Mint every SED-ML id from one document-wide allocator

[jcschaff]

Last of the nightly SEDML_SBML_IT failures. biomodel_28625786 has an application and a simulation both named compartmental, so the exported archive carried

<model id="compartmental" name="compartmental" .../>
<uniformTimeCourse id="compartmental" name="compartmental" .../>

which the OMEX validator rejects: "Each identified SED object must have a unique id. Multiple objects have the following ids: [compartmental]".

Why the previous fix didn't cover it

SED-ML SIds share one namespace across every kind of object, but SEDMLExporter built them from several independent recipes — application names, simulation names, mangled variable names, per-kind counters — with no check that two hadn't landed on the same string. #1840 added a uniqueness guard within the application namespace only; simulation ids were still minted unguarded in createSedMLSim.

Rather than repeat that guard per kind, every mint now goes through one allocator:

private String uniqueId(String base, String kindTag) {
    String candidate = base;
    int collisionCount = 0;
    while (!this.usedSedmlIds.add(candidate)) {
        candidate = base + "_" + kindTag + collisionCount++;
    }
    return candidate;
}

A free base is returned unchanged, so ids move only where they previously collided — a correct archive is unaffected. Applications are reserved up front (reserveSimContextIds) so they keep first refusal on their name, which matters because an application's id also names the SBML file written for it. #1840's _app<n> is now one case of the general _<kind><n>.

For biomodel_28625786 the entire id diff against master is one line:

duplicate ids NEW: NONE
compartmental model : ['compartmental', 'compartmental_0', 'compartmental_1']
compartmental sim   : [('compartmental_sim0', 'compartmental')]
ids removed vs master: []
ids added   vs master: ['compartmental_sim0']

Latent collisions this also closes

Neither had been hit by a test yet:

• Override-derived models are <simContextId>_<overrideCount>, which can land on another application's mangled id. This very model has applications simple_1 and simple_1.5 → simple_1_5; a sixth override variant of simple_1 would have collided.
• Data generator ids are dataGen_<task>_<mangled var>, and TokenMangler.mangleToSName is many-to-one — species a.b and a-b both mangle to a_b.

The 'time' data generator is now looked up by the id it was actually given rather than by rebuilding that id at the use site, since the allocator is free to move it.

Test expectations

biomodel_82065439 had the same duplicate-compartmental fault recorded and now round-trips cleanly, so it failed with "passed SEDML Round trip, but knownSEDMLFault was set" — entry removed. (This is the usual signal on this suite: a fix surfaces as a failure.)

biomodel_28625786 still fails, on a validator bug rather than on the archive, and is recorded as a known fault with the diagnosis inline. biosimulators-utils 0.2.3 checks every repeatedTask setValue target against the last model in the document instead of the one the change references — sedml/validation.py:509 passes a leaked model loop variable where it means change.model. Here that lands on HPC_070907; species targets happen to exist there so they pass, the kinetic parameters do not, and the errors are then reported against models (simple_1, simple_1_5) that were never searched. Verified with lxml that those parameters do resolve against their own model sources:

<parameter constant="true" id="Km_PKA_activates_Raf" units="Unit_umol_l_1" value="0.5"/>

0.2.3 is the latest release on PyPI, so there's nothing to upgrade to; the entry comes out when the upstream fix lands.

Testing

PR CI cannot exercise this — SEDML_SBML_IT runs only in regression.yml, and these models sit behind test.include.slow. Locally:

mvn -o test -pl vcell-core -Dgroups=SEDML_SBML_IT -Dtest=SEDMLExporterSBMLTest \
    -Dtest.include.slow=true -Dtest.only=biomodel_28625786,biomodel_82065439,biomodel_220138948

The duplicate-id error is gone; biomodel_82065439 now passes outright. The full 264-model group is still running — any other model whose fault this resolves will announce itself the same way, and I'll push a follow-up commit if more turn up before merge.

🤖 Generated with Claude Code

https://claude.ai/code/session_01SvqmME7MkRUNEYje5HpiLt

[jcschaff]

Full-group verification complete — the run flagged as pending in the description has finished:

mvn -o test -pl vcell-core -Dgroups=SEDML_SBML_IT -Dtest=SEDMLExporterSBMLTest -Dtest.include.slow=true

[INFO] Tests run: 264, Failures: 0, Errors: 0, Skipped: 0, Time elapsed: 7679 s
[INFO] BUILD SUCCESS

All 264 models including the slow set, i.e. the full suite the nightly runs. No additional models had a knownSEDMLFault resolved by this change, so the two expectation updates already in this PR are the complete set and no follow-up commit is needed.

With this, the nightly SEDML_SBML_IT group goes from 6 failures to 0 (#1827, #1835, #1840, #1841, #1843, and this).