The Bug That Wasn't a Bug Yet: Hunting a Dormant API Contract Violation in Meshery

This bug had never fired. Every test passed. I found it by reading a style guide and taking it literally. It's meshery/meshery#21582, and my fix is PR #21583, which is still open.

The rule

Meshery's AGENTS.md says wire format is camelCase, and that includes URL query params. There was already a correct example in server/handlers/utils.go. getPaginationParams reads pageSize first and falls back to the legacy pagesize.

But that's one function. I wanted to know how many handlers skipped it:

grep -rn '\.Get("pagesize")' server/handlers/

The answer was 20 call sites across 14 files. They read only pagesize, with no fallback.

Why "dormant"

Today the UI's hand-written clients all send pagesize, so everything works. But #18526 tracks moving those clients onto hooks generated from @meshery/schemas, and those emit pageSize. The day one of them migrates, the handler gets an empty string, falls back to its default, and pagination quietly breaks. The migration PR would look correct in review, because it is. The bug sits in code nobody is touching.

I checked open issues and PRs first so I wouldn't duplicate anyone's work. The project had already hit the same class of bug with orgId/orgID, which told me the fix shape was the right one.

The fix

I kept it boring on purpose. I added one helper:

func getPageSizeParam(urlValues url.Values) string {
    if v := urlValues.Get("pageSize"); v != "" {
        return v
    }
    return urlValues.Get("pagesize")
}

getPaginationParams now uses it, and so do all 20 call sites. One spot in connections_handlers.go parsed the value inline, so I simplified it to use the helper too. Anyone still sending pagesize sees no change.

Testing it

I wrote a table test, TestGetPageSizeParam, with four cases: canonical only, legacy only, both present, and neither present. The "both present" case matters most. During a rolling migration, old and new clients can hit the same handler, so pageSize has to win clearly.

Then I ran go build, go test and gofmt -l on every file I touched, plus one last grep to confirm no unguarded .Get("pagesize") reads were left. "I fixed the ones I found" and "I fixed all of them" are different claims, and only the grep backs up the second one.

The review

CodeRabbit asked me to document the contract in the API docs. The OpenAPI spec already documents pageSize, and the helper has a doc comment explaining the fallback. I replied with that reasoning and the thread was resolved with no diff. Sometimes the thing a reviewer asks for is already true.

What I learned

If a repo has a written "this is how we do it" rule and lots of call sites, some of them have drifted from it. Grepping for the drift is cheap, and fixing it before the contract changes costs a code review. Fixing it after costs someone an afternoon of debugging.