fix(BLO-803): add secondary sort by ID for deterministic handler ordering

[Copilot]

Fix Deterministic Ordering for Generators with Same Weight

Implementation Plan

• Explore repository structure and understand the issue
• Confirm the bug in registry.go at line 40-50 where map iteration causes random ordering
• Verify current test infrastructure and existing test patterns
• Fix the ListConversationHandlers() function in registry.go to add secondary sort by ID
• Create unit tests in registry_test.go to verify deterministic ordering
• Verify the fix doesn't break any existing tests
• Build the server to ensure no compilation errors
• Remove accidentally committed binary and update .gitignore
• Run code review to ensure quality
• Run CodeQL security scan
• Fix failing integration test in evm-events-calls

Problem

In registry.go, the ListConversationHandlers() function iterates over a map (which has non-deterministic iteration order in Go) and then sorts only by weight. When multiple handlers have the same weight (e.g., evm-events-calls and evm-events-calls-raw both have weight 82), they would randomly swap positions between runs, leading to inconsistent user experience.

Solution

Added a secondary sort key (handler ID) to ensure deterministic alphabetical ordering within the same weight group. The sort now:

1. First sorts by weight (highest first) - primary sort
2. Then sorts alphabetically by ID for handlers with the same weight - secondary sort

This ensures consistent user experience where choices appear in the same order every time.

Changes Made:

1. Modified registry.go: Updated ListConversationHandlers() function to add secondary sort by ID for deterministic ordering

   • Primary sort: Weight (highest first)
   • Secondary sort: ID (alphabetically) for same weight

2. Created registry_test.go: Added comprehensive unit tests

   • TestListConversationHandlers_DeterministicOrdering: Verifies consistent ordering across 10 runs
   • TestListConversationHandlers_WeightOrdering: Verifies weight-based sorting
   • TestListConversationHandlers_RealWorldEVMScenario: Tests actual EVM generator scenario

3. Updated .gitignore: Added compiled binaries to prevent future accidents

4. Fixed evm-events-calls/convo_test.go: Corrected TestInvalidWrappedABIFormat test

   • Test was incorrectly expecting a SeqMsg wrapper when error messages are returned directly as *pbconvo.SystemOutput
   • Updated test to properly handle direct SystemOutput return

Verification:

✅ All new tests pass
✅ Build succeeds without errors
✅ All tests pass (including the previously failing integration test)
✅ Binary removed and .gitignore updated
✅ Code review passed with no comments
✅ CodeQL security scan found 0 vulnerabilities

Security Summary

No security vulnerabilities were introduced or found in this change. CodeQL analysis completed with 0 alerts.

Fixes https://linear.app/streamingfast/issue/BLO-803/substreams-codegen-ordering-issue

---

💬 We'd love your input! Share your thoughts on Copilot coding agent in our 2 minute survey.

[maoueh]

@copilot There is failing integration tests, I don't think it's related to your work, but can you inspect and fix them.

[Copilot]

@copilot There is failing integration tests, I don't think it's related to your work, but can you inspect and fix them.

Fixed in commit 158787a. The test was incorrectly expecting the error message to be wrapped in a SeqMsg, but error messages from c.Msg().Messagef() are returned directly as *pbconvo.SystemOutput. Updated the test to handle the direct return value.