Summary

vcpkg crashes intermittently with "unreachable code was reached" when track_elapsed_us() receives a negative elapsed time value. This happens in containerized/VM environments where std::chrono::high_resolution_clock can go backwards.

Error Message

Elapsed time to handle boost-utility:x64-linux: -1.29e+08 ns
vcpkg internal error: unreachable code was reached
...

Root Cause

In src/vcpkg/metrics.cpp, the track_elapsed_us() function has this assertion:

void MetricsSubmission::track_elapsed_us(double value)
{
    if (!isfinite(value) || value <= 0.0)
    {
        Checks::unreachable(VCPKG_LINE_INFO);  // Crashes here
    }
    elapsed_us = value;
}

The problem is that std::chrono::high_resolution_clock is not guaranteed to be monotonic. In VMs, containers, or when NTP adjusts time, the clock can go backwards, producing negative elapsed time.

Environment

• OS: CBL-Mariner 2.0 (Linux container) (Also happens in Azure Linux 3.0)
• Context: CI/CD pipeline (Azure DevOps)
• vcpkg version: latest (cloned from main)

Workaround

Disable metrics with -disableMetrics flag or VCPKG_DISABLE_METRICS=1 environment variable.

Disabling metrics still invokes track_elapsed_us.

Suggested Fix

Either:

1. Use std::chrono::steady_clock instead of high_resolution_clock (guaranteed monotonic)
2. Clamp negative elapsed times to 0 or a small positive value instead of crashing
3. Skip metrics submission when elapsed time is invalid

Example fix for option 3:

void MetricsSubmission::track_elapsed_us(double value)
{
    if (!isfinite(value) || value <= 0.0)
    {
        return;  // Skip invalid measurements instead of crashing
    }
    elapsed_us = value;
}