test_statistics_window.c File Reference

C unit tests for window lifecycle, scheduling, and weighting. More...

#include "test_support.h"
#include "statistics_window.h"
#include "field_catalog.h"

Include dependency graph for test_statistics_window.c:

Go to the source code of this file.

Functions
static PicurvWindowDefinition	StepWindow (const char *name, PetscReal start, PetscReal end, PetscBool bounded, PicurvWeighting weighting, PetscInt cadence)
	Builds a step-cadence definition.
static PetscErrorCode	RunUniform (PicurvWindow *w, PetscInt steps, PetscReal dt, PetscReal *total_weight, PetscInt *count)
	Drives a uniform-dt sequence and returns the accumulated weight and count.
static PetscErrorCode	TestWeightingsAgreeAtConstantTimestep (void)
	The two weightings must agree on a constant-timestep run.
static PetscErrorCode	TestOriginStateAnchorsWithoutSampling (void)
	A state at the window origin anchors without becoming a sample.
static PetscErrorCode	TestLateFirstObservationMovesOrigin (void)
	A window first seen after its requested start moves its origin forward.
static PetscErrorCode	TestVariableTimestepWeighting (void)
	Variable timestep weighting follows the actual elapsed intervals.
static PetscErrorCode	TestStrideAndOffScheduleNoOp (void)
	Stride skips states, and the accepted weight still spans the whole gap.
static PetscErrorCode	TestEndClippingAndCompletion (void)
	A bounded window clips its final interval and then accepts nothing.
static PetscErrorCode	TestDuplicateEventRejected (void)
	The same completed step offered twice is counted once.
static PetscErrorCode	TestTimeCadenceOvershootAcceptedOnce (void)
	A step overshooting several time targets is accepted once, losing no time.
static PetscErrorCode	TestInvalidDefinitionsRejected (void)
	Invalid definitions are rejected at initialization.
static PetscErrorCode	HashAndCompare (const PicurvWindowDefinition *baseline, const PicurvWindowDefinition *variant, PetscBool *same, PetscInt *first_difference)
	Hashes a definition and reports which property group changed against a baseline.
static PicurvWindowDefinition	HashWindow (void)
	Builds the hash fixture: a Ucat/P window with a second moment and a covariance.
static PetscErrorCode	TestHashExclusionsAndStability (void)
	The hash is stable, and excludes exactly the properties the spec excludes.
static PetscErrorCode	TestHashDetectsEachProperty (void)
	Every hashed property changes the digest and is named by its group digest.
static PetscErrorCode	SerializeGroupDigests (const PicurvWindowDefinition *definition, char *out, size_t out_size)
	Serializes a definition's group digests the way a checkpoint records them.
static PetscErrorCode	TestFirstHashDifferenceLocalization (void)
	Saved group digests must localize a change to the property that caused it.
int	main (int argc, char **argv)
	Entry point for the window lifecycle suite.

Detailed Description

C unit tests for window lifecycle, scheduling, and weighting.

Covers the Stage 3 acceptance items in Field Statistics Phase 2 Implementation Specification section 14: cadence stride, start and end clipping, variable timestep weighting, duplicate event rejection, off-schedule no-ops, and the property that sample and physical-time weighting agree on a constant-timestep run.

Definition in file test_statistics_window.c.

Function Documentation

StepWindow()

static PicurvWindowDefinition StepWindow ( const char *  name, PetscReal  start, PetscReal  end, PetscBool  bounded, PicurvWeighting  weighting, PetscInt  cadence  )

static

Builds a step-cadence definition.

Definition at line 18 of file test_statistics_window.c.

{
    PicurvWindowDefinition d;
    memset(&d, 0, sizeof(d));
    strncpy(d.name, name, PICURV_WINDOW_NAME_LENGTH - 1);
    d.start_time = start; d.end_time = end; d.bounded = bounded;
    d.weighting = weighting; d.cadence_kind = PICURV_CADENCE_STEP; d.step_cadence = cadence;
    return d;
}

Here is the caller graph for this function:

RunUniform()

static PetscErrorCode RunUniform ( PicurvWindow *  w, PetscInt  steps, PetscReal  dt, PetscReal *  total_weight, PetscInt *  count  )

static

Drives a uniform-dt sequence and returns the accumulated weight and count.

Definition at line 31 of file test_statistics_window.c.

