning:

warning: unnecessary path disambiguator
   --> components/config/prefs.rs:130:51
    |
130 |         accessor_type = crate::pref_util::Accessor::<Prefs, crate::pref_util::PrefValue>,
    |                                                   ^^ try removing `::`

See: https://stackoverflow.com/questions/54710797/how-to-disable-unnecessary-path-disambiguator-warning

Several of the preferences in use were not represented in prefs.json. Some of these may be in error, but it is hard to tell. For example js.offthread_compilation.enabled vs js.ion.offthread_compilation.enabled could be different preferences or could be intended to have the same value.
Previously, several pieces of code provided default values when accessing a preference that may not be present. For example:
```
let DBL_CLICK_TIMEOUT = Duration::from_millis(
    PREFS
        .get("dom.document.dblclick_timeout")
        .as_u64()
        .unwrap_or(300),
);
```
Fallback values don't really make sense now and I've added these defaults to prefs.json. Does this sound right?
I have kept PrefValue::Missing, though it doesn't seem very useful any more. An error might be more appropriate now for an incorrect preference key. I've kept this partly because webdriver_server uses it.

This change is

highfive · 2019-02-21T22:04:28Z

Heads up! This PR modifies the following files:

@jgraham: components/webdriver_server/lib.rs
@emilio: components/style/stylesheets/viewport_rule.rs, components/style/global_style_data.rs, components/style/properties/properties.mako.rs
@KiChjang: components/net/filemanager_thread.rs, components/net/lib.rs, components/script/dom/bindings/guard.rs, components/script/dom/htmlmetaelement.rs, components/script/dom/servoparser/mod.rs and 16 more
@asajeffrey: components/script/dom/bindings/guard.rs, components/script/dom/htmlmetaelement.rs, components/script/dom/servoparser/mod.rs, components/script/dom/mouseevent.rs, components/script/dom/htmlcanvaselement.rs and 15 more
@cbrewster: components/constellation/pipeline.rs, components/constellation/constellation.rs
@paulrouget: ports/servo/browser.rs, components/servo/lib.rs, ports/servo/non_android_main.rs, components/constellation/pipeline.rs, components/constellation/constellation.rs

highfive · 2019-02-21T22:04:30Z

Warning

These commits modify unsafe code. Please review it carefully!

asajeffrey · 2019-02-22T02:38:17Z

This looks like an improvement to me! @jdm, I believe you had some plans around config, is this PR in line with what you were thinking?

peterjoel · 2019-02-22T17:22:47Z

@asajeffrey There is some discussion with @jdm on the original issue #8539.

In the end, I was able to make both the static and key-based interface mutable.

bors-servo · 2019-02-26T07:11:17Z

☔ The latest upstream changes (presumably #22940) made this pull request unmergeable. Please resolve the merge conflicts.

bors-servo · 2019-02-27T18:50:37Z

☔ The latest upstream changes (presumably #22939) made this pull request unmergeable. Please resolve the merge conflicts.

bors-servo · 2019-02-27T21:32:15Z

☔ The latest upstream changes (presumably #22945) made this pull request unmergeable. Please resolve the merge conflicts.

asajeffrey

Mainly this looks really good! A few questions though...

components/config/pref_util.rs

asajeffrey · 2019-02-28T20:32:06Z

components/config/pref_util.rs

+
+pub struct Accessor<P, V> {
+    pub getter: Box<Fn(&P) -> V + Sync>,
+    pub setter: Box<Fn(&mut P, V) + Sync>,


Hmm, this is doing dynamic dispatch. Do we ever access a pref inside a hot loop?

Yes, prefs are accessed all the time during CSS parsing.

These accessors are only used for accessing preferences using their string keys. In most places the value is accessed directly from the structs (usually via the get_pref! macro.

I will take a look at the CSS code to check if it is using the string names.

I also realise that the accessors are used in every place that a preference is changed. In places where the preference name is statically known, I can actually change that to update the structs directly.

@asajeffrey With regard to accessing preferences in CSS, I don't follow the Python code generation very well. Are there existing benchmarks for the CSS to make sure that nothing is degraded? I can't see a ./mach command for that.

Don't know I'm afraid. @emilio?

I don't think I know of anything else that isn't reloading a page full of stylesheets over and over. In any case, looks like we don't have that many pref-gated properties, so probably not something that should concern us much.

