// ----------------------------------------------------------------
static void make_accs(
slls_t* paccumulator_names, // Input
lhmsv_t* acc_field_to_acc_state) // Output
{
acc_t* ppercentile_acc = NULL;
for (sllse_t* pc = paccumulator_names->phead; pc != NULL; pc = pc->pnext) {
// for "sum", "count"
char* acc_name = pc->value;
if (is_percentile_acc_name(acc_name)) {
if (ppercentile_acc == NULL) {
ppercentile_acc = acc_percentile_alloc();
}
lhmsv_put(acc_field_to_acc_state, acc_name, ppercentile_acc);
} else {
acc_t* pacc = make_acc(acc_name);
if (pacc == NULL) {
fprintf(stderr, "%s stats1: accumulator \"%s\" not found.\n",
MLR_GLOBALS.argv0, acc_name);
exit(1);
}
lhmsv_put(acc_field_to_acc_state, acc_name, pacc);
}
}
}
// ----------------------------------------------------------------
void make_stats1_accs(
char* value_field_name, // input
slls_t* paccumulator_names, // input
int allow_int_float, // input
lhmsv_t* acc_field_to_acc_state) // output
{
stats1_acc_t* ppercentile_acc = NULL;
for (sllse_t* pc = paccumulator_names->phead; pc != NULL; pc = pc->pnext) {
// for "sum", "count"
char* stats1_acc_name = pc->value;
if (is_percentile_acc_name(stats1_acc_name)) {
if (ppercentile_acc == NULL) {
ppercentile_acc = stats1_percentile_alloc(value_field_name, stats1_acc_name, allow_int_float);
} else {
stats1_percentile_reuse(ppercentile_acc);
}
lhmsv_put(acc_field_to_acc_state, stats1_acc_name, ppercentile_acc, NO_FREE);
} else {
stats1_acc_t* pstats1_acc = make_stats1_acc(value_field_name, stats1_acc_name, allow_int_float);
if (pstats1_acc == NULL) {
fprintf(stderr, "%s stats1: accumulator \"%s\" not found.\n",
MLR_GLOBALS.argv0, stats1_acc_name);
exit(1);
}
lhmsv_put(acc_field_to_acc_state, stats1_acc_name, pstats1_acc, NO_FREE);
}
}
}
// ----------------------------------------------------------------
static char* test_lhmsv() {
lhmsv_t *pmap = lhmsv_alloc();
mu_assert_lf(pmap->num_occupied == 0);
mu_assert_lf(!lhmsv_has_key(pmap, "w")); mu_assert_lf(lhmsv_get(pmap, "w") == NULL);
mu_assert_lf(!lhmsv_has_key(pmap, "x")); mu_assert_lf(lhmsv_get(pmap, "x") == NULL);
mu_assert_lf(!lhmsv_has_key(pmap, "y")); mu_assert_lf(lhmsv_get(pmap, "y") == NULL);
mu_assert_lf(!lhmsv_has_key(pmap, "z")); mu_assert_lf(lhmsv_get(pmap, "z") == NULL);
mu_assert_lf(lhmsv_check_counts(pmap));
lhmsv_put(pmap, "x", "3");
mu_assert_lf(pmap->num_occupied == 1);
mu_assert_lf(!lhmsv_has_key(pmap, "w")); mu_assert_lf(lhmsv_get(pmap, "w") == NULL);
mu_assert_lf( lhmsv_has_key(pmap, "x")); mu_assert_lf(streq(lhmsv_get(pmap, "x"), "3"));
mu_assert_lf(!lhmsv_has_key(pmap, "y")); mu_assert_lf(lhmsv_get(pmap, "y") == NULL);
mu_assert_lf(!lhmsv_has_key(pmap, "z")); mu_assert_lf(lhmsv_get(pmap, "z") == NULL);
mu_assert_lf(lhmsv_check_counts(pmap));
lhmsv_put(pmap, "y", "5");
mu_assert_lf(pmap->num_occupied == 2);
mu_assert_lf(!lhmsv_has_key(pmap, "w")); mu_assert_lf(lhmsv_get(pmap, "w") == NULL);
mu_assert_lf( lhmsv_has_key(pmap, "x")); mu_assert_lf(streq(lhmsv_get(pmap, "x"), "3"));
mu_assert_lf( lhmsv_has_key(pmap, "y")); mu_assert_lf(streq(lhmsv_get(pmap, "y"), "5"));