{
    PetscFunctionBeginUser;
    for (PetscInt s = 0; s <= steps; ++s) {
        PetscBool accepted = PETSC_FALSE;
        PetscReal weight = 0.0;
        PetscCall(PicurvWindowOfferState(w, s, (PetscReal)s * dt, &accepted, &weight));
    }
    *total_weight = w->total_weight;
    *count = w->sample_count;
    PetscFunctionReturn(0);
}

Here is the call graph for this function:

Here is the caller graph for this function:

TestWeightingsAgreeAtConstantTimestep()

static PetscErrorCode TestWeightingsAgreeAtConstantTimestep ( void  )

static

The two weightings must agree on a constant-timestep run.

This is the property that fixed the initial-state rule: counting the state at the window origin under sample weighting alone would make the two disagree by O(1/N) for no physical reason.

Definition at line 52 of file test_statistics_window.c.

{
    PicurvWindow sample_w, time_w;
    PetscReal sample_weight = 0.0, time_weight = 0.0;
    PetscInt sample_count = 0, time_count = 0;
    const PetscReal dt = 0.25;
    const PetscInt steps = 20;
 
    PetscFunctionBeginUser;
    {
        PicurvWindowDefinition d = StepWindow("s", 0.0, 0.0, PETSC_FALSE, PICURV_WEIGHTING_SAMPLE, 1);
        PetscCall(PicurvWindowInit(&sample_w, &d));
    }
    {
        PicurvWindowDefinition d = StepWindow("t", 0.0, 0.0, PETSC_FALSE, PICURV_WEIGHTING_PHYSICAL_TIME, 1);
        PetscCall(PicurvWindowInit(&time_w, &d));
    }
    PetscCall(RunUniform(&sample_w, steps, dt, &sample_weight, &sample_count));
    PetscCall(RunUniform(&time_w, steps, dt, &time_weight, &time_count));
 
    /* The state at t=0 anchors the origin and is not a sample under either rule. */
    PetscCall(PicurvAssertIntEqual(steps, sample_count, "sample weighting counts one per interval"));
    PetscCall(PicurvAssertIntEqual(steps, time_count, "time weighting counts one per interval"));
    PetscCall(PicurvAssertIntEqual(sample_count, time_count,
                                   "both weightings must accept exactly the same states"));
    PetscCall(PicurvAssertRealNear((PetscReal)steps, sample_weight, 1.0e-12, "sample total weight"));
    PetscCall(PicurvAssertRealNear((PetscReal)steps * dt, time_weight, 1.0e-12, "time total weight"));
    /* A mean is weight-normalized, so equal counts with proportional weights means
     * the two weightings produce the identical mean for a uniform-dt run. */
    PetscCall(PicurvAssertRealNear(dt, time_weight / sample_weight, 1.0e-12,
                                   "weights differ only by the constant timestep"));
    PetscCall(PicurvAssertRealNear((PetscReal)steps * dt, time_w.represented_time, 1.0e-12,
                                   "represented time spans the whole run"));
    PetscFunctionReturn(0);
}

Here is the call graph for this function:

Here is the caller graph for this function:

TestOriginStateAnchorsWithoutSampling()

static PetscErrorCode TestOriginStateAnchorsWithoutSampling ( void  )

static

A state at the window origin anchors without becoming a sample.

Definition at line 89 of file test_statistics_window.c.

{
    PicurvWindow w;
    PicurvWindowDefinition d = StepWindow("anchor", 5.0, 0.0, PETSC_FALSE,
                                          PICURV_WEIGHTING_PHYSICAL_TIME, 1);
    PetscBool accepted = PETSC_FALSE;
    PetscReal weight = 0.0;
 
    PetscFunctionBeginUser;
    PetscCall(PicurvWindowInit(&w, &d));
    PetscCall(PicurvAssertIntEqual(PICURV_WINDOW_PENDING, w.state, "window starts pending"));
 
    /* Before the start: no effect at all. */
    PetscCall(PicurvWindowOfferState(&w, 10, 4.5, &accepted, &weight));
    PetscCall(PicurvAssertBool((PetscBool)!accepted, "states before the start are ignored"));
    PetscCall(PicurvAssertIntEqual(PICURV_WINDOW_PENDING, w.state, "an early state does not activate"));
 
    /* Exactly at the start: activates and anchors, but is not a sample. */
    PetscCall(PicurvWindowOfferState(&w, 11, 5.0, &accepted, &weight));
    PetscCall(PicurvAssertBool((PetscBool)!accepted, "the origin state is not a sample"));
    PetscCall(PicurvAssertIntEqual(PICURV_WINDOW_ACTIVE, w.state, "the origin state activates the window"));
    PetscCall(PicurvAssertIntEqual(0, w.sample_count, "anchoring records no sample"));
 
    /* Next state carries the interval back to the origin. */
    PetscCall(PicurvWindowOfferState(&w, 12, 5.5, &accepted, &weight));
    PetscCall(PicurvAssertBool(accepted, "the state after the origin is a sample"));
    PetscCall(PicurvAssertRealNear(0.5, weight, 1.0e-12, "first sample carries the interval from the origin"));
    PetscFunctionReturn(0);
}

Here is the call graph for this function:

Here is the caller graph for this function:

TestLateFirstObservationMovesOrigin()

static PetscErrorCode TestLateFirstObservationMovesOrigin ( void  )

static

A window first seen after its requested start moves its origin forward.

