Understand the root cause of the following 8 issues and fix them.


<file name="src/streaming/protocol/command-begin-set-end-v2.c">

<violation number="1" location="src/streaming/protocol/command-begin-set-end-v2.c:61">
Encoding integer baselines now loses precision for values &gt;2^53 because the value is routed through a double before being cast back to int64, corrupting streamed state for large counters.</violation>
</file>

<file name="src/streaming/protocol/command-begin-set-end-v1.c">

<violation number="1" location="src/streaming/protocol/command-begin-set-end-v1.c:32">
Casting the collected double to int64 before printing truncates all fractional dimension values, so streaming SET lines still transmit integers instead of floats.</violation>
</file>

<file name="src/streaming/stream-replication-sender.c">

<violation number="1" location="src/streaming/stream-replication-sender.c:244">
Casting the integer baseline through rrddim_last_collected_as_double() truncates large counters (&gt;2^53) before sending them. Use the raw int accessor so replication baselines remain exact.</violation>
</file>

<file name="src/database/rrdset-collection.c">

<violation number="1" location="src/database/rrdset-collection.c:678">
Reset handling for PCENT_OVER_DIFF_TOTAL now calls the float setter unconditionally, so integer dimensions never update their last value after an overflow and continue to report negative deltas.</violation>

<violation number="2" location="src/database/rrdset-collection.c:681">
Summing rrddim_*_as_double() into total_number (long long) truncates every float dimension’s contribution, breaking percentage calculations for non-integer dimensions.</violation>
</file>

<file name="src/plugins.d/pluginsd_parser.c">

<violation number="1" location="src/plugins.d/pluginsd_parser.c:1111">
Float dimensions are reset using the int-only accessor, so their collected_value fields remain stale after END2; use the float setter when rrddim_is_float(rd).</violation>
</file>

<file name="src/database/rrddim.c">

<violation number="1" location="src/database/rrddim.c:618">
`rrddim_set_by_pointer_double()` truncates floats to 64‑bit integers before storing, so the new float DIMENSION support never records fractional data.</violation>

<violation number="2" location="src/database/rrddim.c:645">
The float branch of `rrddim_timed_set_by_pointer()` returns the last value cast to `collected_number`, so even the new double API always reports an integer result.</violation>
</file>