@emilio can you point me to where some pref-gated CSS properties are implemented?

Sure, just grep for servo_pref= in components/style/properties. Looks like we have a few of them gated with layout.columns.enabled, and then layout.writing-mode.enabled.

components/config/pref_util.rs

components/config/prefs.rs

jdm · 2019-02-28T22:38:44Z

I just want to chime in and say that this was clearly more complex than I originally thought, and that I'm impressed by your solution!

emilio

It is a bit unfortunate to have to pay the cost of locking and arc-cloning to access simple prefs. Ideally bool prefs should be just a relaxed atomic read. Though given this simplifies the API and makes it easy to change it under the hood, it may be worth to land it and improve that as a follow-up...

components/config/prefs.rs

peterjoel · 2019-03-01T10:55:03Z

@emilio wrote:
It is a bit unfortunate to have to pay the cost of locking and arc-cloning to access simple prefs. Ideally bool prefs should be just a relaxed atomic read. Though given this simplifies the API and makes it easy to change it under the hood, it may be worth to land it and improve that as a follow-up...

The existing implementation stores the preferences in a Arc<RwLock<HashMap<String, Pref>>>, but each Pref also wraps the value in another Arc. Hopefully, this new implementation is at least no worse than that.

The choice was between locking the entire structure or locking individual preferences, and I went with the simplest one. Since most code can use the macros, as you say, it can be changed under the hood.

components/config/opts.rs

servo-wpt-sync · 2019-03-01T11:34:49Z

Opened new PR for upstreamable changes.

Completed upstream sync of web-platform-test changes at web-platform-tests/wpt#15617.

components/webdriver_server/lib.rs

servo-wpt-sync · 2019-03-01T12:10:10Z

Transplanted upstreamable changes to existing PR.

Completed upstream sync of web-platform-test changes at web-platform-tests/wpt#15617.

servo-wpt-sync · 2019-03-01T12:13:05Z

Transplanted upstreamable changes to existing PR.

Completed upstream sync of web-platform-test changes at web-platform-tests/wpt#15617.

servo-wpt-sync · 2019-03-01T12:30:35Z

Transplanted upstreamable changes to existing PR.

Completed upstream sync of web-platform-test changes at web-platform-tests/wpt#15617.

bors-servo · 2019-03-19T18:14:22Z

📌 Commit 73e47e9 has been approved by asajeffrey


        Auto merge of #22923 - peterjoel:issue_8539_prefs_refactor, r=asajeffrey

#8539 Config preferences backend restructure  A procedural macro for generating static structures for use as the backend for config preferences, as well a mapping from string names to accessors. Preferences can be accessed and updated via a map-like interface with `String` keys, and now also via a convenience macro: `get_pref!(path.to.pref)`. Various `serde`-compatible field attributes are also supported, including renaming the string keys. This could be useful when changing the backend structure without breaking existing usages. I have added the choice to use `i64` as well as `f64` for storing numbers. As it turns out, none of the existing preferences used non-integer values. Setting a floating point value from a command-line argument requires a decimal point in order to parse correctly. ---  - [X] `./mach build -d` does not report any errors - [X] `./mach test-tidy` does not report any errors - [X] These changes fix #8539  - [X] There are tests for these changes OR - [ ] These changes do not require tests because ___   ----- I have a few outstanding problems or questions: 1. I am unable to get rid of this warning: ``` warning: unnecessary path disambiguator --> components/config/prefs.rs:130:51 | 130 | accessor_type = crate::pref_util::Accessor::<Prefs, crate::pref_util::PrefValue>, | ^^ try removing `::` ``` See: https://stackoverflow.com/questions/54710797/how-to-disable-unnecessary-path-disambiguator-warning 2. Several of the preferences in use were not represented in `prefs.json`. Some of these may be in error, but it is hard to tell. For example `js.offthread_compilation.enabled` vs `js.ion.offthread_compilation.enabled` could be different preferences or could be intended to have the same value. 3. Previously, several pieces of code provided default values when accessing a preference that may not be present. For example: ```Rust let DBL_CLICK_TIMEOUT = Duration::from_millis( PREFS .get("dom.document.dblclick_timeout") .as_u64() .unwrap_or(300), ); ``` Fallback values don't really make sense now and I've added these defaults to `prefs.json`. Does this sound right? 4. I have kept `PrefValue::Missing`, though it doesn't seem very useful any more. An error might be more appropriate now for an incorrect preference key. I've kept this partly because [`webdriver_server` uses it](https://github.com/servo/servo/blob/master/components/webdriver_server/lib.rs#L224).  --- This change is [<img src="https://reviewable.io/review_button.svg" height="34" align="absmiddle" alt="Reviewable"/>](https://reviewable.io/reviews/servo/servo/22923)

bors-servo · 2019-03-19T18:29:12Z

⌛ Testing commit 73e47e9 with merge db1685a...

