Skip to content

Pull Request Checklist

Structure Aspect

  • Names of concepts are self-explanatory.
  • Concepts have aliases.
  • The cardinality of children and references was checked.
  • Properties need to be properties and shouldn't be referencable ( use children or references otherwise).
  • Concepts/Interfaces are located in the correct virtual package.
  • New concepts were added to the language that groups similar concepts e.g. data mappings to the data mappings language.
  • Concepts have a short description.
  • Enumerations are the right choice and should not be replaced by concepts.
  • A decision was made between constraint datatypes and property constraints.
  • Multiple references disable smart references. A custom substitute menu was used instead.

Editor Aspect

  • Code completion works.
  • Deletion of elements (doesn't) work.
  • No duplicated entries in the code completion menu.
  • Code completion menu descriptions are sufficient to differentiate entries.
  • Collapsible blocks work correctly.
    • Error messages still have enough information when collapsed.
    • Cardinality inconsistency: the real value vs. the expected value is shown in the error message.
  • Substitute menus vs. grammar cells was decided
  • A decision was made on how to enter stuff
    • for example: code completion vs. context assistant cell vs. side transformations.
  • Intentions are used when side transformations are too confusing for the user.
  • Links are consistent (e.g. always use ctrl + click).
  • Strange errors in the editor are fixed or reported (maybe upstream platform or MPS issue).
  • Important entries in the code completion menu are at the first positions (priorities).
  • Code completion + side transformations: longer descriptions are used instead of just single words.
  • The feature was tried from a user perspective. What annoyed you? Can you enter everything and not too much?
  • Check usability: Nothing should be deletable when it is mandatory. It also should not be deletable and raise an error.
  • The user can input something wrong and is notified instead of preventing him from entering something.

Constraints Aspect

  • Scopes tests were added when many scopes were changed or when the logic was complicated.
  • Constraints are set correctly for new features
    • For example: only specific containers, parents, links, etc. are allowed.
    • By default, the scope of the model is used.
  • Scopes work in tests and generators.
  • Properties have constraints for their values or implement common interfaces/concepts like INamedConcept.
    • Alternatively, the constraints can be enforced on the type system level.
  • When using KernelF: unnecessary concepts are filtered through the can be ancestor constraint.

Behavior Aspect

  • All abstract methods are implemented.
  • Methods that should be overridable have the virtual flag set.
  • Methods that should override a super method have been marked as overrides.
  • Default implementations exist for used concepts that have abstract unimplemented methods.
  • Usage of static flags was checked for all methods.
  • Methods are generalized correctly and are moved to a super concept if needed.
  • Methods that return a collection return an empty list if there is no result.
  • Languages with multiple generators don't pollute the behavior aspect with generator-specific methods.

Typesystem Aspect

  • Tests were added for every single new type system error/warning.
  • Checks are displayed at places where it is clear where the error is. Alternatively: more information was added to make the error locatable.
  • Structural checks consider all relevant properties/children/references.
  • When using types, it is checked if they must match exactly or just be subtypes.
  • Checks cover all corner cases.
  • Checks were done for the right concepts e.g. enum literals vs. enum type and are displayed at the right node.
  • A quick fix or an (error) intention is used and not both.

Plugin Aspect

  • A plugin solution is used instead of a plugin aspect.
  • Actions were added to groups so that the user could invoke them from the UI.
  • The right type of plugin was chosen (project vs. application plugin).
  • All functionality in plugin solutions/aspects was manually tested and don't throw errors.
  • Actions have keyboard shortcuts.
  • Actions have the right kind of access (command, editor command, read, none).
  • All actions were disabled (NonDumbAwareActions) that shouldn't be available while indexing.

Test Aspect

  • All corner cases are tested (see: testing).
  • Test methods have reasonable names that explain what they are testing.
  • Test methods don't test too many different things.
  • Test aspects have each a TestInfo node.
  • Complex migrations are tested using a MigrationTestCase.
  • Complex scopes are tested using scope tests.
  • has error and has warning checks have references to the error statement.
  • The before tests block is used to set up tests.
  • The after tests block is used to clean up tests.
  • Temporary models created in tests are disposed of after testing.
  • The generator was called manually for tests where the generator tests were not fine-grained enough.
  • Visual diffing through the MPS diff tool is integrated for more complex tests.