mu_assert_lf(!lhmsv_has_key(pmap, "z")); mu_assert_lf(lhmsv_get(pmap, "z") == NULL);
mu_assert_lf(lhmsv_check_counts(pmap));
lhmsv_put(pmap, "x", "4");
mu_assert_lf(pmap->num_occupied == 2);
mu_assert_lf(!lhmsv_has_key(pmap, "w")); mu_assert_lf(lhmsv_get(pmap, "w") == NULL);
mu_assert_lf( lhmsv_has_key(pmap, "x")); mu_assert_lf(streq(lhmsv_get(pmap, "x"), "4"));
mu_assert_lf( lhmsv_has_key(pmap, "y")); mu_assert_lf(streq(lhmsv_get(pmap, "y"), "5"));
mu_assert_lf(!lhmsv_has_key(pmap, "z")); mu_assert_lf(lhmsv_get(pmap, "z") == NULL);
mu_assert_lf(lhmsv_check_counts(pmap));
lhmsv_put(pmap, "z", "7");
mu_assert_lf(pmap->num_occupied == 3);
mu_assert_lf(!lhmsv_has_key(pmap, "w")); mu_assert_lf(lhmsv_get(pmap, "w") == NULL);
mu_assert_lf( lhmsv_has_key(pmap, "x")); mu_assert_lf(streq(lhmsv_get(pmap, "x"), "4"));
mu_assert_lf( lhmsv_has_key(pmap, "y")); mu_assert_lf(streq(lhmsv_get(pmap, "y"), "5"));
mu_assert_lf( lhmsv_has_key(pmap, "z")); mu_assert_lf(streq(lhmsv_get(pmap, "z"), "7"));
mu_assert_lf(lhmsv_check_counts(pmap));
lhmsv_remove(pmap, "y");
mu_assert_lf(pmap->num_occupied == 2);
mu_assert_lf(!lhmsv_has_key(pmap, "w")); mu_assert_lf(lhmsv_get(pmap, "w") == NULL);
mu_assert_lf( lhmsv_has_key(pmap, "x")); mu_assert_lf(streq(lhmsv_get(pmap, "x"), "4"));
mu_assert_lf(!lhmsv_has_key(pmap, "y")); mu_assert_lf(lhmsv_get(pmap, "y") == NULL);
mu_assert_lf( lhmsv_has_key(pmap, "z")); mu_assert_lf(streq(lhmsv_get(pmap, "z"), "7"));
mu_assert_lf(lhmsv_check_counts(pmap));
lhmsv_free(pmap);
return NULL;
}
// ----------------------------------------------------------------
void multi_lrec_writer_output_srec(multi_lrec_writer_t* pmlw, lrec_t* poutrec, char* filename_or_command,
file_output_mode_t file_output_mode, int flush_every_record, context_t* pctx)
{
lrec_writer_and_fp_t* pstate = lhmsv_get(pmlw->pnames_to_lrec_writers_and_fps, filename_or_command);
if (pstate == NULL) {
pstate = mlr_malloc_or_die(sizeof(lrec_writer_and_fp_t));
pstate->plrec_writer = lrec_writer_alloc(pmlw->pwriter_opts);
MLR_INTERNAL_CODING_ERROR_IF(pstate->plrec_writer == NULL);
pstate->filename_or_command = mlr_strdup_or_die(filename_or_command);
char* mode_string = get_mode_string(file_output_mode);
char* mode_desc = get_mode_desc(file_output_mode);
if (file_output_mode == MODE_PIPE) {
pstate->is_popen = TRUE;
pstate->output_stream = popen(filename_or_command, mode_string);
if (pstate->output_stream == NULL) {
perror("popen");
fprintf(stderr, "%s: failed popen for %s on \"%s\".\n",
MLR_GLOBALS.bargv0, mode_desc, filename_or_command);
exit(1);
}
} else {
pstate->is_popen = FALSE;
pstate->output_stream = fopen(filename_or_command, mode_string);
if (pstate->output_stream == NULL) {
perror("fopen");
fprintf(stderr, "%s: failed fopen for %s on \"%s\".\n",
MLR_GLOBALS.bargv0, mode_desc, filename_or_command);
exit(1);
}
}
lhmsv_put(pmlw->pnames_to_lrec_writers_and_fps, mlr_strdup_or_die(filename_or_command), pstate, FREE_ENTRY_KEY);
}
pstate->plrec_writer->pprocess_func(pstate->plrec_writer->pvstate, pstate->output_stream, poutrec, pctx);
if (poutrec != NULL) {
if (flush_every_record)
fflush(pstate->output_stream);
