microsoft / microsoft/TypeScript
Identical resolved program produces 157 vs 662,155 diagnostics depending on whether the tsconfig is passed directly or through a pass-through extends
Personne n'a encore pris cette issue.
- Langage dominant
- Go
- Étoiles
- 111k
- Forks
- 14.3k
- Merge moyen
- 2 j 4 h
- PR mergées (30 j)
- 132
Description
On a large private monorepo package (~12k source files, React + Radix polymorphic component library), `tsgo --noEmit -p tsconfig.typecheck.json --pretty false` produces a stable 157 diagnostics. Creating a second config file in the SAME directory whose entire content is:
```json
{ "extends": "./tsconfig.typecheck.json" }
```
and running `--noEmit -p tsconfig.passthrough.json --pretty false` produces **662,155** diagnostics - 661,006 of them TS2859 "Excessive complexity comparing types" on comparisons as trivial as:
```
error TS2859: Excessive complexity comparing types '"confidential"' and
'"confidential" | "cui" | "internal" | "public" | "restricted" | "secret"'.
```
`--showConfig` for both invocations reports identical compilerOptions and an identical 8,743-file program (the only textual difference is path display relative to each config's location). Both results are stable across repeated runs - deterministic divergence keyed by config indirection, not flakiness.
Additional data points from the same program:
- A third shape - `extends` plus an explicitly enumerated `include` list that resolves to the same file set - produces a THIRD stable result (152 diagnostics), which additionally DROPS 5 real TS2353 errors that the direct run reports.
- Reproduced on `@typescript/native-preview 7.0.0-dev.20260707.2` (there the direct run was the clean-ish one) AND on stable `typescript@7.0.2` (there the DIRECT run explodes to the same 662,155).
- Composition-sensitivity at scale: growing the include set toward the full program flips diagnostic totals between 152 / 645,161 / 648,896 depending on which include units are present; the minimal trigger for the phantom `... is not assignable to type 'never'` class on polymorphic `as`-prop components (`const Component = as || Primitive.div`) is including the package's whole-fleet root barrel.
- tsc 5.9.3 cannot referee whole-program on this package (it dies at the V8 relation-cache Map cap - the motivating case for the native port); bounded subsets of the same files are green under tsc 5.9.3, 6.0.3, preview tsgo, and 7.0.2 alike.
Likely related ("Type Ordering"): microsoft/typescript-go#1605, microsoft/typescript-go#2830, microsoft/typescript-go#1278.
The repo is private, so this report ships observations rather than a runnable repro; happy to run instrumented builds or provide `--listFiles`/`--showConfig`/raw diagnostic captures for the A/B pair on request.
**Repro shape (for any large program):**
```bash
tsc-or-tsgo --noEmit -p tsconfig.json --pretty false | wc -l
printf '{ "extends": "./tsconfig.json" }\n' > tsconfig.passthrough.json
tsc-or-tsgo --noEmit -p tsconfig.passthrough.json --pretty false | wc -l
```
Guide de contribution
Ouvrir le guide de contribution
Par où commencer
- Lisez l'issue en entier, puis le guide de contribution du projet.
- Signalez en commentaire que vous la prenez — cela évite que deux personnes fassent le même travail.
- Forkez le dépôt et travaillez sur une branche.
- Ouvrez une pull request qui référence le numéro de l'issue.
Piste de recherche
Commencez par les exécutions A/B signalées de tsconfig.typecheck.json et tsconfig.passthrough.json, en comparant --showConfig, --listFiles et les diagnostics bruts pour les configurations directe et pass-through. Vérifiez la variante de include explicitement énumérée ainsi que les issues Type Ordering associées #1605, #2830 et #1278. C’est terminé lorsque la divergence des diagnostics dépendante de la configuration est identifiée et corrigée, tout en conservant le même programme et les diagnostics attendus.
Rédigé par le modèle d'indexation à partir du texte de l'issue.
Évaluation
- Stack technique
- go, react, typescript
- Domaine
- compilers
- Type d'issue
- Bug
- Difficulté
- 5/5
- Temps estimé
- Plus d'une semaine
- Activité
- Calme
- Clarté
- Plutôt claire
- Accessibilité débutants
- 35/100