Definition at line 120 of file test_statistics_window.c.

{
    PicurvWindow w;
    PicurvWindowDefinition d = StepWindow("late", 5.0, 0.0, PETSC_FALSE,
                                          PICURV_WEIGHTING_PHYSICAL_TIME, 1);
    PetscBool accepted = PETSC_FALSE;
    PetscReal weight = 0.0;
 
    PetscFunctionBeginUser;
    PetscCall(PicurvWindowInit(&w, &d));
    /* Resumed at t=8 with no earlier observation: the window must not claim [5,8]. */
    PetscCall(PicurvWindowOfferState(&w, 40, 8.0, &accepted, &weight));
    PetscCall(PicurvAssertBool((PetscBool)!accepted, "the first observed state anchors the moved origin"));
    PetscCall(PicurvAssertRealNear(8.0, w.effective_start, 1.0e-12,
                                   "effective start moves to the first observed time"));
    PetscCall(PicurvWindowOfferState(&w, 41, 8.5, &accepted, &weight));
    PetscCall(PicurvAssertBool(accepted, "the next state is a sample"));
    PetscCall(PicurvAssertRealNear(0.5, weight, 1.0e-12, "weight is measured from the moved origin"));
    PetscCall(PicurvAssertRealNear(0.5, w.represented_time, 1.0e-12,
                                   "a window never claims time it did not observe"));
    PetscFunctionReturn(0);
}

Here is the call graph for this function:

Here is the caller graph for this function:

TestVariableTimestepWeighting()

static PetscErrorCode TestVariableTimestepWeighting ( void  )

static

Variable timestep weighting follows the actual elapsed intervals.

Definition at line 144 of file test_statistics_window.c.

{
    PicurvWindow w;
    PicurvWindowDefinition d = StepWindow("vardt", 0.0, 0.0, PETSC_FALSE,
                                          PICURV_WEIGHTING_PHYSICAL_TIME, 1);
    const PetscReal times[4] = {0.0, 0.5, 2.0, 2.25};
    const PetscReal expected[4] = {0.0, 0.5, 1.5, 0.25};
    PetscBool accepted = PETSC_FALSE;
    PetscReal weight = 0.0;
 
    PetscFunctionBeginUser;
    PetscCall(PicurvWindowInit(&w, &d));
    for (PetscInt i = 0; i < 4; ++i) {
        PetscCall(PicurvWindowOfferState(&w, i, times[i], &accepted, &weight));
        PetscCall(PicurvAssertRealNear(expected[i], weight, 1.0e-12,
                                       "variable-dt weight equals the elapsed interval"));
    }
    PetscCall(PicurvAssertIntEqual(3, w.sample_count, "three intervals become samples"));
    PetscCall(PicurvAssertRealNear(2.25, w.total_weight, 1.0e-12, "weights sum to the span"));
    PetscFunctionReturn(0);
}

Here is the call graph for this function:

Here is the caller graph for this function:

TestStrideAndOffScheduleNoOp()

static PetscErrorCode TestStrideAndOffScheduleNoOp ( void  )

static

Stride skips states, and the accepted weight still spans the whole gap.

Definition at line 167 of file test_statistics_window.c.

{
    PicurvWindow w;
    PicurvWindowDefinition d = StepWindow("stride", 0.0, 0.0, PETSC_FALSE,
                                          PICURV_WEIGHTING_PHYSICAL_TIME, 3);
    PetscBool accepted = PETSC_FALSE;
    PetscReal weight = 0.0;
    const PetscReal dt = 0.1;
 
