Code Review for phet-io instrumentation

[zepumph]

NOTE! Prior to doing a code review, copy this checklist to a GitHub issue for the repository being reviewed.

PhET code-review checklist

Build and Run Checks

• [x] Does the sim build without warnings or errors?
• [x] Does the sim start up? (requirejs and built versions)
• [x] Does the sim experience any assertion failures? (run with query parameter 'ea')
• [x] Does the sim pass a scenery fuzzer test? (run with query parameter 'fuzzMouse&ea')

Internationalization

• [x] Are there any strings that are not being internationalized? (run with query parameter 'stringTest=x', you should see nothing but 'x' strings)
• [x] Does the sim layout gracefully handle internationalized strings that are twice as long as the English strings? (run with query parameter 'stringTest=double')
• [x] Does the sim layout gracefully handle internationalized strings that are exceptionally long? (run with query parameter 'stringTest=long')
• [x] Does the sim layout gracefully handle internationalized strings that are shorter than the English strings? (run with query parameter 'stringTest=X')
• [x] Does the sim stay on the sim page (doesn't redirect to an external page) when running with the query parameter 'stringTest=xss' (test passes if sim does not redirect, OK if sim crashes or fails to fully start). Only test on one desktop platform.
• [x] Make sure the string keys are all perfect, they are difficult to change after 1.0.0 is published. Strings keys should generally match the values, such as {binaryProbability: "Binary Probability"}. Screen names should use camelcase, like so screen.screenName. Message patterns and long paragraphs will also use a different pattern.

Repository structure

• [x] Are all required files and directories present?

  For a sim repository named “my-repo”, the general structure should look like this (where assets/, audio/, images/, or dependencies.json may be omitted if the sim doesn’t have those types of assets).

   my-repo/
      assets/
      audio/
         license.json
      doc/
         model.md
         implementation-notes.md
      images/
         license.json
      js/
         (see section below)
      dependencies.json
      .gitignore
      my-repo_en.html
      my-repo-strings_en.json
      Gruntfile.js
      LICENSE
      package.json
      README.md

For a common-code repository, the structure is similar, but some of the files and directories may not be present if the repo doesn’t have audio, images, strings, or a demo application.

• [x] Is the js/ directory properly structured?

  All JavaScript source should be in the js/ directory. There should be a subdirectory for each screen (this also applies for single-screen sims, where the subdirectory matches the repo name). For a multi-screen sim, code shared by 2 or more screens should be in a js/common/ subdirectory. Model and view code should be in model/ and view/ subdirectories for each screen and common/. For example, for a sim with screens “Introduction” and “Lab”, the general directory structure should look like this:

   my-repo/
      js/
         common/
            model/
            view/
         introduction/
            model/
            view/
         lab/
            model/
            view/
         my-repo-config.js
         my-repo-main.js
         myRepo.js

• [x] Is there a file in assets/ for every resource file in audio/ and images/? Note that there is not necessarily a 1:1 correspondence between asset and resource files; for example, several images may be in the same .ai file.
• [x] Was the README.md generated by grunt published-README or grunt unpublished-README?
• [x] Does package.json refer to any dependencies that are not used by the sim?
• [x] Does sim-config.js refer to any dependencies that are not used by the sim?
• [x] Is the LICENSE file correct? (GPL v3 for sims, MIT for common code)
• [x] Does .gitignore match the one in simula-rasa?
• [x] Are the credits in sim-main.js correct?
• [x] Are there git repository branches that are no longer used and should be pruned?

Coding conventions

• [x] Does the code meet PhET's code-style guidelines, as described in the “Coding Style Guidelines” section of "PhET Development Overview"?
• [x] Is the code formatted according to PhET conventions? See phet-idea-code-style.xml for IntelliJ IDEA code style.

