Define and enforce a runtime minimum scale for Base2ExponentialHistogramAggregation

[ThomsonTan]

Describe your environment

• Repository revision: PR #4324
• Reproduced on Windows with MSVC 19.44 using a CMake/Ninja Release build
• The runtime scale behavior predates PR [SDK] Downscale base2 exponential histogram buckets in place #4324; that PR only exposed it during review

Steps to reproduce

Configure a base2 exponential histogram at the lowest currently accepted max_scale_ and max_size_, then record values at both ends of the finite double range:

using namespace opentelemetry::sdk::metrics;

Base2ExponentialHistogramAggregationConfig config;
config.max_scale_ = kMaxScaleMin;  // -10
config.max_size_  = kMaxSizeMin;   // 2

Base2ExponentialHistogramAggregation aggregation{&config};
aggregation.Aggregate((std::numeric_limits<double>::denorm_min)());
aggregation.Aggregate((std::numeric_limits<double>::max)());

const auto point =
    opentelemetry::nostd::get<Base2ExponentialHistogramPointData>(aggregation.ToPoint());
EXPECT_GE(point.scale_, kMaxScaleMin);

The current implementation accepts -10 as the configured maximum scale in aggregation_config.h, but automatic scale reduction never applies a lower bound.

What is the expected behavior?

The aggregation should define and enforce a reasonable runtime minimum scale for every automatic reduction path.

If -10 is selected as the C++ SDK's runtime minimum, point.scale_ should never become less than -10. When an input range still cannot fit within max_size_ at that floor, the implementation should apply an explicit, documented policy while preserving all recorded counts.

What is the actual behavior?

The reproduction produces:

point.scale_ == -11

At scale -10, denorm_min() and max() map to bucket indices -2 and 0. Their three-bucket span exceeds max_size_ == 2, so GetScaleReduction() requests one additional reduction. Base2ExponentialHistogramAggregation::Downscale() then subtracts that reduction without consulting any runtime floor:

const uint32_t scale_reduction =
    GetScaleReduction(start_index, end_index, point_data_.max_buckets_);
Downscale(scale_reduction);
// ...
point_data_.scale_ -= static_cast<int32_t>(by);

Additional context

Related:

• Reviewer discussion on PR #4324
• Configuration validation issue #4250

_{Tip: React with 👍 to help prioritize this issue. Please use comments to provide useful context, avoiding +1 or me too, to help us triage it. Learn more here.}