Reworked the sub-type relationship infrastructure #58

JohannesMeierSE · 2025-01-20T17:09:44Z

This PR contributes:

Sub-type relationships are now explicitly stored in the type graph
- They are not calculated by the asymmetric analyzeIsSubTypeOf and analyzeIsSuperTypeOf anymore
- Types store their sub-type-reationships themselves in the type graph
- no caching of (transitive) sub-type relationships anymore
- new feature: now sub-type relationships can be explicitly defined by the user of Typir
This is the precondition to improve assignability checks:
- arbitrary paths of implicit conversion and sub-type relationships can be exploited now for advanced assignability checks, by path search in the type graph
- implementation for finding and controlling best matches of overloaded functions/operators
moved the existing graph algorithms into its own dedicated service in order to reuse and to customize them (+ created variants of the existing algorithms)
I started to tackle the terminology issue with "problems" as result of e.g. sub-type checking by introducing SubTypeProblem and SubTypeSuccess interfaces, combined as type SubTypeResult = SubTypeSuccess | SubTypeProblem
provided some predefined language elements for reuse in test cases inside typir in packages/typir/src/test/predefined-language-nodes.ts (they are strongly inspired from our joint work @insafuhrmann)
more test utilities
more test cases

Aspects for the review:

How to produce meaningful subProblems for failed checks of sub-type and assignability? As an example, a user is interested, why two structually typed classes are not sub-types.
How to make the signatures in GraphAlgorithms in graph-algorithms.ts more type-safe?
@insafuhrmann conversion.test.ts contains a test case to check for cycles in conversion rules: Could you check, whether is reflects the kind of test cases you looked for you?

…tils

…ns/operators, more test cases for assignability

… order to reuse and to customize them

… in the type graph, exploit them for advanced assignability checks (WIP)

…ced assignability result+success(+problem)

Lotes

All in all just a bit coding style and a bit knowledge transfer -Typescript-wise. I still do not get the purpose of the 300+-changes file which I commented as latest. I hope you will find these comments helpful :). Thanks for your work.

packages/typir/test/kinds/function/operator-overloaded.test.ts

packages/typir/src/graph/graph-algorithms.ts

packages/typir/src/graph/type-node.ts

packages/typir/src/kinds/bottom/bottom-type.ts

packages/typir/src/kinds/class/top-class-type.ts

packages/typir/src/kinds/function/function-initializer.ts

packages/typir/src/test/predefined-language-nodes.ts

packages/typir/src/services/assignability.ts

packages/typir/src/services/subtype.ts

packages/typir/src/kinds/function/function-initializer.ts

insafuhrmann

Thanks for this work @JohannesMeierSE. I am pretty happy to get rid of analyzeIsSubTypeOf and analyzeIsSuperTypeOf. The rest looks fine to me too, where I had something to add, I placed a comment. Let's discuss the first two "aspects for review" in the meeting. I do not have a ready made solution at hand.

README.md

packages/typir/src/initialization/type-selector.ts

packages/typir/src/kinds/class/class-validation.ts

packages/typir/src/kinds/function/function-initializer.ts

insafuhrmann · 2025-01-24T11:13:28Z

packages/typir/src/kinds/function/function-initializer.ts

+            }
+            matchingOverloads.splice(index); // keep only the matches with the same highest priority
+            // TODO review: should we make this implementation more efficient?
+


If I understand correctly you are sorting the overloads and then iterate until priority drops (place where prio changes) to then cut the rest, right? I think it is OK, though it might be worth considering to use another data structure in the first place, in the way of a priority queue. Not sure it is worth it, though.

Maybe do the compare right away and if you get something better, drop everything you have so far and keep the new one instead, if equal, add, if less, drop the candidate, to avoid collecting all of them and having to sort. But in the end, how many candidates do we expect?

You are completely right, I used the algorithm you sketched and it is much easier than I thought. Now we have linear effort instead of square effort, thanks!

how many candidates do we expect?

The operator/function with the most overloaded signatures is usually "add" (+), which is applied to strings and numbers (and maybe one or two more special types). How many number types might be used in a DSL? Maybe 10, but probably not more than 20 ...

packages/typir/src/services/assignability.ts

packages/typir/src/test/predefined-language-nodes.ts

insafuhrmann · 2025-01-24T11:37:48Z

packages/typir/test/kinds/function/operator-overloaded.test.ts

+        });
+
+        test('integer to string', () => {
+            expectAssignmentValid(new IntegerLiteral(123), new StringLiteral('456'), 'SubTypeEdge', 'ConversionEdge');


A bit surprising, had to check back with the definitions above :). But OK for test purposes.

Yes, I agree. We could integrate "language types" into our predefined test nodes, but than we have another file some types in it and I wasn't sure, whether that is a good idea. But it would improve the readability of our test cases. What do you think?

Might be good to discuss together with @Lotes tomorrow. I think it might be worth the effort.

