[adamjohnwright]

Spec 015 (specs/015-graph-exports), T069. With #754 (T070, merged), node's .biopax endpoint now has a service behind it. Nothing is routed until T071.

The service:

• GET /biopax/{stId} answers one event's BioPAX Level 3, from BioPAX3Exporter's library (main at 64c716a, built from source). It carries the release's per-event xml:base.
• Files are kept per release on disk, written beside their final name and moved into place, as SBML's are.
• SbmlFiles and SbmlController became ExportFiles and ExportController, with one ExportFiles per export.
• The cap of two conversions at once is shared by both exports, so the heap holds at most two whatever node sends.

The build:

• The pom holds Kotlin at the library's 2.4.21. Spring Boot 2.7 would bring 1.6.21.
• The Dockerfile cmps the library jar and checks the jar's kotlin-stdlib and paxtools-core versions.
• PINS.md has a new row, and pins.spec.ts follows. Its ARG pattern now allows a digit: as written it couldn't see BIOPAX3_EXPORTER.

check.sh:

• A BioPAX stage over BioPAX3Exporter's fixture graph: one event of each of five kinds across both species, each checked by that commit's contained.py against its species' expected file, with the per-event base, plus a 404 and a 400.
• A SERVICE_PORT setting, so it can run beside a deployment's own service.

Checked here:

• The image built, with its unit tests and jar checks.
• check.sh passed: SBML's three files valid and identical to expected, and the five BioPAX events contained and whole.
• Against beta's current image, which has no BioPAX, the check failed with five 404s.

Measured on a private copy of release 98's graph, in a container held to the compose limits (1.5 GiB, 1 GiB heap). The BioPAX and SBML of Signal Transduction and Metabolism were all asked at once:

• Memory: 1.10 GiB at peak, with no kill.
• Time: the cap converted Metabolism's pair in 156–158 s, then Signal Transduction's, finishing at 300–310 s.
• Consequence: in that worst case, the queued pair's readers get node's 503 ("took too long") at 290 s. The files are still made and kept, so the next request is answered at once. Alone, each conversion is about 150 s.

Unit tests:

• ExportFilesTest (11) and ExportControllerTest (8).
• New: an export with no diagram source never fetches one; each export keeps its own files; the cap is shared across exports; BioPAX is answered as RDF with SBML's errors.

Also, from #754's review:

• Node's comment on the export shares now says what they protect.
• BioPAX's own 503 messages are tested.
• The slashed form joins the forms of the id that are tested.

🤖 Generated with Claude Code