Generator Aspect

  • Tests were added/changed when modifications were made in generators.
  • Preprocessing and postprocessing scripts are used sparingly.
  • Custom generation plans are used when needed or priority rules are set.
  • Performance impacts were checked (see: performance
    • Slow operations in the generator were saved in variables/in a cache.
    • Code is not executed more often than needed.
  • Mapping configurations have an applicable block set.
  • The generator runs deterministically.
  • Necessary attributes are not lost between generation steps.
  • Pick output is used instead of get output to have scoping checks.
  • Naming convention: for templates meant to be used via explicit calls from other templates instead of in reduction rules, the "map_" prefix was instead of "reduce_".
  • No usages of .forEach in generator use foreach var in iterator (performance)
  • COPY_SRCL was used instead of LOOP+CALL
  • Model pruning is used to increase the performance.
  • The exact generator settings were decided (strict mode, parallel mode, apply transformations in place, etc.).

Intentions Aspect

  • The is applicable block was set up correctly.
  • Helpful additional intentions were added such as an intention to remove all items in a list.
  • Intentions were grouped using the language com.mbeddr.mpsutil.intentions.
  • When changing the model, the focus and selection are correctly set using the editor context.
Documentation Aspect
  • Documentation was written for new features.
  • There are no instances of technical details and terms in the documentation that the users don't understand.
  • New features that are not mentioned in the documentation are self-explanatory.
    • Generated files are self-explanatory as well or contain comments.
  • Screenshots/embedded nodes are used and have captions.
  • It was decided which kind of documentation is necessary
    • For example: Word documents, HTML pages, and documentation inside MPS.
  • Interactive exercises are provided for users.
Actions Aspect
  • For common substitutions, node factories are set up.
Feedback Aspect
  • The aspect was used to override error messages that are not helpful for the user (e.g. structural checks, failing property constraints).
Refactorings Aspect
  • Refactorings were used instead of directly renaming concepts/properties etc. (they update the resolve info too).
Scripts Aspect
  • Enhancement scripts were used to let the user perform changes to the models in batches.
Find Usages Aspect
  • To help the user find specific nodes more easily, custom finders were added.
Data Flow Aspect
  • Unreachable statements/expressions are checked.
  • Unitialized reads are checked.
  • Unused assignments are checked.
  • Unused variables are checked.
  • Returns are checked.
  • The Data Flow check is integrated into the type system.
  • Nullable analysis is implemented.
TextGen Aspect
  • TextGen is used for scenarios where the generator is not needed.
  • The class TextGeneratorEngine is used to get the generated text when needed.
  • Text gen components have the right extension and file encoding.
  • The generated files have valid names according to the operation system or used software.
Migration Aspect
  • Refactorings were used to move methods, concepts, etc. to not break models.
  • For deprecated concepts is was decided if they need an error message, if entering them should be disallowed, or if they should be removed completely.
  • Migrations were tested locally or through migration tests.
  • New migrations were executed in the project.
  • Old code is not immediately deleted but deprecated first. The code is deleted after a certain period.
  • When using the refactoring menu, a migration script or refactoring log was written for situations where existing models need to be updated.
  • Users are not using different distributions with different language versions to create models.
Accessory Models
  • It was decided how to ship libraries for users.
    • Should they be customizable?
    • Part of the user repository?
    • Be Overridable?
Listener Aspect
  • Listeners are the right choice and simple buttons, intentions or quick fixes can't solve the problem.
  • Constructors, node factories, reference handlers, and similar features can't solve the problem.
  • Listeners are not added for BaseConcept to preserve performance.
  • Updates of the models don't cause jumping cursors or lost focus in the editor.
  • Big updates are done in background threads to not freeze the editor.
Last update: October 2, 2023

Comments