Repository navigation
Conversation
boxel-cli's local-mount server now passes each module's URL to transpileJS, so the classes it serves carry identity tags like a realm-server's do. Host tests tag their fixture classes as a loader records a module's exports, covering every shim path rather than only the test adapter's, and tagging a class with the first module its loader credits it to. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
A class a factory makes, as in `export const Status = enumField(...)`, has no body to put the identity member in. The plugin now follows such a binding with a statement that tags its value once it exists, when that value is a function carrying no identity of its own. `export default makeClass()` is bound to a name first so the statement can reach it. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The gallery job paired this commit's host harness with a published CLI's test engine. The engine compiles the realm's modules for the host, which now identifies the classes they export by what the compiler records in them, so a published CLI's compiler no longer fits a newer host. Build the CLI here instead of installing a release. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Compiled modules are cached in-process, in module_transpile_cache, and by clients holding their validators, all keyed by source alone. Tagging exported classes changes what unchanged source compiles to, so output from before it would outlive the change until a reindex wiped the cache. The module ETag now carries a transpiler version: a row or validator from an earlier version no longer matches and the module is compiled over. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
A realm now tags its modules with their URL, and the loader canonicalizes the tag when it reads it. What a module compiles to no longer depends on the realm prefixes registered at compile time, so a prefix registered later spells the module the same way identify() does. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
setupTestRealm tags the classes in its module contents under the module's realm URL, and tests that shim modules directly go through shimTestModule (or tagTestModuleExports for a lazily resolved shim). A class keeps the first tag it gets in a test, which is the module a loader credits it to. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Both realms in the operator-mode acceptance setup exported one class object as `person`, which no two modules can do: a class is declared by one of them. The second realm now exports a subclass, so each class's identity names the module that declares it. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
identify() now answers only from the identity tag compiled into a class, canonicalized through the loader's virtual network. The loader no longer records an identity for every function a module exports, and the bundled-base identity registry is gone, along with the build step that published it and the assertion that compared the two systems. getLoaderFor keeps the record of which loader loaded each export. The bundled-base tests now assert what compile-time tags make true: a class only a bundled sibling names is identified by the module that declares it, whichever modules the loader was asked for. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
identify() built a new object from a class's tag on every call, where the loader's record it replaced handed back one object per class. Callers compared identities by reference: the new-card menu's de-duplication of recent card types listed a type twice and dropped the next one. identify() now returns one object per class for as long as its module canonicalizes the same way, and the menu compares types by module and name. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
ef4
force-pushed
the
loader-identify-reform
branch
from
October 23, 2026 15:51
318ccaf to
a2d473c
Compare
It was written for the bundled-base identity registry, which is gone. Its fixtures now tag their own classes, so what it exercises is the test helper; a base class identified wrongly fails the many tests that load base cards. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The plugin tagged the value of any exported binding computed by a call or a reference, which emitted a runtime statement for template-only components, modifiers, symbols and data alongside the handful of factory-made classes. It now tags an exported binding only when its value is a call to a listed class factory, named by module and export: base's enumField and the catalog's statusField and priorityField. A call is recognized however the module reaches the factory: by a default, named or namespace import, by a realm alias or relative specifier, or as a function the module exports itself. The host names base modules, and resolves base's relative imports, with @embroider/reverse-exports; the realm canonicalizes imports through its virtual network. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
ef4
commented
Oct 23, 2026
| if: needs.change-check.outputs.choreo == 'true' || github.ref == 'refs/heads/main' || needs.change-check.outputs.run_all == 'true' | ||
| runs-on: ubuntu-latest | ||
| timeout-minutes: 15 | ||
| timeout-minutes: 25 |
Contributor
Author
There was a problem hiding this comment.
This PR changes some tests to build boxes-cli instead of consuming the published version. That probably slows down CI and is undesirable, but it's necessary until a published boxer-cli is compatible with the changes in this PR.
| let ref = identifyCard(card.constructor); | ||
| let name = cardTypeDisplayName(card); | ||
| if (isResolvedCodeRef(ref)) { | ||
| if (items.find((item) => item.ref === ref && item.name === name)) { |
Contributor
Author
There was a problem hiding this comment.
This implicitly assumed that we could trust object stability of ref. Mostly we do have that stability, because we're efficient about reusing objects, but it's more correct to compare by value.
| if: needs.change-check.outputs.choreo == 'true' || github.ref == 'refs/heads/main' || needs.change-check.outputs.run_all == 'true' | ||
| runs-on: ubuntu-latest | ||
| timeout-minutes: 15 | ||
| timeout-minutes: 25 |
Contributor
Author
There was a problem hiding this comment.
This PR switches to building boxel-cli in CI instead of consuming the published one, because the published one is not compatible with our changes.
We could go back to using the published one after this lands and we publish.
| // The second realm's module declares a class of its own, as any two | ||
| // modules do: one class object can only have been declared by one of | ||
| // them. | ||
| class SecondRealmPerson extends Person {} |
Contributor
Author
There was a problem hiding this comment.
The tests were creating a situation that's impossible in real realms -- the same class can't actually be declared in two different realms.
The plugin no longer carries a list of class factories of its own; it tags the results of the ones its caller names. The realm's transpile and the host's babel config both pass the list in runtime-common/class-factories.ts. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Resolving the base realm's alias inside the plugin made it import the realm prefixes, which brought the base card types into every package that type-checks the transpiler, and failed boxel-cli's lint. The plugin now only resolves relative imports; callers spell other forms through canonicalModule. The realm canonicalizes through its virtual network, and a compile without one maps the base realm's alias, beside the factory list. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Contributor
Author
|
This will see CI failures until #6936 is landed. |
ef4
marked this pull request as ready for review
October 23, 2026 17:16
Contributor
Author
|
@lukemelia this is ready for review and merge, it's the followup from my latest work with @backspace |
This branch has not been deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
This switches our strategy for identifying which module a card/field definition comes from.
Before it was a custom module loader extension, now it is a build step that tags classes with a symbol.
Since we already have required custom build steps, this is an overall reduction in ways we diverge from standard module loading, and it eliminates special handling that was previously required for base realm files inside host.