Documentation

• [x] Are documentation conventions followed, as described in the “Coding Style Guidelines” section of the PhET Development Overview?
• [x] Does model.md adequately describe the model, in terms appropriate for teachers?
• [x] Does implementation-notes.md adequately describe the implementation, with an overview that will be useful to future maintainers?
• [x] Are sim-specific query parameters (if any) identified and documented in one .js file in js/common/ or js/ (if there is no common/)? The .js file should be named {{REPO}}QueryParameters, for example ArithmeticQueryParameters for the aritmetic repository.

Math Libraries

• [x] Check that dot.Util.roundSymmetric is used instead of Math.round. Math.round does not treat positive and negative numbers symmetrically, see https://github.com/phetsims/dot/issues/35#issuecomment-113587879.
• [x] dot.Util.toFixed or dot.Util.toFixedNumber should be used instead of toFixed. JavaScript's toFixed is notoriously buggy. Behavior differs depending on browser, because the spec doesn't specify whether to round or floor.
• [x] Check that random numbers are generated using phet.joist.random, and are doing so after modules are declared (non-statically). For instance, the following methods (and perhaps others) should not be used:
• Math.random
• _.shuffle
• _.sample
• _.random
• new Random()

Organization, Readability, Maintainability

• [x] Does the organization and structure of the code make sense? Do the model and view contain types that you would expect (or guess!) by looking at the sim? Do the names of things correspond to the names that you see in the user interface?
• [x] Are appropriate design patterns used?
• [x] Is inheritance used where appropriate? Does the type hierarchy make sense?
• [x] Is there any unnecessary coupling? (e.g., by passing large objects to constructors, or exposing unnecessary properties/functions)
• [x] Is there too much unnecessary decoupling? (e.g. by passing all of the properties of an object independently instead of passing the object itself)?
• [x] Are the source files reasonable in size? Scrutinize large files with too many responsibilities - can responsibilities be broken into smaller delegates?
• [x] Are any significant chunks of code duplicated? This will be checked manually as well as with https://github.com/danielstjules/jsinspect
• [x] Is there anything that should be generalized and migrated to common code?
• [x] Are there any TODO or FIXME comments in the code? They should be addressed or promoted to GitHub issues.
• [x] Does the implementation rely on any specific constant values that are likely to change in the future? Identify constants that might be changed in the future. (Use your judgement about which constants are likely candidates.) Does changing the values of these constants break the sim? For example, see https://github.com/phetsims/plinko-probability/issues/84.
• [x] The simulation should use Property instead of PropertySet
• [x] Are all dependent properties modeled as DerivedProperty instead of Property?

Performance, Usability

• [x] Does the sim perform as desired across the range of supported platforms? (eg, not too slow on slow platforms, not too fast on fast platforms)
• [x] If the sim uses WebGL, does it have a fallback? Does the fallback perform reasonably well? (run with query parameter 'webgl=false')
• [x] Are UI components sufficiently responsive? (especially continuous UI components, such as sliders)
• [x] Are pointer areas optimized, especially for touch? (run with query parameter 'showPointerAreas')
• [x] Do pointer areas overlap? (run with query parameter 'showPointerAreas')
• [x] Is the timestep dt capped appropriately? Try switching applications or browser tabs, then switch back. Did the model take one big/long/awkward step forward? If so, dt may need to be capped. Example from faradays-law.FaradaysLawModel:

// Cap large dt values, which can occur when the tab containing
// the sim had been hidden and then re-shown
dt = Math.min( 0.1, dt );