@Lotes and @JohannesMeierSE had no convincing idea for an easy solution today.

packages/typir/test/kinds/function/operator-overloaded.test.ts

insafuhrmann · 2025-01-24T11:40:15Z

packages/typir/test/services/conversion.test.ts

+        typir.Conversion.markAsConvertible(integerType, doubleType, 'IMPLICIT_EXPLICIT');
+        expect(() => typir.Conversion.markAsConvertible(doubleType, integerType, 'IMPLICIT_EXPLICIT'))
+            .toThrowError('Adding the conversion from double to integer with mode IMPLICIT_EXPLICIT has introduced a cycle in the type graph.');
+    });


Yep, this was what I'd been thinking of. Thanks!

If you like, feel free to contribute more test cases like this in a separate PR. I didn't investigated in this PR, how many test cases we need here 🙂

…ld use loops to define multiple relationships at once

…tions

…omments

…onsTypeInferenceRule, improved comments

JohannesMeierSE · 2025-01-27T10:07:32Z

Thanks @insafuhrmann and @Lotes for your helpful reviews!

I just pushed the corresponding improvements (and resolved the corresponding discussions here), but some discussions are still open. I suggest to continue them in our next joint meeting.

JohannesMeierSE · 2025-01-28T15:19:29Z

@Lotes @insafuhrmann I just pushed the last improvements according to our discussions today.

JohannesMeierSE added 14 commits January 10, 2025 14:11

some minor fixes

22b305a

started with some predefined language elements for testing and test u…

c6550d8

…tils

assignability informs about the reason for successful assignability now

67e9dbe

now sub-type relationships can be explicitly defined

92d7583

started implementation for finding best matches of overloaded functio…

1652024

…ns/operators, more test cases for assignability

fixed bug

02b5a3d

more test cases

7437b23

new test case for conversions with cyclic rules

20d7f41

extend conversion API to return all types to convert to

cc8874d

moved the existing graph algorithms into its own dedicated service in…

bf7c90d

… order to reuse and to customize them

reworked the management of sub-type relationships: use explicit edges…

0d0283c

… in the type graph, exploit them for advanced assignability checks (WIP)

make the whole assignability path in the type graph explicit, introdu…

8bc6e71

…ced assignability result+success(+problem)

sub-type relationships: check for cycles, controlled by a new option

6862143

improved README

e7434f3

JohannesMeierSE added New Feature New feature or request Infrastructure labels Jan 20, 2025

JohannesMeierSE requested review from Lotes and insafuhrmann January 20, 2025 17:09

JohannesMeierSE marked this pull request as draft January 21, 2025 12:15

JohannesMeierSE added 4 commits January 21, 2025 20:42

graph algorithms use only existing edges

fe96e20

removed methods of types to analyze their subType relationships

6ef3e51

improved comments, fixed bug

03dc6b6

improved test cases to explicitly check the resulting path

17449e5

JohannesMeierSE marked this pull request as ready for review January 21, 2025 19:59

Lotes reviewed Jan 22, 2025

View reviewed changes

insafuhrmann approved these changes Jan 24, 2025

View reviewed changes

JohannesMeierSE added 4 commits January 27, 2025 08:36

simplified APIs for sub-type and conversion, since users of Typir cou…

95644b6

…ld use loops to define multiple relationships at once

improved the algorithm to select the best matches of overloading func…

a018854

…tions

introduced test fixtures for literals

3d38164

fixed typos

3605669

JohannesMeierSE added 5 commits January 27, 2025 10:11

prefixed methods in listeners with on

fcbb582

made methods in TypeGraphListener optional

8ee03f9

refactoring: replaced nested ifs by chain of inverted ifs, improved c…

673f05f

…omments

refactoring to enable the exchange of the predefined OverloadedFuncti…

7852b23

…onsTypeInferenceRule, improved comments

added information about this PR into the CHANGELOG.md

4417beb

JohannesMeierSE added 2 commits January 28, 2025 16:12

simplified assignability test cases

1882722

refactoring: split existing file into two more files

145ac57

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Reworked the sub-type relationship infrastructure #58

Reworked the sub-type relationship infrastructure #58

JohannesMeierSE commented Jan 20, 2025 •

edited

Loading

Lotes left a comment

insafuhrmann left a comment

insafuhrmann Jan 24, 2025

insafuhrmann Jan 24, 2025

JohannesMeierSE Jan 27, 2025

insafuhrmann Jan 24, 2025

JohannesMeierSE Jan 27, 2025

insafuhrmann Jan 27, 2025 •

edited

Loading

JohannesMeierSE Jan 28, 2025

insafuhrmann Jan 24, 2025

JohannesMeierSE Jan 27, 2025

JohannesMeierSE commented Jan 27, 2025

JohannesMeierSE commented Jan 28, 2025

Reworked the sub-type relationship infrastructure #58

Are you sure you want to change the base?

Reworked the sub-type relationship infrastructure #58

Conversation

JohannesMeierSE commented Jan 20, 2025 • edited Loading

Lotes left a comment

Choose a reason for hiding this comment

insafuhrmann left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

insafuhrmann Jan 27, 2025 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

JohannesMeierSE commented Jan 27, 2025

JohannesMeierSE commented Jan 28, 2025

JohannesMeierSE commented Jan 20, 2025 •

edited

Loading

insafuhrmann Jan 27, 2025 •

edited

Loading