static really_inline
void roseEodExec_i(const struct RoseEngine *t, u8 *state, u64a offset,
struct hs_scratch *scratch, const char is_streaming) {
assert(t);
assert(scratch->core_info.buf || scratch->core_info.hbuf);
assert(!scratch->core_info.buf || !scratch->core_info.hbuf);
assert(!can_stop_matching(scratch));
// Fire the special EOD event literal.
if (t->hasEodEventLiteral) {
DEBUG_PRINTF("firing eod event id %u at offset %llu\n",
t->eodLiteralId, offset);
const struct core_info *ci = &scratch->core_info;
size_t len = ci->buf ? ci->len : ci->hlen;
assert(len || !ci->buf); /* len may be 0 if no history is required
* (bounds checks only can lead to this) */
roseRunEvent(len, t->eodLiteralId, &scratch->tctxt);
if (can_stop_matching(scratch)) {
DEBUG_PRINTF("user told us to stop\n");
return;
}
}
roseCheckNfaEod(t, state, scratch, offset, is_streaming);
if (!t->eodIterOffset && !t->ematcherOffset) {
DEBUG_PRINTF("no eod accepts\n");
return;
}
// Handle pending EOD reports.
int itrv = roseEodRunIterator(t, state, offset, scratch);
if (itrv == MO_HALT_MATCHING) {
return;
}
// Run the EOD anchored matcher if there is one.
if (t->ematcherOffset) {
assert(t->ematcherRegionSize);
// Unset the reports we just fired so we don't fire them again below.
mmbit_clear(getRoleState(state), t->rolesWithStateCount);
mmbit_clear(getActiveLeafArray(t, state), t->activeArrayCount);
sidecar_enabled_populate(t, scratch, state);
hwlmcb_rv_t rv = roseEodRunMatcher(t, offset, scratch, is_streaming);
if (rv == HWLM_TERMINATE_MATCHING) {
return;
}
cleanupAfterEodMatcher(t, state, offset, scratch);
// Fire any new EOD reports.
roseEodRunIterator(t, state, offset, scratch);
roseCheckEodSuffixes(t, state, offset, scratch);
}
}
static really_inline
void do_rebuild(const struct RoseEngine *t, const struct HWLM *ftable,
struct hs_scratch *scratch) {
assert(!can_stop_matching(scratch));
size_t len = MIN(scratch->core_info.hlen, t->delayRebuildLength);
const u8 *buf = scratch->core_info.hbuf + scratch->core_info.hlen - len;
DEBUG_PRINTF("BEGIN FLOATING REBUILD over %zu bytes\n", len);
scratch->core_info.status &= ~STATUS_DELAY_DIRTY;
hwlmExec(ftable, buf, len, 0, roseDelayRebuildCallback, scratch,
scratch->tctxt.groups);
assert(!can_stop_matching(scratch));
}
int roseAnchoredCallback(u64a start, u64a end, u32 id, void *ctx) {
struct hs_scratch *scratch = ctx;
assert(scratch && scratch->magic == SCRATCH_MAGIC);
struct RoseContext *tctxt = &scratch->tctxt;
struct core_info *ci = &scratch->core_info;
const struct RoseEngine *t = ci->rose;
u64a real_end = ci->buf_offset + end; // index after last byte
DEBUG_PRINTF("MATCH id=%u offsets=[???,%llu]\n", id, real_end);
DEBUG_PRINTF("STATE groups=0x%016llx\n", tctxt->groups);
if (can_stop_matching(scratch)) {
DEBUG_PRINTF("received a match when we're already dead!\n");
return MO_HALT_MATCHING;
}
const size_t match_len = 0;
/* delayed literals need to be delivered before real literals; however
* delayed literals only come from the floating table so if we are going
* to deliver a literal here it must be too early for a delayed literal */
/* no history checks from anchored region and we are before the flush
* boundary */
if (real_end <= t->floatingMinLiteralMatchOffset) {
roseFlushLastByteHistory(t, scratch, real_end);
tctxt->lastEndOffset = real_end;
}
const u32 *programs = getByOffset(t, t->litProgramOffset);
assert(id < t->literalCount);
const u8 flags = ROSE_PROG_FLAG_IN_ANCHORED;
if (roseRunProgram(t, scratch, programs[id], start, real_end, match_len,
