Parser improvements

[Jenson3210]

Summary

Parser fixes for failures found while analyzing a parse-failure report (org.openrewrite.table.ParseFailures) across ~18 open source repositories.

Docker (rewrite-docker)

• Whitespace before a comma in JSON arrays was lost (CMD ["a", "b" , "c"]), so the printed output differed from the input. A new CommaPrefix marker stores the whitespace before the comma on the preceding element, and DockerPrinter prints it back. Covers VOLUME, SHELL and the exec forms of CMD/ENTRYPOINT/RUN/HEALTHCHECK/COPY/ADD.
• = in ADD/COPY paths (ADD https://host/r?filepath=org/x.jar /dest) split the token, and ANTLR error recovery dropped text. sourcePath and destinationPath now accept EQUALS, and tokens touching an = are regrouped into one argument. The lexer is unchanged, so ENV/LABEL/ARG behave as before.
• DockerParser.java / .interp are regenerated.

JavaScript / TypeScript (rewrite-javascript)

• Source files are now decoded by decodeSourceBytes: a FF FE BOM is read as UTF-16LE, FE FF as UTF-16BE, and everything else as UTF-8 as before. Previously a UTF-16 file was read as UTF-8 and TypeScript reported diagnostic 1490 ("File appears to be binary").
• The BOM stays in the text as a leading U+FEFF, the same way a UTF-8 BOM already did, so the existing BOM handling round-trips it.
• Known limitation: the tree records the charset as utf8 for a UTF-16 file because the charset is detected after decoding. Printing is unaffected.

JSON (rewrite-json)

• Invalid JSON (YAML saved as .json, template placeholders, a missing closing brace, ...) previously produced either an opaque NullPointerException in JsonParserVisitor or a mangled tree that failed requirePrintEqualsInput.
• JsonParser now records ANTLR syntax errors and, before the visitor runs, fails the file with a ParseError: Syntax error in <path> at line L:C <msg> (plus a count of further errors).
• Behavior change: any input with an ANTLR syntax error is now a ParseError; before, some could still yield a document if it happened to print back identically. onError now receives one JsonParsingException per file. The existing suite (comments, single quotes, trailing commas, hex numbers) passes unchanged.

YAML (rewrite-yaml)

• Tagged block mappings (UserData: !Base64 followed by a newline and Fn::Sub:, !workspace, !!python/object/apply:...) re-printed the first key with extra indentation. The tag suffix was cut with a hard-coded 2-character indent, so it only worked for a mapping nested exactly two deep (a root-level tag threw StringIndexOutOfBoundsException). The suffix now runs up to the last newline before the first key.
• Helm templates: a new findHelmTemplateEnd() replaces the single-line regex. It handles multi-line {{/* */}} comments, multi-line {{- ... }} actions and ... -}} closers, and braces inside quoted strings (e.g. "${2}"). A {{ it cannot delimit falls back to the old regex. A standalone template line also no longer loses its leading/trailing whitespace, which caused print-idempotence failures.
• Not addressed: templates that share a line with other content (include ... | nindent), and non-YAML such as Jinja {% %}.

Testing

• New tests for every fix, written first and confirmed failing before the fix. rewrite-docker: 10 new tests (415 total, 0 failures). rewrite-json: parameterized syntax-error test over 9 inputs plus an empty-file test. rewrite-yaml: 8 tag tests and 13 Helm tests (737 total, 0 failures). rewrite-javascript: encoding.test.ts (UTF-16LE/BE, UTF-8 with and without BOM).
• Corpus check of rewrite-yaml against 5,809 YAML files containing {{ from the cloned repos: parse + print-identical went from 5,744 to 5,782, non-idempotent from 4 to 0, parse errors from 61 to 27, with no regressions. All 16 tag-bearing files now round-trip.
• Only each changed module's own tests were run; modules depending on the JSON/YAML parsers were not.

[Jenson3210]

User organization for:
https://github.com/aws-observability/aws-otel-collector
https://github.com/Azure/aad-auth-proxy
https://github.com/Azure/apiforfhir-migration-tool
https://github.com/Azure/data-product-streaming
https://github.com/Azure/employee-retention-databricks-kubernetes-poc
https://github.com/Azure/gitops-flux2-kustomize-helm-mt
https://github.com/Azure/prometheus-collector
https://github.com/Azure/SQL-Migration-AzureSQL-PoC
https://github.com/Azure/terraform-provider-azapi
https://github.com/chtrembl/azure-cloud
https://github.com/cilium/cilium
https://github.com/grafana/loki
https://github.com/grafana/mimir
https://github.com/grafana/tempo
https://github.com/IBM/zmodstack-deploy
https://github.com/kubernetes/autoscaler
https://github.com/kubernetes/kops
https://github.com/kubernetes/minikube
https://github.com/NVIDIA/k8s-nim-operator
https://github.com/openclarity/openclarity
https://github.com/openshift/origin
https://github.com/redhat-cop/agnosticd
https://github.com/VictoriaMetrics/VictoriaMetrics

Run FindParseFailures

[Jenson3210]

@timtebeek @greg-at-moderne, I identified these issues while testing a parser on my end, no rush / direct need for it. It's just this ParseFailures cluttered my output.

Feel free to close / cherry pick fixes that are worth picking.

[Jenson3210]

on the failed pipeline:

1. The CI run had empty credentials. The job log shows both ORG_GRADLE_PROJECT_codegenomeUsername: and ORG_GRADLE_PROJECT_codegenomePassword: with no value
2. The codegenome repo needs those credentials and serves jsonrpc.
3. The version that works isn't on Maven Central. I queried Central directly. jsonrpc-1.0.12.pom returns 200 and jsonrpc-1.3.0.pom returns 404. latest.release therefore resolves to 1.0.12 without codegenome
   access, and JsonRpc.tags(...), which arrived in 1.3.0, doesn't compile. My local runs had the credentials and got 1.3.0.