sensordata Offline Checks

sensordata_check is a plain main() program with its own CHECK() macro, not a registered test framework, because this repository has no test harness to register with. It needs no tag and no capture: every case is a small SQLite log it builds itself, or a committed fixture.

cmake -DBUILD_SENSORDATA_CHECKS=ON <build-dir>
cmake --build <build-dir> --target sensordata_check
<build-dir>/Release/bin/sensordata_check

It prints each failing check with its line, then a count, and exits non-zero if any failed. Scratch logs go to $TMPDIR, or /tmp.

What it covers

Every answer is compared with a brute-force computation that shares no code with the library: linear scans for column edges, extremes and ranges.

Area	Cases
view()	Random data with spikes at bucket and power-of-two boundaries, a constant run with a repeated extreme, signed zeros, infinities, a NaN run longer than a bucket, a NULL; two segments 1,000 s apart; windows across, inside and outside the data and the gap, zero-width and empty; 1 to 100,000 columns, and 300 random windows
range(), StreamInfo	Every fixed window above; whole-stream min, max, first and last time
valueAt()	Both picks against a linear scan, values not representable as float, non-contiguous rowids, NaN and infinite times
Out-of-order tables	Shuffled times with many ties: stable sort, rowid of each sample, forEachRow() order
Edge cases	A single row, an empty table, and metadata that must be refused: no streams table or rows, a missing table or column, an unknown kind, an incomplete row, an ungrouped or mixed record set
Derived streams	forEachRow() inputs, addDerived() on a shared and on its own axis, rejected lengths and times, find() preferring the newest, remove(), and forEachRow() staying available while any stream uses the table
Gaps	Irregular steps with injected gaps (at the start, before the last sample, long, clustered several to a column, random): the gap count and typical interval against an independent implementation of the rule; views in both gap modes over windows that start, end or lie inside a gap; gap markers; inheritance by computed streams; valueAt() inside and beside gaps; an irregular but continuous table with no gaps
Fixtures	Every log in host/docs/fixtures/sensorviz/: the stream list equals the metadata rows with data, in order; times match ORDER BY time; views match brute force; forEachRow() returns the SQLite doubles

--gaps <path>...

Prints, for each log, every table's sample count, typical interval and gap count. The counts for the committed fixtures and other logs are in the decimating backend results.

--log <path>

Loads a real or synthetic log and prints the load time, the peak memory footprint (macOS) or peak RSS, the cost of 2,000-column views of the longest streams over the full log, one hour, one minute and ten seconds, and whether each view matched brute force. Results for the worst-case IMUTag log are in the decimating backend results.