flags) == HWLM_TERMINATE_MATCHING) {
assert(can_stop_matching(scratch));
DEBUG_PRINTF("caller requested termination\n");
return MO_HALT_MATCHING;
}
DEBUG_PRINTF("DONE groups=0x%016llx\n", tctxt->groups);
if (real_end > t->floatingMinLiteralMatchOffset) {
recordAnchoredLiteralMatch(t, scratch, id, real_end);
}
return MO_CONTINUE_MATCHING;
}
static really_inline
hwlmcb_rv_t roseCallback_i(size_t start, size_t end, u32 id, void *ctxt) {
struct hs_scratch *scratch = ctxt;
struct RoseContext *tctx = &scratch->tctxt;
const struct RoseEngine *t = scratch->core_info.rose;
u64a real_end = end + tctx->lit_offset_adjust;
#if defined(DEBUG)
DEBUG_PRINTF("MATCH id=%u offsets=[%llu,%llu]: ", id,
start + tctx->lit_offset_adjust, real_end);
printMatch(&scratch->core_info, start + tctx->lit_offset_adjust, real_end);
printf("\n");
#endif
DEBUG_PRINTF("last end %llu\n", tctx->lastEndOffset);
DEBUG_PRINTF("STATE groups=0x%016llx\n", tctx->groups);
if (can_stop_matching(scratch)) {
DEBUG_PRINTF("received a match when we're already dead!\n");
return HWLM_TERMINATE_MATCHING;
}
hwlmcb_rv_t rv = flushQueuedLiterals(t, scratch, real_end);
/* flushDelayed may have advanced tctx->lastEndOffset */
if (real_end >= t->floatingMinLiteralMatchOffset) {
roseFlushLastByteHistory(t, scratch, real_end);
tctx->lastEndOffset = real_end;
}
if (rv == HWLM_TERMINATE_MATCHING) {
return HWLM_TERMINATE_MATCHING;
}
size_t match_len = end - start + 1;
rv = roseProcessMatchInline(t, scratch, real_end, match_len, id);
DEBUG_PRINTF("DONE groups=0x%016llx\n", tctx->groups);
if (rv != HWLM_TERMINATE_MATCHING) {
return tctx->groups;
}
assert(can_stop_matching(scratch));
DEBUG_PRINTF("user requested halt\n");
return HWLM_TERMINATE_MATCHING;
}
static never_inline
void soleOutfixStreamExec(struct hs_stream *stream_state,
struct hs_scratch *scratch) {
assert(stream_state);
assert(scratch);
assert(!can_stop_matching(scratch));
const struct RoseEngine *t = stream_state->rose;
assert(t->outfixEndQueue == 1);
assert(!t->amatcherOffset);
assert(!t->ematcherOffset);
assert(!t->fmatcherOffset);
const struct NFA *nfa = getNfaByQueue(t, 0);
struct mq *q = scratch->queues;
initOutfixQueue(q, 0, t, scratch);
if (!scratch->core_info.buf_offset) {
nfaQueueInitState(nfa, q);
pushQueueAt(q, 0, MQE_START, 0);
pushQueueAt(q, 1, MQE_TOP, 0);
pushQueueAt(q, 2, MQE_END, scratch->core_info.len);
} else {
nfaExpandState(nfa, q->state, q->streamState, q->offset,
queue_prev_byte(q, 0));
pushQueueAt(q, 0, MQE_START, 0);
pushQueueAt(q, 1, MQE_END, scratch->core_info.len);
}
if (nfaQueueExec(q->nfa, q, scratch->core_info.len)) {
nfaQueueCompressState(nfa, q, scratch->core_info.len);
} else if (!told_to_stop_matching(scratch)) {
scratch->core_info.status |= STATUS_EXHAUSTED;
}
}
static never_inline
void soleOutfixEodExec(hs_stream_t *id, hs_scratch_t *scratch) {
const struct RoseEngine *t = id->rose;
if (can_stop_matching(scratch)) {
DEBUG_PRINTF("stream already broken\n");
return;
}
if (isAllExhausted(t, scratch->core_info.exhaustionVector)) {
DEBUG_PRINTF("stream exhausted\n");
return;
}
assert(t->outfixEndQueue == 1);
assert(!t->amatcherOffset);
assert(!t->ematcherOffset);
assert(!t->fmatcherOffset);
const struct NFA *nfa = getNfaByQueue(t, 0);
struct mq *q = scratch->queues;
initOutfixQueue(q, 0, t, scratch);
if (!scratch->core_info.buf_offset) {
DEBUG_PRINTF("buf_offset is zero\n");
return; /* no vacuous engines */
}
nfaExpandState(nfa, q->state, q->streamState, q->offset,
queue_prev_byte(q, 0));
assert(nfaAcceptsEod(nfa));