    PetscFunctionBeginUser;
    PetscCall(PicurvWindowInit(&w, &d));
    for (PetscInt s = 0; s <= 6; ++s) {
        PetscReal before_weight = w.total_weight;
        PetscInt before_count = w.sample_count;
        PetscCall(PicurvWindowOfferState(&w, s, (PetscReal)s * dt, &accepted, &weight));
        if (s % 3 != 0) {
            PetscCall(PicurvAssertBool((PetscBool)!accepted, "off-schedule states are not sampled"));
            PetscCall(PicurvAssertRealNear(before_weight, w.total_weight, 1.0e-15,
                                           "an off-schedule state changes no scientific state"));
            PetscCall(PicurvAssertIntEqual(before_count, w.sample_count,
                                           "an off-schedule state records no sample"));
        }
    }
    /* Steps 3 and 6 are sampled; each represents three timesteps. */
    PetscCall(PicurvAssertIntEqual(2, w.sample_count, "stride 3 over 6 steps yields two samples"));
    PetscCall(PicurvAssertRealNear(0.6, w.total_weight, 1.0e-12,
                                   "strided weights still cover the full elapsed span"));
    PetscFunctionReturn(0);
}

Here is the call graph for this function:

Here is the caller graph for this function:

TestEndClippingAndCompletion()

static PetscErrorCode TestEndClippingAndCompletion ( void  )

static

A bounded window clips its final interval and then accepts nothing.

Definition at line 198 of file test_statistics_window.c.

{
    PicurvWindow w;
    PicurvWindowDefinition d = StepWindow("bounded", 0.0, 1.0, PETSC_TRUE,
                                          PICURV_WEIGHTING_PHYSICAL_TIME, 1);
    PetscBool accepted = PETSC_FALSE;
    PetscReal weight = 0.0;
 
    PetscFunctionBeginUser;
    PetscCall(PicurvWindowInit(&w, &d));
    PetscCall(PicurvWindowOfferState(&w, 0, 0.0, &accepted, &weight));   /* anchor */
    PetscCall(PicurvWindowOfferState(&w, 1, 0.4, &accepted, &weight));
    PetscCall(PicurvAssertRealNear(0.4, weight, 1.0e-12, "interior interval"));
    PetscCall(PicurvWindowOfferState(&w, 2, 0.8, &accepted, &weight));
    PetscCall(PicurvAssertRealNear(0.4, weight, 1.0e-12, "interior interval"));
 
    /* Overshoots the bound: the final weight clips to the requested end. */
    PetscCall(PicurvWindowOfferState(&w, 3, 1.3, &accepted, &weight));
    PetscCall(PicurvAssertBool(accepted, "the overshooting state still contributes its clipped interval"));
    PetscCall(PicurvAssertRealNear(0.2, weight, 1.0e-12, "final interval clips to the requested end"));
    PetscCall(PicurvAssertIntEqual(PICURV_WINDOW_COMPLETE, w.state, "reaching the bound completes the window"));
    PetscCall(PicurvAssertRealNear(1.0, w.represented_time, 1.0e-12,
                                   "represented time equals the requested span exactly"));
    PetscCall(PicurvAssertRealNear(1.0, PicurvWindowProgress(&w), 1.0e-12, "a complete window reports full progress"));
 
    /* Nothing further is accepted. */
    PetscCall(PicurvWindowOfferState(&w, 4, 1.5, &accepted, &weight));
    PetscCall(PicurvAssertBool((PetscBool)!accepted, "a complete window accepts nothing further"));
    PetscCall(PicurvAssertIntEqual(3, w.sample_count, "completion does not add samples"));
    PetscFunctionReturn(0);
}

Here is the call graph for this function:

Here is the caller graph for this function:

TestDuplicateEventRejected()

static PetscErrorCode TestDuplicateEventRejected ( void  )

static

The same completed step offered twice is counted once.

Definition at line 231 of file test_statistics_window.c.

{
    PicurvWindow w;
    PicurvWindowDefinition d = StepWindow("dup", 0.0, 0.0, PETSC_FALSE,
                                          PICURV_WEIGHTING_PHYSICAL_TIME, 1);
    PetscBool accepted = PETSC_FALSE;
    PetscReal weight = 0.0;
 
    PetscFunctionBeginUser;
    PetscCall(PicurvWindowInit(&w, &d));
    PetscCall(PicurvWindowOfferState(&w, 0, 0.0, &accepted, &weight));
    PetscCall(PicurvWindowOfferState(&w, 1, 0.5, &accepted, &weight));
    PetscCall(PicurvAssertBool(accepted, "first offer of a step is accepted"));
 
    PetscCall(PicurvWindowOfferState(&w, 1, 0.5, &accepted, &weight));
    PetscCall(PicurvAssertBool((PetscBool)!accepted, "the same step must not be counted twice"));
    PetscCall(PicurvAssertIntEqual(1, w.sample_count, "a duplicate offer records no extra sample"));
    PetscCall(PicurvAssertRealNear(0.5, w.total_weight, 1.0e-12, "a duplicate offer adds no weight"));
    PetscFunctionReturn(0);
}

Here is the call graph for this function:

Here is the caller graph for this function:

TestTimeCadenceOvershootAcceptedOnce()

static PetscErrorCode TestTimeCadenceOvershootAcceptedOnce ( void  )

static

A step overshooting several time targets is accepted once, losing no time.