} else {
if (pstate->is_popen) {
// Sadly, pclose returns an error even on well-formed commands. For example, if the popened
// command was "grep nonesuch" and the string "nonesuch" was not encountered, grep returns
// non-zero and popen flags it as an error. We cannot differentiate these from genuine
// failure cases so the best choice is to simply call pclose and ignore error codes.
// If a piped-to command does fail then it should have some output to stderr which the
// user can take advantage of.
(void)pclose(pstate->output_stream);
} else {
if (fclose(pstate->output_stream) != 0) {
perror("fclose");
fprintf(stderr, "%s: fclose error on \"%s\".\n", MLR_GLOBALS.bargv0, filename_or_command);
exit(1);
}
}
pstate->output_stream = NULL;
}
}
static sllv_t* mapper_step_process(lrec_t* pinrec, context_t* pctx, void* pvstate) {
mapper_step_state_t* pstate = pvstate;
if (pinrec == NULL)
return sllv_single(NULL);
// ["s", "t"]
slls_t* pvalue_field_values = mlr_selected_values_from_record(pinrec, pstate->pvalue_field_names);
slls_t* pgroup_by_field_values = mlr_selected_values_from_record(pinrec, pstate->pgroup_by_field_names);
if (pgroup_by_field_values->length != pstate->pgroup_by_field_names->length) {
lrec_free(pinrec);
return NULL;
}
lhmsv_t* group_to_acc_field = lhmslv_get(pstate->groups, pgroup_by_field_values);
if (group_to_acc_field == NULL) {
group_to_acc_field = lhmsv_alloc();
lhmslv_put(pstate->groups, slls_copy(pgroup_by_field_values), group_to_acc_field);
}
sllse_t* pa = pstate->pvalue_field_names->phead;
sllse_t* pb = pvalue_field_values->phead;
// for x=1 and y=2
for ( ; pa != NULL && pb != NULL; pa = pa->pnext, pb = pb->pnext) {
char* value_field_name = pa->value;
char* value_field_sval = pb->value;
int have_dval = FALSE;
double value_field_dval = -999.0;
lhmsv_t* acc_field_to_acc_state = lhmsv_get(group_to_acc_field, value_field_name);
if (acc_field_to_acc_state == NULL) {
acc_field_to_acc_state = lhmsv_alloc();
lhmsv_put(group_to_acc_field, value_field_name, acc_field_to_acc_state);
}
// for "delta", "rsum"
sllse_t* pc = pstate->pstepper_names->phead;
for ( ; pc != NULL; pc = pc->pnext) {
char* step_name = pc->value;
step_t* pstep = lhmsv_get(acc_field_to_acc_state, step_name);
if (pstep == NULL) {
pstep = make_step(step_name, value_field_name);
if (pstep == NULL) {
fprintf(stderr, "mlr step: stepper \"%s\" not found.\n",
step_name);
exit(1);
}
lhmsv_put(acc_field_to_acc_state, step_name, pstep);
}
if (pstep->psprocess_func != NULL) {
pstep->psprocess_func(pstep->pvstate, value_field_sval, pinrec);
}
if (pstep->pdprocess_func != NULL) {
if (!have_dval) {
value_field_dval = mlr_double_from_string_or_die(value_field_sval);
have_dval = TRUE;
}
pstep->pdprocess_func(pstep->pvstate, value_field_dval, pinrec);
}
}
}
return sllv_single(pinrec);
}
// ----------------------------------------------------------------
static void mapper_stats1_ingest(lrec_t* pinrec, mapper_stats1_state_t* pstate) {
// E.g. ["s", "t"]
// To do: make value_field_values into a hashmap. Then accept partial
// population on that, but retain full-population requirement on group-by.