bors-servo · 2019-03-19T18:50:27Z

💔 Test failed - linux-rel-css

jdm · 2019-03-19T19:24:13Z

dom.testbinding.preference_value.quote_string_test


        #8539 Config preferences backend restructure

peterjoel · 2019-03-20T15:14:41Z

@jdm I also found browser.display.background-color and browser.display.foreground-color, which I've also added now. Hopefully that's the last.🤞

peterjoel · 2019-03-20T15:31:19Z

@jdm I've come across some mention of prefs in some Python tests, e.g: https://github.com/servo/servo/blob/master/tests/wpt/web-platform-tests/tools/wptrunner/wptrunner/tests/test_wpttest.py#L14 :

I'm not exactly sure what these are doing though, or how to make sure that I've run them properly. Are these trying to set and reset a pref called "a"?

jdm · 2019-03-20T17:05:09Z

Those are python-only unit tests for wptrunner and don't run against Servo.

jdm · 2019-03-20T17:06:04Z

@bors-servo r+

bors-servo · 2019-03-20T17:06:05Z

📌 Commit 8bfd4dc has been approved by jdm

bors-servo · 2019-03-20T18:33:39Z

⌛ Testing commit 8bfd4dc with merge 5ec7254...


        Auto merge of #22923 - peterjoel:issue_8539_prefs_refactor, r=jdm

#8539 Config preferences backend restructure  A procedural macro for generating static structures for use as the backend for config preferences, as well a mapping from string names to accessors. Preferences can be accessed and updated via a map-like interface with `String` keys, and now also via a convenience macro: `get_pref!(path.to.pref)`. Various `serde`-compatible field attributes are also supported, including renaming the string keys. This could be useful when changing the backend structure without breaking existing usages. I have added the choice to use `i64` as well as `f64` for storing numbers. As it turns out, none of the existing preferences used non-integer values. Setting a floating point value from a command-line argument requires a decimal point in order to parse correctly. ---  - [X] `./mach build -d` does not report any errors - [X] `./mach test-tidy` does not report any errors - [X] These changes fix #8539  - [X] There are tests for these changes OR - [ ] These changes do not require tests because ___   ----- I have a few outstanding problems or questions: 1. I am unable to get rid of this warning: ``` warning: unnecessary path disambiguator --> components/config/prefs.rs:130:51 | 130 | accessor_type = crate::pref_util::Accessor::<Prefs, crate::pref_util::PrefValue>, | ^^ try removing `::` ``` See: https://stackoverflow.com/questions/54710797/how-to-disable-unnecessary-path-disambiguator-warning 2. Several of the preferences in use were not represented in `prefs.json`. Some of these may be in error, but it is hard to tell. For example `js.offthread_compilation.enabled` vs `js.ion.offthread_compilation.enabled` could be different preferences or could be intended to have the same value. 3. Previously, several pieces of code provided default values when accessing a preference that may not be present. For example: ```Rust let DBL_CLICK_TIMEOUT = Duration::from_millis( PREFS .get("dom.document.dblclick_timeout") .as_u64() .unwrap_or(300), ); ``` Fallback values don't really make sense now and I've added these defaults to `prefs.json`. Does this sound right? 4. I have kept `PrefValue::Missing`, though it doesn't seem very useful any more. An error might be more appropriate now for an incorrect preference key. I've kept this partly because [`webdriver_server` uses it](https://github.com/servo/servo/blob/master/components/webdriver_server/lib.rs#L224).  --- This change is [<img src="https://reviewable.io/review_button.svg" height="34" align="absmiddle" alt="Reviewable"/>](https://reviewable.io/reviews/servo/servo/22923)

bors-servo · 2019-03-20T20:13:13Z

☀️ Test successful - android-mac, arm32, arm64, linux-rel-css, linux-rel-wpt, mac-rel-css1, mac-rel-css2, mac-rel-wpt1, mac-rel-wpt2, mac-rel-wpt3, mac-rel-wpt4, magicleap, status-taskcluster
Approved by: jdm
Pushing 5ec7254 to master...

