Langium 4 and Service registry.

[cdietrich]

Hi,

whilest evaluating the langium 4 alpha we see the following problem.

instead of our custom service registry we see this be used (at least in a test)

the uri is

_formatted =
'memory://lotsemetadata.langium'

which is langium itself.

=> now somehow a reference into the grammar exists at runtime that seems to be processed
(Grammar key)

i wonder what has changed that this does no longer work

maybe related to the singleton thing in the registry that was removed

[msujew]

Hey @cdietrich,

what is the call stack for the invocation of the service registry?

maybe related to the singleton thing in the registry that was removed

This was prior to 3.5.0 already, so nothing should have changed for the next release of v4.

[cdietrich]

@msujew can finally reproduce
https://github.com/cdietrich/lang4-ser-reg-jest

[cdietrich]

@msujew reproducer in langium codebase

test('xxxxx', () => {
    const from = URI.parse('inmemory://mygrammar.langium');
    expect(UriUtils.extname(from)).toBe('.langium');
});

[cdietrich]

test('xxxxx2', () => {
    const from = URI.parse('inmemory://aaaa/mygrammar.langium');
    expect(UriUtils.extname(from)).toBe('.langium');
});

this one does work, note the add. segment
=> maybe we need to have a different way for the in memory grammar uris

(loadGrammarFromJson)

[cdietrich]

indeed

services.shared.workspace.LangiumDocumentFactory.fromModel(astNode, URI.parse(`memory://grammars/${astNode.name ?? 'grammar'}.langium`));

seems to solve this one ...

[cdietrich]

or alternatiely using triple slash
const from = URI.parse('inmemory:///mygrammar.langium');

[msujew]

I see. Due to the test expectation error, Jest is attempting to stringify the AST nodes. This includes the whole CST, which in turn contains references to the loaded grammar. However, this grammar is not part of a real document, which leads to issues down the line.

I guess, it's a trivial fix, even though I'd argue that Jest shouldn't do that in the first place. Note that the original result for this call was incorrect as well (since it returned the language ID of the adopter language instead of langium).

[cdietrich]

actually (in 3.5)

[msujew]

See #1980 for a fix.