docs(revdep): Analyze and document reverse dependency problems #2500

Copilot · 2025-11-16T15:44:39Z

Analyzed six packages with newly broken checks to determine root causes and assess whether issues stem from igraph bugs or downstream package issues.

Analysis

Created revdep/problems-analysis.md with detailed findings for all issues:

1. Cascade - Namespace Collision (Low Severity)

New circulant() export conflicts with magic::circulant. Non-breaking warning only.

Root cause: make_circulant() added in v2.2.1.9003 exports constructor alias that collides with magic package.

2. DiagrammeR - neighbors() Requires Single Vertex (High Severity)

Package breaks when passing multiple vertices to neighbors().

Root cause: igraph added stricter validation requiring exactly one vertex for neighbors(). DiagrammeR's get_leverage_centrality() passes a vector.

Recommendation: Iterate over vertices individually using lapply().

3. jewel - Integer Validation (High Severity)

Package breaks on non-integer niter values:

# Fails with "not representable as an integer"
rewire(g, keeping_degseq(niter = 2.45))

Root cause: rewire_impl() uses as.numeric() but C code now strictly validates integers. Previously silently truncated.

Recommendation: Use ceiling(p * 0.05) or add as.integer() in rewire_keeping_degseq() for backward compatibility.

4. manynet - Scalar Integer Validation (High Severity)

Package breaks when passing vector to sample_last_cit().

Root cause: igraph added stricter validation for scalar parameters. C code now validates that scalar parameters are indeed scalars.

Recommendation: Ensure scalar values or iterate if multiple values intended.

5. rSpectral - Modularity Changes (Medium Severity)

Tests fail due to different modularity values.

Root cause: modularity_impl() auto-detects weight attribute when weights = NULL:

if (is.null(weights) && "weight" %in% edge_attr_names(graph)) {
  weights <- E(graph)$weight
}

Workaround discovered: Passing weights = numeric() effectively disables auto-detection:

modularity(g, membership, weights = numeric())  # Forces unweighted calculation

This works because numeric() is not NULL (skips auto-detection), but !all(is.na(numeric())) is FALSE, causing the code to set weights <- NULL internally.

6. sfnetworks - from Parameter Requires Single Vertex (High Severity)

Package breaks when passing multiple vertices to all_shortest_paths().

Root cause: igraph added stricter validation requiring exactly one vertex for from parameter.

Recommendation: Handle single/multiple vertices explicitly or iterate.

Deliverables

revdep/problems-analysis.md: Root cause analysis, assessment, and recommendations for all six issues
revdep/examples/*.R: Simplified runnable minimal reproducible examples (no cat(), no tryCatch())
revdep/examples/*.md: Reprex-style markdown outputs showing expected results for each example
revdep/notify-maintainers.sh: Automation script to notify package maintainers via GitHub issues or email drafts
revdep/NOTIFY-README.md: Documentation for using the notification script

Example Format

All examples follow a simplified, clean format suitable for reprex::reprex():

No cat() statements: Documentation via comments only
No tryCatch() blocks: Error cases are commented out for clarity
Paired files: Each issue has both .R (runnable script) and .md (markdown output) files
Reprex-style: Clean, self-contained examples that demonstrate the issue

Issue Categorization

Category	Count	Packages
API Tightening	3	DiagrammeR, manynet, sfnetworks
Type Validation Bug	1	jewel
Namespace Collision	1	Cascade
Behavior Change	1	rSpectral

Maintainer Notification Script

Created notify-maintainers.sh to automate notifying package maintainers:

Intelligent routing: Determines upfront whether package is on GitHub, then creates either a GitHub issue or an email draft (not both)
GitHub Integration: Uses gh CLI to create GitHub issues directly for packages hosted on GitHub
Email Fallback: Generates email draft templates only for packages not accessible via GitHub
Efficient output:
- For GitHub packages: Issues created directly, no local files saved
- For non-GitHub packages: Only email draft files created in notifications/ directory
Template-based: Creates well-formatted issue/email content with:
- Problem summary and error messages
- Root cause analysis
- Minimal reproducible examples
- Suggested fixes with code examples
- Severity ratings
- References to igraph repository

Usage: ./notify-maintainers.sh

The script checks repository accessibility first and creates only the appropriate output type per package, eliminating redundant files.

See NOTIFY-README.md for detailed instructions on using the script, authentication setup, and troubleshooting.

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

Original prompt

This section details on the original issue you should resolve

<issue_title>Verify revdep problems</issue_title>
<issue_description>The file revdep/problems.md contains packages that now fail their checks, compared to the most recent CRAN version. Create minimal reproducible examples for the underlying problems and assess whether we uncovered a bug in a downstream package or if we inadvertently changed behavior.</issue_description>

Comments on the Issue (you are @copilot in this section)

Fixes Verify revdep problems #2499

✨ Let Copilot coding agent set things up for you — coding agent works faster and does higher quality work when set up for your repo.