This is the self-correcting property of right-rectangle weighting: the accepted weight is the actual elapsed interval, not the nominal cadence, so skipped targets neither drop nor double-count represented time.

Definition at line 259 of file test_statistics_window.c.

{
    PicurvWindow w;
    PicurvWindowDefinition d;
    PetscBool accepted = PETSC_FALSE;
    PetscReal weight = 0.0;
 
    PetscFunctionBeginUser;
    memset(&d, 0, sizeof(d));
    strncpy(d.name, "tcad", PICURV_WINDOW_NAME_LENGTH - 1);
    d.start_time = 0.0; d.bounded = PETSC_FALSE;
    d.weighting = PICURV_WEIGHTING_PHYSICAL_TIME;
    d.cadence_kind = PICURV_CADENCE_TIME; d.time_cadence = 0.1;
    PetscCall(PicurvWindowInit(&w, &d));
 
    PetscCall(PicurvWindowOfferState(&w, 0, 0.0, &accepted, &weight));  /* anchor */
    PetscCall(PicurvAssertBool((PetscBool)!accepted, "origin anchors under time cadence too"));
 
    /* One large step jumps past targets 0.1, 0.2, 0.3, 0.4 and lands on 0.45. */
    PetscCall(PicurvWindowOfferState(&w, 1, 0.45, &accepted, &weight));
    PetscCall(PicurvAssertBool(accepted, "the overshooting state is accepted"));
    PetscCall(PicurvAssertRealNear(0.45, weight, 1.0e-12,
                                   "weight is the actual elapsed interval, not the nominal cadence"));
    PetscCall(PicurvAssertIntEqual(1, w.sample_count, "several skipped targets yield exactly one sample"));
 
    /* The schedule resumes on the absolute grid rather than drifting from 0.45. */
    PetscCall(PicurvWindowOfferState(&w, 2, 0.46, &accepted, &weight));
    PetscCall(PicurvAssertBool((PetscBool)!accepted, "the next target has not been reached yet"));
    PetscCall(PicurvWindowOfferState(&w, 3, 0.51, &accepted, &weight));
    PetscCall(PicurvAssertBool(accepted, "the state reaching the 0.5 target is accepted"));
    PetscCall(PicurvAssertRealNear(0.06, weight, 1.0e-12, "weight resumes from the last accepted state"));
    PetscCall(PicurvAssertRealNear(0.51, w.represented_time, 1.0e-12,
                                   "no represented time is lost or double counted across the overshoot"));
    PetscFunctionReturn(0);
}

Here is the call graph for this function:

Here is the caller graph for this function:

TestInvalidDefinitionsRejected()

static PetscErrorCode TestInvalidDefinitionsRejected ( void  )

static

Invalid definitions are rejected at initialization.

Definition at line 296 of file test_statistics_window.c.

{
    PicurvWindow w;
    PetscErrorCode ierr_cadence = 0, ierr_name = 0, ierr_span = 0;
 
    PetscFunctionBeginUser;
    {
        PicurvWindowDefinition d = StepWindow("zero", 0.0, 0.0, PETSC_FALSE, PICURV_WEIGHTING_SAMPLE, 0);
        PetscCall(PetscPushErrorHandler(PetscIgnoreErrorHandler, NULL));
        ierr_cadence = PicurvWindowInit(&w, &d);
        PetscCall(PetscPopErrorHandler());
    }
    PetscCall(PicurvAssertIntEqual(PETSC_ERR_ARG_OUTOFRANGE, ierr_cadence, "non-positive stride is rejected"));
    {
        PicurvWindowDefinition d = StepWindow("", 0.0, 0.0, PETSC_FALSE, PICURV_WEIGHTING_SAMPLE, 1);
        PetscCall(PetscPushErrorHandler(PetscIgnoreErrorHandler, NULL));
        ierr_name = PicurvWindowInit(&w, &d);
        PetscCall(PetscPopErrorHandler());
    }
    PetscCall(PicurvAssertIntEqual(PETSC_ERR_ARG_OUTOFRANGE, ierr_name, "an empty window name is rejected"));
    {
        PicurvWindowDefinition d = StepWindow("bad", 5.0, 5.0, PETSC_TRUE, PICURV_WEIGHTING_SAMPLE, 1);
        PetscCall(PetscPushErrorHandler(PetscIgnoreErrorHandler, NULL));
        ierr_span = PicurvWindowInit(&w, &d);
        PetscCall(PetscPopErrorHandler());
    }
    PetscCall(PicurvAssertIntEqual(PETSC_ERR_ARG_OUTOFRANGE, ierr_span, "end must exceed start"));
    PetscFunctionReturn(0);
}

