fix(debug-files): a dry run must not fail on a map with no debug-id

[krassx]

Item §4 from the CLI-in-JS-flows audit: the whole flow cannot be dry-run. Independent of #45/#46 (different file), so it can merge in any order.

The problem

--dry-run is the documented safe diagnostic, and it could not be used on a freshly built output directory at all:

• sourcemaps inject --dry-run writes nothing, by design.
• so every map still has no debug-id,
• so debug-files upload --type sourcemaps --dry-run exited 11 on the first one.

@bugsee/bundler-plugin-core works around it by skipping the upload step entirely on a dry run — meaning the one safe way to preview the flow never exercises the flow.

The fix

A dry run sends nothing, so an un-keyed map cannot register the unfindable symbol the real run refuses over. It is now reported and counted instead:

$ bugsee-cli sourcemaps inject . --dry-run
INFO injected debug-id path=./app.js debug_id=c33a6517-…
INFO inject complete js_injected=1 … maps_updated=1 dry_run=true

$ bugsee-cli debug-files upload . --type sourcemaps --dry-run …
WARN dry run: no debug_id — `sourcemaps inject` keys it before a real upload path=./app.js.map
WARN dry run: no debug_id — `sourcemaps inject` keys it before a real upload path=./orphan.js.map
INFO dry-run complete: no map carries a debug_id yet — `sourcemaps inject` keys them unkeyed=2 skipped=0
exit 0

$ bugsee-cli debug-files upload . --type sourcemaps …        # real run
exit 11        # unchanged

When NOTHING is keyed, the run completes as a success saying so, rather than falling into the "all 0 source maps are stylesheet or type-declaration maps" branch — plainly the wrong message there, and the one it used to produce.

Unchanged: a real run still exits 11 on the first un-keyed map (uploading it would register a symbol nothing can find), --uuid still keys a map on a dry run, and a dry run still packs everything it can key.

Tests

Three unit tests (real run still refuses + dry run succeeds and packs the keyed map; an entirely un-injected build; --uuid on a dry run) and two e2e flows against the mock server, asserting the warning text and that nothing is posted. Four mutants caught: tolerance removed, tolerance extended to the real run, the all-unkeyed branch bypassed, the counter not incremented.

Gates: cargo fmt --check, clippy --all-targets -D warnings (0), 496 unit tests + all suites, e2e_flows.py ALL PASS.

Follow-on in the JS repo once this is released: @bugsee/bundler-plugin-core can drop its dry-run workaround and preview the real flow.

🤖 Generated with Claude Code

[claude]

Code review

debug-files upload --type sourcemaps --dry-run no longer fails on a map with no debug-id: an un-keyed map is now counted (unkeyed) and reported via tracing::warn! instead of erroring, and a dry run where nothing is keyed returns Ok with an accurate log line instead of falling into the (previously misleading) "all N maps are stylesheet maps" branch. The real-run path is untouched (still errors via the same typed input_invalid, still exit 11), the --uuid override path is untouched, and the new logging goes through tracing to stderr, so stdout purity and the exit-code contract both hold.

I traced the full run_sourcemap_upload flow against CLAUDE.md's stdout/exit-code/wire-shape rules and checked the new/changed branches for edge cases (mixed skipped+unkeyed candidates, dry_run with a partially-keyed batch, the --uuid-override bypass) — all behave as documented, and the new unit tests plus the two added e2e flows (sourcemaps_uninjected_dry_run_succeeds, sourcemaps_dry_run_uploads_nothing) exercise exactly the cases described in the PR body. No --help/clap doc drift (no flag or command surface changed), no Cargo.toml/dist/daemon changes in this diff.

Findings: None.

Looks sound — merge as is.

[claude]

Code review

Makes debug-files upload --type sourcemaps --dry-run report un-keyed maps (unkeyed counter + a tracing::warn!) instead of failing on the first one, so the documented SAFE diagnostic actually works on a freshly built, uninjected directory; a real (non-dry) run still exits 11 on the first un-keyed map, unchanged. Traced through the logic carefully (the planned.is_empty() branch, the interaction with the pre-existing stylesheet-only-maps error path, and the final dry-run complete log) and it holds together: unkeyed is only ever nonzero on a dry run, the stylesheet-only-maps case is untouched (still fails loudly in both modes, per the existing test), and --uuid still overrides as before. New unit tests cover the three relevant cases, and the e2e script + README/CHANGELOG were updated consistently with the code. No stdout writes were added (all new messages go through tracing to stderr), no exit-code meanings changed, and no wire-format/JSON output is touched.

Findings: None.

Looks sound — merge as is.