Co-authored-by: krlmlr <[email protected]>

krlmlr · 2025-11-16T16:11:03Z

@copilot: Check if passing weights = numeric() would work for rSpectral. Update code and assessment as needed.

Co-authored-by: krlmlr <[email protected]>

Copilot · 2025-11-16T16:17:38Z

@copilot: Check if passing weights = numeric() would work for rSpectral. Update code and assessment as needed.

Yes! weights = numeric() works perfectly as a workaround.

The mechanism: numeric() is not NULL (skips auto-detection), but !all(is.na(numeric())) evaluates to FALSE, causing the code to set weights <- NULL internally.

Updated analysis and examples to document this workaround. Commit: 3351e96

krlmlr · 2025-11-16T20:23:39Z

@copilot: Create a script that uses gh and creates issues or PRs in the corresponding repositories. If not on GitHub, create a script that creates a GMail draft.

Co-authored-by: krlmlr <[email protected]>

Copilot · 2025-11-16T20:33:24Z

@copilot: Create a script that uses gh and creates issues or PRs in the corresponding repositories. If not on GitHub, create a script that creates a GMail draft.

Created notify-maintainers.sh script that:

Uses gh CLI to create GitHub issues for all three packages (Cascade, jewel, rSpectral)
Falls back to generating email drafts when GitHub is not accessible
Creates well-formatted issue templates with root cause, examples, and fixes

Usage: ./notify-maintainers.sh

Outputs to notifications/ directory:

{Package}-issue.md - GitHub issue template
{Package}-email.txt - Email draft

See NOTIFY-README.md for detailed instructions. Commit: 7ad2400

krlmlr · 2025-11-16T20:44:10Z

@copilot: Determine now if the package is on GitHub or not, the script should either create an issue or draft an e-mail, not both.

Co-authored-by: krlmlr <[email protected]>

Copilot · 2025-11-16T20:53:17Z

@copilot: Determine now if the package is on GitHub or not, the script should either create an issue or draft an e-mail, not both.

Updated script to determine upfront if package is on GitHub:

Checks repository accessibility first
Creates either GitHub issue (via gh) or email draft, not both
For GitHub repos: Issue created directly, no local files
For non-GitHub repos: Only email draft file created

Previously created 6 files (3 issue templates + 3 emails), now creates only what's needed. Commit: 1ad205a

krlmlr · 2026-01-27T08:06:25Z

@copilot: Updated results. Reassess. Simplify the reproducible scripts: avoid cat() in favor of comments, avoid tryCatch() in favor of running the script with reprex::reprex() and storing the Markdown output along with the script.

- Simplified example scripts: removed cat() in favor of comments, removed tryCatch() - Created markdown output files (.md) for reprex-style documentation - Added examples for 3 new issues: DiagrammeR, manynet, sfnetworks - Updated problems-analysis.md with all 6 issues - All examples now follow clean, runnable format suitable for reprex Co-authored-by: krlmlr <[email protected]>

Copilot · 2026-01-27T08:18:40Z

@copilot: Updated results. Reassess. Simplify the reproducible scripts: avoid cat() in favor of comments, avoid tryCatch() in favor of running the script with reprex::reprex() and storing the Ma...

Updated analysis for new revdep results. Simplified examples per feedback:

Changes:

Removed cat() - using comments instead
Removed tryCatch() - error cases are commented out
Added .md files with reprex-style output for each example
Clean, runnable format suitable for reprex::reprex()

New issues added:

DiagrammeR (neighbors() requires single vertex)
manynet (scalar integer validation)
sfnetworks (from requires single vertex)

All 6 issues now documented with simplified R scripts and markdown outputs. Commit: eead493

maelle · 2026-02-03T09:10:23Z

For the rSpectral workaround, do we want to implement an actual non hacky solution in igraph? Or to document and test this workaround? Just so we don't recommend a risky solution to a revdep.

maelle · 2026-02-03T08:59:12Z

revdep/NOTIFY-README.md

+The script will:
+
+1. Check if `gh` CLI is available
+2. For each package (Cascade, jewel, rSpectral):


what do these three names mean?

not suitable for long-term docs

maelle · 2026-02-03T08:59:27Z

revdep/NOTIFY-README.md

+
+1. Check if `gh` CLI is available
+2. For each package (Cascade, jewel, rSpectral):
+   - Check if the GitHub repository is accessible


from what data? DESCRIPTION only?

maelle · 2026-02-03T09:00:12Z

revdep/NOTIFY-README.md

+   - If accessible: Create a GitHub issue directly using `gh issue create`
+   - If not accessible: Create an email draft in `notifications/`
+
+The script determines upfront which action to take and only creates the appropriate output (either GitHub issue OR email draft, not both).


I'm wondering whether we should still send an email when there's an issue, but the email would link to the issue.

maelle · 2026-02-03T09:01:12Z

revdep/NOTIFY-README.md

+
+- `{Package}-email.txt` - Complete email draft with subject and body
+
+For packages with accessible GitHub repositories, issues are created directly and no local files are saved.


are links to issues stored somewhere? they could be useful for follow-up.