Here is the call graph for this function:

Here is the caller graph for this function:

HashAndCompare()

static PetscErrorCode HashAndCompare ( const PicurvWindowDefinition *  baseline, const PicurvWindowDefinition *  variant, PetscBool *  same, PetscInt *  first_difference  )

static

Hashes a definition and reports which property group changed against a baseline.

Definition at line 327 of file test_statistics_window.c.

{
    char base_digest[65], variant_digest[65];
    char base_groups[PICURV_WINDOW_HASH_GROUP_COUNT][PICURV_WINDOW_HASH_GROUP_LENGTH];
    char variant_groups[PICURV_WINDOW_HASH_GROUP_COUNT][PICURV_WINDOW_HASH_GROUP_LENGTH];
 
    PetscFunctionBeginUser;
    PetscCall(PicurvWindowComputeHash(baseline, base_digest, base_groups));
    PetscCall(PicurvWindowComputeHash(variant, variant_digest, variant_groups));
    PetscCall(PetscStrcmp(base_digest, variant_digest, same));
    *first_difference = -1;
    for (PetscInt group = 0; group < PICURV_WINDOW_HASH_GROUP_COUNT; ++group) {
        PetscBool group_same = PETSC_FALSE;
 
        PetscCall(PetscStrcmp(base_groups[group], variant_groups[group], &group_same));
        if (!group_same) { *first_difference = group; break; }
    }
    PetscFunctionReturn(0);
}

Here is the call graph for this function:

Here is the caller graph for this function:

HashWindow()

static PicurvWindowDefinition HashWindow ( void  )

static

Builds the hash fixture: a Ucat/P window with a second moment and a covariance.

Definition at line 350 of file test_statistics_window.c.

{
    PicurvWindowDefinition d = StepWindow("production", 10.0, 20.0, PETSC_TRUE,
                                          PICURV_WEIGHTING_PHYSICAL_TIME, 5);
    d.field_count = 2;
    d.fields[0].field_id = FIELD_ID_UCAT; d.fields[0].want_second = PETSC_TRUE;
    d.fields[1].field_id = FIELD_ID_P;    d.fields[1].want_second = PETSC_FALSE;
    d.covariance_count = 1;
    d.covariances[0].first = FIELD_ID_UCAT;
    d.covariances[0].second = FIELD_ID_P;
    return d;
}

Here is the call graph for this function:

Here is the caller graph for this function:

TestHashExclusionsAndStability()

static PetscErrorCode TestHashExclusionsAndStability ( void  )

static

The hash is stable, and excludes exactly the properties the spec excludes.

Definition at line 364 of file test_statistics_window.c.

{
    const PicurvWindowDefinition baseline = HashWindow();
    PicurvWindowDefinition variant;
    PetscBool same = PETSC_FALSE;
    PetscInt group = -1;
 
    PetscFunctionBeginUser;
    /* Stability: hashing the same definition twice gives the same digest. */
    variant = baseline;
    PetscCall(HashAndCompare(&baseline, &variant, &same, &group));
    PetscCall(PicurvAssertBool(same, "hashing an unchanged definition is stable"));
 
    /* end_time is excluded so a bounded window can be extended forward. */
    variant = baseline;
    variant.end_time = 40.0;
    PetscCall(HashAndCompare(&baseline, &variant, &same, &group));
    PetscCall(PicurvAssertBool(same, "extending end_time does not change the identity"));
 
    /* Listing order is excluded so a reordered configuration still continues. */
    variant = baseline;
    variant.fields[0] = baseline.fields[1];
    variant.fields[1] = baseline.fields[0];
    PetscCall(HashAndCompare(&baseline, &variant, &same, &group));
    PetscCall(PicurvAssertBool(same, "reordering the field list does not change the identity"));
 
    /* A covariance pair is symmetric, so its member order is excluded too. */
    variant = baseline;
    variant.covariances[0].first = FIELD_ID_P;
    variant.covariances[0].second = FIELD_ID_UCAT;
    PetscCall(HashAndCompare(&baseline, &variant, &same, &group));
    PetscCall(PicurvAssertBool(same, "swapping covariance members does not change the identity"));
    PetscFunctionReturn(0);
}

Here is the call graph for this function:

Here is the caller graph for this function:

TestHashDetectsEachProperty()

static PetscErrorCode TestHashDetectsEachProperty ( void  )

static

Every hashed property changes the digest and is named by its group digest.