Memory Leaks

• [x] Does a heap comparison using Chrome Developer Tools indicate a memory leak? (Describing this process is beyond the scope of this document.)
• [x] For each common-code component (sun, scenery-phet, vegas, …) that opaquely registers observers or listeners, is there a call to that component’s dispose function, or documentation about why dispose is unnecessary?
• [x] Are there leaks due to registering observers or listeners? These guidelines should be followed, or documentation added about why following them is not necessary:
  • [x] AXON: Property.link is accompanied by Property.unlink.
  • [x] AXON: PropertySet.link is accompanied by PropertySet.unlink.
  • [x] AXON: Creation of DerivedProperty is accompanied by dispose.
  • [x] AXON: Creation of Multilink is accompanied by dispose.
  • [x] AXON: Events.on is accompanied by Events.off.
  • [x] AXON: Emitter.addListener is accompanied by Emitter.removeListener.
  • [x] SCENERY: Node.on is accompanied by Node.off
  • [x] TANDEM: tandem.addInstance is accompanied by tandem.removeInstance.
• [x] Do all types that require a dispose function have one? This should expose a public dispose function that calls this.disposeMyType(), where disposeMyType is a private function declared in the constructor. MyType should exactly match the filename.

PhET-iO

• [x] If the simulation is supposed to be instrumented for PhET-iO, please see [How to Instrument a PhET Simulation for PhET-iO] (https://github.com/phetsims/phet-io/blob/master/doc/how-to-instrument-a-phet-simulation-for-phet-io.md) for the PhET-iO development process.

[zepumph]

@samreid do these credits seem alright to you?

credits: {
  leadDesign: 'Michael Dubson',
  softwareDevelopment: 'Michael Dubson, John Blanco',
  team: 'Wendy Adams, Mindy Gratny, Ariel Paul',
  thanks: 'Thanks to Mobile Learner Labs for working with the PhET development team\nto convert this simulation to HTML5.'
}

[samreid]

The credits seem reasonable, nothing unexpected jumping out at me.

[zepumph]

For each common-code component (sun, scenery-phet, vegas, …) that opaquely registers observers or listeners, is there a call to that component’s dispose function, or documentation about why dispose is unnecessary?

There is no dispose function for the ResistanceInAWireScreenView, but there is this call:

    this.addChild( new ResetAllButton( {
      listener: function() { model.reset(); },
      radius: 30,
      right: slidersBox.right,
      centerY: slidersBox.bottom + 60
    } ) );

I don't think more has to be done here, but there is no documentation about that listener being removed.

[zepumph]

@samreid there are just a few points I want to go over with you, otherwise this issue is basically complete.

[zepumph]

@samreid please close if you feel like all looks good.

[samreid]

I did some more refactoring in the above commits, SlidersBox still needs attention.

[zepumph]

@samreid in https://github.com/phetsims/resistance-in-a-wire/commit/588c5ce7900b8eea7cd4b32dc16f5df46b6a0e57 I noticed that you moved the Node.call() down to the end of the constructor. I am thinking that that is because the tandem must be passed as the last thing in the constructor.

Could you comment on your implementation vs:

function Slider(){

  Node.call( {x:x, y:y . . .}

... all of the constructor

  Node.mutate( {tandem: tandem})

[samreid]

I am thinking that that is because the tandem must be passed as the last thing in the constructor.

No, the tandem doesn't need to be last. The main reason for the change was to move from an imperative style to a declarative style. Here's a simplification of the change:

// imperative
var x = new Node();
x.addChild(child1);
x.addChild(child2);

// declarative
var y = new Node({
  children: [child1, child2]
});

The main principles for why to prefer the 2nd approach:

1. It is preferable to create the node with its correct state rather than creating it in the incorrect state and mutating it over a period of time.
2. Less opportunity for other code/bugs to sneak in between the addChild calls
3. Factoring out the addChild calls
4. Easy to see the structure of the node (more relevant when there are many lines of code between the addChild calls.

It is not always possible to use the declarative approach, but it seems generally preferable where applicable.

[zepumph]

@veillette did some great work on this front. I think it still needs another pass to make sure all is good. But I think we are close.

[zepumph]

@samreid I think this is ready to close, I refactored further and got things pretty crisp and clean. Please have a look and comment if there is more to be done. Thanks!

[samreid]

Things look much better, thanks! Closing.