fix: write GML without explicit vertex IDs

[LeonidasZhak]

Summary

Allow write_graph(..., format = "gml") to use the GML writer's default vertex IDs when id is omitted. Fixes #2710.

Thanks to maintainers

Thanks for triaging the report, requesting a regression test, refreshing the branch, and re-running the checks.

Issue or motivation

Writing a graph as GML currently fails unless the caller supplies one ID per vertex. Earlier versions generated node IDs when none were supplied, and the C writer still supports that path.

Root cause

Stimulus treated the C writer's nullable id pointer as a required vector. The generated R wrapper converted NULL to numeric(0), so C received an empty vector instead of a null pointer and rejected its length.

Change

Mark id as optional in the Stimulus metadata and regenerate the R/C interface layers. The added regression test writes a graph without IDs and reads it back to confirm graph isomorphism.

The branch is now merged through igraph/rigraph@a52ba87 and remains mergeable against the current main; the PR diff is still limited to the GML interface metadata, generated wrapper/C interface, and the focused regression test.

Tests

• air format --check .
• Regenerated the Stimulus interface targets; the generated GML wrapper files remained reproducible.
• Focused test-foreign.R: 18 passed, 0 failed, 0 warnings.
• Full testthat::test_local(reporter = "check", stop_on_failure = TRUE): 9208 passed, 0 failed, 3 skipped.
• R CMD build: succeeded, including vignette creation.
• R CMD check --no-manual: package tests and vignette rebuild passed; local status was 2 WARNINGs from vendored GLPK/compiler diagnostics, with no errors or notes.
• GitHub Actions rcc run 33739775688, attempt 2: succeeded. Its jobs included the stock-R smoke test, matrix collection, pkgdown build, and R CMD check jobs on macOS, Windows, Windows ARM, Ubuntu, and Ubuntu ARM for R 4.2 through devel. The separate benchmark receive workflow was skipped.

Scope

This does not change explicit IDs, other graph writers, GML serialization details, vendored GLPK, unrelated lifecycle/CI behavior, or the igraph C library.

I used AI assistance to help check for issues, refine the code, and run tests.

I'm fine with GPL 2 or later and FDL.

• By submitting this pull request, I assign the copyright of my contribution to The igraph development team.

[schochastics]

LGTM if CI is green

[krlmlr]

Working on fixing checks on main, let's merge main into this branch when done to get it green.

[LeonidasZhak]

Summary

Merged the current main branch into this PR branch as requested. The GML fix itself is unchanged.

Thanks to maintainers

Thanks for reviewing the fix and for the guidance on when to refresh the branch.

Issue or motivation

The branch was seven commits behind main, including the recent package-cleanup changes, so its CI result no longer represented the current repository state.

Root cause

The previous PR checks ran against the older base at acd63244. The branch needed the current main history before CI could reassess the patch.

Change

Merged igraph/rigraph@e772e9e7 without conflicts. Relative to current main, the PR still changes only the four original GML interface/test files.

Tests

• air format .
• focused test-foreign.R: four non-skipped tests passed; six CRAN-gated tests skipped
• NOT_CRAN=true testthat::test_local(..., stop_on_failure = TRUE)
• full vignette build with R CMD build
• R CMD check --no-manual: tests and vignette rebuild passed; two existing vendored-GLPK warnings and one local rgl/X11 note remain

Scope

This update only syncs the PR with main. It does not alter the GML patch, workflows, vendored GLPK, or unrelated current-main CI failures.

[LeonidasZhak]

Quick follow-up on the CI gate: the PR is still mergeable and the GML patch is unchanged after the requested main sync. The rcc and Continuous Benchmarks (Receive) runs both currently show action_required, with no jobs or test logs created.

Could a maintainer approve or re-run those workflows when convenient? I’m happy to refresh the branch again if the current base requires it.