Definition at line 400 of file test_statistics_window.c.

{
    const PicurvWindowDefinition baseline = HashWindow();
    PicurvWindowDefinition variant;
    PetscBool same = PETSC_FALSE;
    PetscInt group = -1;
    PetscBool named = PETSC_FALSE;
 
    PetscFunctionBeginUser;
    variant = baseline;
    strncpy(variant.name, "other", PICURV_WINDOW_NAME_LENGTH - 1);
    PetscCall(HashAndCompare(&baseline, &variant, &same, &group));
    PetscCall(PicurvAssertBool((PetscBool)!same, "a renamed window is a different window"));
    PetscCall(PicurvAssertIntEqual(0, group, "the name group reports the difference"));
 
    variant = baseline;
    variant.start_time = 11.0;
    PetscCall(HashAndCompare(&baseline, &variant, &same, &group));
    PetscCall(PicurvAssertBool((PetscBool)!same, "a moved start is a different window"));
    PetscCall(PicurvAssertIntEqual(1, group, "the start_time group reports the difference"));
 
    variant = baseline;
    variant.weighting = PICURV_WEIGHTING_SAMPLE;
    PetscCall(HashAndCompare(&baseline, &variant, &same, &group));
    PetscCall(PicurvAssertBool((PetscBool)!same, "a changed weighting is a different window"));
    PetscCall(PicurvAssertIntEqual(2, group, "the weighting group reports the difference"));
 
    variant = baseline;
    variant.step_cadence = 7;
    PetscCall(HashAndCompare(&baseline, &variant, &same, &group));
    PetscCall(PicurvAssertBool((PetscBool)!same, "a changed cadence is a different window"));
    PetscCall(PicurvAssertIntEqual(3, group, "the cadence group reports the difference"));
 
    /* Dropping a requested moment changes what the saved state means. */
    variant = baseline;
    variant.fields[0].want_second = PETSC_FALSE;
    PetscCall(HashAndCompare(&baseline, &variant, &same, &group));
    PetscCall(PicurvAssertBool((PetscBool)!same, "a changed moment set is a different window"));
    PetscCall(PicurvAssertIntEqual(4, group, "the fields group reports the difference"));
 
    variant = baseline;
    variant.covariance_count = 0;
    PetscCall(HashAndCompare(&baseline, &variant, &same, &group));
    PetscCall(PicurvAssertBool((PetscBool)!same, "a dropped covariance is a different window"));
    PetscCall(PicurvAssertIntEqual(5, group, "the covariances group reports the difference"));
 
    /* Group names exist for every group and are bounded. */
    for (PetscInt g = 0; g < PICURV_WINDOW_HASH_GROUP_COUNT; ++g) {
        PetscCall(PetscStrcmp(PicurvWindowHashGroupName(g), "unknown", &named));
        PetscCall(PicurvAssertBool((PetscBool)!named, "every hashed group has a stable name"));
    }
    PetscCall(PetscStrcmp(PicurvWindowHashGroupName(PICURV_WINDOW_HASH_GROUP_COUNT), "unknown", &named));
    PetscCall(PicurvAssertBool(named, "an out-of-range group index is named unknown"));
    PetscFunctionReturn(0);
}

Here is the call graph for this function:

Here is the caller graph for this function:

SerializeGroupDigests()

static PetscErrorCode SerializeGroupDigests ( const PicurvWindowDefinition *  definition, char *  out, size_t  out_size  )

static

Serializes a definition's group digests the way a checkpoint records them.

Definition at line 457 of file test_statistics_window.c.

{
    char digest[65];
    char groups[PICURV_WINDOW_HASH_GROUP_COUNT][PICURV_WINDOW_HASH_GROUP_LENGTH];
    size_t used = 0;
 
    PetscFunctionBeginUser;
    out[0] = '\0';
    PetscCall(PicurvWindowComputeHash(definition, digest, groups));
    for (PetscInt group = 0; group < PICURV_WINDOW_HASH_GROUP_COUNT; ++group) {
        PetscCall(PetscStrlen(out, &used));
        PetscCall(PetscSNPrintf(out + used, out_size - used, "%s%s", group ? "," : "", groups[group]));
    }
    PetscFunctionReturn(0);
}

Here is the call graph for this function:

Here is the caller graph for this function:

TestFirstHashDifferenceLocalization()

static PetscErrorCode TestFirstHashDifferenceLocalization ( void  )

static

Saved group digests must localize a change to the property that caused it.