jdm · 2019-03-20T20:37:13Z

Woohoo! Great work, @peterjoel.

highfive assigned asajeffrey Feb 21, 2019

highfive added the S-awaiting-review label Feb 21, 2019

highfive added the S-needs-rebase label Feb 26, 2019

peterjoel force-pushed the peterjoel:issue_8539_prefs_refactor branch 3 times, most recently from 89d7382 to 08e3c40 Feb 27, 2019

jdm removed the S-needs-rebase label Feb 27, 2019

highfive added the S-needs-rebase label Feb 27, 2019

peterjoel force-pushed the peterjoel:issue_8539_prefs_refactor branch from 08e3c40 to 2a1e5d8 Feb 27, 2019

peterjoel force-pushed the peterjoel:issue_8539_prefs_refactor branch from 2a1e5d8 to 30f2c23 Feb 28, 2019

asajeffrey reviewed Feb 28, 2019

View changes

emilio reviewed Mar 1, 2019

View changes

components/config/prefs.rs Show resolved Hide resolved

components/config/prefs.rs Outdated Show resolved Hide resolved

peterjoel reviewed Mar 1, 2019

View changes

components/config/opts.rs Outdated Show resolved Hide resolved

peterjoel reviewed Mar 1, 2019

View changes

components/config/opts.rs Show resolved Hide resolved

servo-wpt-sync mentioned this pull request Mar 1, 2019

#8539 Config preferences backend restructure web-platform-tests/wpt#15617

Closed

peterjoel reviewed Mar 1, 2019

View changes

components/webdriver_server/lib.rs Show resolved Hide resolved

peterjoel force-pushed the peterjoel:issue_8539_prefs_refactor branch from 101eb80 to 675b494 Mar 1, 2019

highfive added S-awaiting-merge and removed S-awaiting-review labels Mar 19, 2019

highfive added S-tests-failed and removed S-awaiting-merge labels Mar 19, 2019

#8539 Config preferences backend restructure

Loading status checks…

8bfd4dc

peterjoel force-pushed the peterjoel:issue_8539_prefs_refactor branch from 73e47e9 to 8bfd4dc Mar 20, 2019

highfive added S-awaiting-review and removed S-tests-failed labels Mar 20, 2019

highfive assigned jdm and unassigned asajeffrey Mar 20, 2019

highfive added S-awaiting-merge and removed S-awaiting-review labels Mar 20, 2019

highfive removed the S-awaiting-merge label Mar 20, 2019

jdm mentioned this pull request Mar 21, 2019

debugmozjs build is broken #23074

Closed

CYBAI mentioned this pull request Apr 4, 2019

about warnings caused by building #23161

Closed

servo / servo

Join GitHub today

#8539 Config preferences backend restructure #22923

#8539 Config preferences backend restructure #22923

Conversation

peterjoel commented Feb 21, 2019 • edited by SimonSapin

highfive commented Feb 21, 2019

highfive commented Feb 21, 2019

asajeffrey commented Feb 22, 2019

peterjoel commented Feb 22, 2019

bors-servo commented Feb 26, 2019

bors-servo commented Feb 27, 2019

bors-servo commented Feb 27, 2019

asajeffrey left a comment

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

jdm commented Feb 28, 2019

emilio left a comment

peterjoel commented Mar 1, 2019

servo-wpt-sync commented Mar 1, 2019

servo-wpt-sync commented Mar 1, 2019

servo-wpt-sync commented Mar 1, 2019

servo-wpt-sync commented Mar 1, 2019

bors-servo commented Mar 19, 2019

bors-servo commented Mar 19, 2019

bors-servo commented Mar 19, 2019

jdm commented Mar 19, 2019

peterjoel commented Mar 20, 2019

peterjoel commented Mar 20, 2019

jdm commented Mar 20, 2019

jdm commented Mar 20, 2019

bors-servo commented Mar 20, 2019

bors-servo commented Mar 20, 2019

bors-servo commented Mar 20, 2019

jdm commented Mar 20, 2019

peterjoel commented Feb 21, 2019 •

edited by SimonSapin