Also, are the issues opened on behalf of the person running the script? This should be specified in the docs, as it is important to know who will get notifications.

maelle · 2026-02-03T09:01:51Z

revdep/NOTIFY-README.md

+1. Review the email content in `notifications/{Package}-email.txt`
+2. Copy the content
+3. Create a new email in your email client
+4. Update the "To:" field with the actual maintainer email (check CRAN package page)


then this email address should be in notifications/{Package}-email.txt, not hard to automate, and a time gain for users.

maelle · 2026-02-03T09:02:24Z

revdep/NOTIFY-README.md

+2. Copy the content
+3. Create a new email in your email client
+4. Update the "To:" field with the actual maintainer email (check CRAN package page)
+5. Paste the subject and body


do we have an email address for igraph development that we could cc? or even use to send the emails? so that it's not a single team member who gets the answers.

maelle · 2026-02-03T09:03:20Z

revdep/NOTIFY-README.md

+- The repository may have issues disabled
+
+### "GitHub repository not accessible"
+- The repository might be private


Suggested change

- The repository might be private

- The repository might be private

- There might be no link to the repository in the package metadata

Initial plan

9af061a

Copilot AI assigned Copilot and krlmlr Nov 16, 2025

Copilot started work on behalf of krlmlr November 16, 2025 15:45 View session

Copilot AI and others added 2 commits November 16, 2025 15:57

Add comprehensive analysis and examples for revdep problems

fcd3c00

Co-authored-by: krlmlr <[email protected]>

Add README for revdep examples and finalize analysis

6ae5ab3

Co-authored-by: krlmlr <[email protected]>

Copilot AI changed the title ~~[WIP] Create minimal reproducible examples for revdep issues~~ docs(revdep): Analyze and document reverse dependency problems Nov 16, 2025

Copilot AI requested a review from krlmlr November 16, 2025 16:02

Copilot finished work on behalf of krlmlr November 16, 2025 16:02

Copilot started work on behalf of krlmlr November 16, 2025 16:11 View session

krlmlr mentioned this pull request Nov 16, 2025

feat: change constructors to use explicit arguments #2467

Merged

13 tasks

Add weights = numeric() workaround for rSpectral modularity issue

3351e96

Co-authored-by: krlmlr <[email protected]>

Copilot finished work on behalf of krlmlr November 16, 2025 16:19

Copilot started work on behalf of krlmlr November 16, 2025 20:24 View session

Add script to notify package maintainers about revdep issues

7ad2400

Co-authored-by: krlmlr <[email protected]>

Copilot finished work on behalf of krlmlr November 16, 2025 20:35

Copilot started work on behalf of krlmlr November 16, 2025 20:44 View session

Update notify script to create either GitHub issue OR email, not both

1ad205a

Co-authored-by: krlmlr <[email protected]>

Copilot finished work on behalf of krlmlr November 16, 2025 20:55

Merge branch 'main' into copilot/verify-revdep-problems

6fb1bf4

Copilot started work on behalf of krlmlr January 27, 2026 08:06 View session

Copilot finished work on behalf of krlmlr January 27, 2026 08:19

maelle reviewed Feb 3, 2026

View reviewed changes


		- `{Package}-email.txt` - Complete email draft with subject and body

		For packages with accessible GitHub repositories, issues are created directly and no local files are saved.

	- The repository might be private
	- The repository might be private
	- There might be no link to the repository in the package metadata

Uh oh!

docs(revdep): Analyze and document reverse dependency problems #2500

Are you sure you want to change the base?

docs(revdep): Analyze and document reverse dependency problems #2500

Conversation

Copilot AI commented Nov 16, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Analysis

1. Cascade - Namespace Collision (Low Severity)

2. DiagrammeR - neighbors() Requires Single Vertex (High Severity)

3. jewel - Integer Validation (High Severity)

4. manynet - Scalar Integer Validation (High Severity)

5. rSpectral - Modularity Changes (Medium Severity)

6. sfnetworks - from Parameter Requires Single Vertex (High Severity)

Deliverables

Example Format

Issue Categorization

Maintainer Notification Script

Comments on the Issue (you are @copilot in this section)

Uh oh!

krlmlr commented Nov 16, 2025

Uh oh!

Copilot AI commented Nov 16, 2025

Uh oh!

krlmlr commented Nov 16, 2025

Uh oh!

Copilot AI commented Nov 16, 2025

Uh oh!

krlmlr commented Nov 16, 2025

Uh oh!

Copilot AI commented Nov 16, 2025

Uh oh!

krlmlr commented Jan 27, 2026

Uh oh!

Copilot AI commented Jan 27, 2026

Uh oh!

maelle commented Feb 3, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Copilot AI commented Nov 16, 2025 •

edited

Loading

maelle commented Feb 3, 2026 •

edited

Loading