This is the only path that turns a refused continuation into an actionable message, and it is reachable at runtime only through a fatal error, so it is exercised directly here rather than through that path.

Definition at line 481 of file test_statistics_window.c.

{
    const PicurvWindowDefinition baseline = HashWindow();
    PicurvWindowDefinition variant;
    char saved[PICURV_WINDOW_HASH_GROUP_COUNT * (PICURV_WINDOW_HASH_GROUP_LENGTH + 4)];
    PetscInt group = 0;
 
    PetscFunctionBeginUser;
    PetscCall(SerializeGroupDigests(&baseline, saved, sizeof(saved)));
 
    /* An unchanged definition reports no differing group. */
    PetscCall(PicurvWindowFirstHashDifference(&baseline, saved, &group));
    PetscCall(PicurvAssertIntEqual(-1, group, "an unchanged definition localizes to no group"));
 
    /* Each changed property is localized to its own group. */
    variant = baseline;
    variant.weighting = PICURV_WEIGHTING_SAMPLE;
    PetscCall(PicurvWindowFirstHashDifference(&variant, saved, &group));
    PetscCall(PicurvAssertIntEqual(2, group, "a changed weighting localizes to the weighting group"));
 
    variant = baseline;
    variant.covariance_count = 0;
    PetscCall(PicurvWindowFirstHashDifference(&variant, saved, &group));
    PetscCall(PicurvAssertIntEqual(5, group, "a dropped covariance localizes to the covariances group"));
 
    /* The earliest differing group wins when several changed at once. */
    variant = baseline;
    variant.start_time = 99.0;
    variant.step_cadence = 3;
    PetscCall(PicurvWindowFirstHashDifference(&variant, saved, &group));
    PetscCall(PicurvAssertIntEqual(1, group, "the first differing group is the one reported"));
 
    /* Degenerate inputs report no localization rather than a wrong property. */
    PetscCall(PicurvWindowFirstHashDifference(&baseline, NULL, &group));
    PetscCall(PicurvAssertIntEqual(-1, group, "absent group digests localize to nothing"));
    PetscCall(PicurvWindowFirstHashDifference(&baseline, "", &group));
    PetscCall(PicurvAssertIntEqual(-1, group, "an empty digest list localizes to nothing"));
 
    variant = baseline;
    variant.covariance_count = 0;
    PetscCall(SerializeGroupDigests(&baseline, saved, sizeof(saved)));
    saved[40] = '\0';   /* truncate mid-digest, so the last segment is short */
    PetscCall(PicurvWindowFirstHashDifference(&variant, saved, &group));
    PetscCall(PicurvAssertIntEqual(-1, group,
                                   "a truncated digest list reports no property rather than a wrong one"));
    PetscFunctionReturn(0);
}

Here is the call graph for this function:

Here is the caller graph for this function:

main()

int main	(	int	argc,
		char **	argv
	)

Entry point for the window lifecycle suite.

Definition at line 532 of file test_statistics_window.c.

{
    PetscErrorCode ierr;
    const PicurvTestCase cases[] = {
        {"weightings-agree-at-constant-timestep", TestWeightingsAgreeAtConstantTimestep},
        {"origin-state-anchors-without-sampling", TestOriginStateAnchorsWithoutSampling},
        {"late-first-observation-moves-origin", TestLateFirstObservationMovesOrigin},
        {"variable-timestep-weighting", TestVariableTimestepWeighting},
        {"stride-and-off-schedule-no-op", TestStrideAndOffScheduleNoOp},
        {"end-clipping-and-completion", TestEndClippingAndCompletion},
        {"duplicate-event-rejected", TestDuplicateEventRejected},
        {"time-cadence-overshoot-accepted-once", TestTimeCadenceOvershootAcceptedOnce},
        {"invalid-definitions-rejected", TestInvalidDefinitionsRejected},
        {"hash-exclusions-and-stability", TestHashExclusionsAndStability},
        {"hash-detects-each-property", TestHashDetectsEachProperty},
        {"first-hash-difference-localization", TestFirstHashDifferenceLocalization},
    };
 
    ierr = PetscInitialize(&argc, &argv, NULL, "PICurv statistics window tests");
    if (ierr) return (int)ierr;
    ierr = PicurvRunTests("unit-statistics-window", cases, sizeof(cases) / sizeof(cases[0]));
    if (ierr) { PetscFinalize(); return (int)ierr; }
    ierr = PetscFinalize();
    return (int)ierr;
}

Here is the call graph for this function: