scout traces show: span durations render ~1000x too small (bars, colors, slow-warning all dead) #26

Problem

scout traces show renders every span's duration roughly 1000x too small, which silences the whole "where did the time go" visualization: progress bars, color thresholds, and the slow-span warning.

The API's span fields duration_ms / exclusive_duration_ms are actually reported in seconds, not milliseconds, despite the name. Confirmed by cross-checking a trace's duration_in_seconds against its root span's duration_ms — they're numerically identical, which they should not be if one were really 1000x the other.

$ scout traces show 12345 --app 6
Trace #12345 — Controller/users/index — 7.9s       # header: correct
 └── Middleware/Summary                        8ms  █░░░░░░░░░░░░░░░░░░░  # root span of the SAME trace

Root cause

internal/output/tree.go:

• renderSpan calls FormatMs(span.DurationMs) directly on the raw API value, formatting seconds as if they were milliseconds ("8ms" for an 8-second span).
• The bar-fill ratio span.DurationMs / totalMs is therefore always ≈0.001 regardless of the span's actual share of the trace, so filled is always forced to the 1-block minimum. This means:
  • Every span's bar looks identical.
  • The color thresholds (ratio > 0.5 yellow, > 0.8 red) can never trigger.
  • The ⚠ slow warning (also gated on ratio > 0.5) is unreachable.

internal/output/format_mb_test.go's TestRenderSpanTreeMemoryFooter currently constructs a TraceSpan with DurationMs: 500 alongside DurationInSeconds: 0.5 on the parent trace — i.e. it bakes in the same wrong assumption and will need updating alongside the fix.

Fix

Scale span.DurationMs / span.ExclusiveDurationMs by 1000 when consuming them (either at the API boundary in internal/api/types.go's TraceSpan, or at the point of use in renderSpan), and rename the struct fields/doc comments to make the actual unit (seconds) explicit rather than relying on a misleading _ms suffix from the API. Add a regression test asserting the header duration and the sum of top-level span durations are on the same order of magnitude.

Found during

A pre-1.0 audit of the whole CLI against a live account (see internal discussion — not filed verbatim here to avoid pasting live production data into a public issue).