// E.g. if accumulating stats of x,y on a,b then skip record with x,y,a but
// process record with x,a,b.
slls_t* pvalue_field_values = mlr_selected_values_from_record(pinrec, pstate->pvalue_field_names);
slls_t* pgroup_by_field_values = mlr_selected_values_from_record(pinrec, pstate->pgroup_by_field_names);
if (pvalue_field_values->length != pstate->pvalue_field_names->length)
return;
if (pgroup_by_field_values->length != pstate->pgroup_by_field_names->length)
return;
lhmsv_t* group_to_acc_field = lhmslv_get(pstate->groups, pgroup_by_field_values);
if (group_to_acc_field == NULL) {
group_to_acc_field = lhmsv_alloc();
lhmslv_put(pstate->groups, slls_copy(pgroup_by_field_values), group_to_acc_field);
}
sllse_t* pa = pstate->pvalue_field_names->phead;
sllse_t* pb = pvalue_field_values->phead;
// for x=1 and y=2
for ( ; pa != NULL && pb != NULL; pa = pa->pnext, pb = pb->pnext) {
char* value_field_name = pa->value;
char* value_field_sval = pb->value;
int have_dval = FALSE;
double value_field_dval = -999.0;
lhmsv_t* acc_field_to_acc_state = lhmsv_get(group_to_acc_field, value_field_name);
if (acc_field_to_acc_state == NULL) {
acc_field_to_acc_state = lhmsv_alloc();
lhmsv_put(group_to_acc_field, value_field_name, acc_field_to_acc_state);
}
// Look up presence of all accumulators at this level's hashmap.
char* presence = lhmsv_get(acc_field_to_acc_state, fake_acc_name_for_setups);
if (presence == NULL) {
make_accs(pstate->paccumulator_names, acc_field_to_acc_state);
lhmsv_put(acc_field_to_acc_state, fake_acc_name_for_setups, fake_acc_name_for_setups);
}
// There isn't a one-to-one mapping between user-specified acc_names
// and internal acc_t's. Here in the ingestor we feed each datum into
// an acc_t. In the emitter, we loop over the acc_names in
// user-specified order. Example: they ask for p10,mean,p90. Then there
// is only one percentiles accumulator to be told about each point. In
// the emitter it will be asked to produce output twice: once for the
// 10th percentile & once for the 90th.
for (lhmsve_t* pc = acc_field_to_acc_state->phead; pc != NULL; pc = pc->pnext) {
char* acc_name = pc->key;
if (streq(acc_name, fake_acc_name_for_setups))
continue;
acc_t* pacc = pc->pvvalue;
if (pacc->psingest_func != NULL) {
pacc->psingest_func(pacc->pvstate, value_field_sval);
}
if (pacc->pdingest_func != NULL) {
if (!have_dval) {
value_field_dval = mlr_double_from_string_or_die(value_field_sval);
have_dval = TRUE;
}
pacc->pdingest_func(pacc->pvstate, value_field_dval);
}
}
}
}