nfaCheckFinalState(nfa, q->state, q->streamState, q->offset, q->cb,
q->som_cb, scratch);
}
void roseBlockEodExec(const struct RoseEngine *t, u64a offset,
struct hs_scratch *scratch) {
assert(t->requiresEodCheck);
assert(t->maxBiAnchoredWidth == ROSE_BOUND_INF
|| offset <= t->maxBiAnchoredWidth);
assert(!can_stop_matching(scratch));
u8 *state = (u8 *)scratch->core_info.state;
// Ensure that history is correct before we look for EOD matches
prepForEod(t, state, scratch->core_info.len, &scratch->tctxt);
roseEodExec_i(t, state, offset, scratch, 0);
}
static really_inline
void rawEodExec(hs_stream_t *id, hs_scratch_t *scratch) {
const struct RoseEngine *rose = id->rose;
if (can_stop_matching(scratch)) {
DEBUG_PRINTF("stream already broken\n");
return;
}
if (isAllExhausted(rose, scratch->core_info.exhaustionVector)) {
DEBUG_PRINTF("stream exhausted\n");
return;
}
roseEodExec(rose, id->offset, scratch);
}
static really_inline
void rawStreamExec(struct hs_stream *stream_state, struct hs_scratch *scratch) {
assert(stream_state);
assert(scratch);
assert(!can_stop_matching(scratch));
DEBUG_PRINTF("::: streaming rose ::: offset = %llu len = %zu\n",
stream_state->offset, scratch->core_info.len);
const struct RoseEngine *rose = stream_state->rose;
assert(rose);
roseStreamExec(rose, scratch);
if (!told_to_stop_matching(scratch) &&
isAllExhausted(rose, scratch->core_info.exhaustionVector)) {
DEBUG_PRINTF("stream exhausted\n");
scratch->core_info.status |= STATUS_EXHAUSTED;
}
}
char nfaExecTamarama0_Q(const struct NFA *n, struct mq *q, s64a end) {
DEBUG_PRINTF("exec\n");
struct mq q1;
char rv = MO_ALIVE;
char copy = 0;
const struct Tamarama *t = getImplNfa(n);
while (q->cur < q->end && q_cur_loc(q) <= end) {
updateQueues(t, q, &q1);
rv = nfaQueueExec_raw(q1.nfa, &q1, end);
q->report_current = q1.report_current;
copy = 1;
if (can_stop_matching(q->scratch)) {
break;
}
}
if (copy) {
copyBack(t, q, &q1);
}
return rv;
}
int roseReportAdaptor(u64a start, u64a end, ReportID id, void *context) {
struct hs_scratch *scratch = context;
assert(scratch && scratch->magic == SCRATCH_MAGIC);
DEBUG_PRINTF("id=%u matched at [%llu,%llu]\n", id, start, end);
const struct RoseEngine *rose = scratch->core_info.rose;
// Our match ID is the program offset.
const u32 program = id;
const size_t match_len = 0; // Unused in this path.
const u8 flags = ROSE_PROG_FLAG_SKIP_MPV_CATCHUP;
hwlmcb_rv_t rv =
roseRunProgram(rose, scratch, program, start, end, match_len, flags);
if (rv == HWLM_TERMINATE_MATCHING) {
return MO_HALT_MATCHING;
}
return can_stop_matching(scratch) ? MO_HALT_MATCHING : MO_CONTINUE_MATCHING;
}
static really_inline
void pureLiteralStreamExec(struct hs_stream *stream_state,
struct hs_scratch *scratch) {
assert(stream_state);
assert(scratch);
assert(!can_stop_matching(scratch));
char *state = getMultiState(stream_state);
const struct RoseEngine *rose = stream_state->rose;
const struct HWLM *ftable = getFLiteralMatcher(rose);
size_t len2 = scratch->core_info.len;
u8 *hwlm_stream_state;
if (rose->floatingStreamState) {
hwlm_stream_state = getFloatingMatcherState(rose, state);
} else {
hwlm_stream_state = NULL;
}
DEBUG_PRINTF("::: streaming rose ::: offset = %llu len = %zu\n",
stream_state->offset, scratch->core_info.len);
// Pure literal cases don't have floatingMinDistance set, so we always
// start the match region at zero.
const size_t start = 0;
hwlmExecStreaming(ftable, scratch, len2, start, rosePureLiteralCallback,
scratch, rose->initialGroups, hwlm_stream_state);
if (!told_to_stop_matching(scratch) &&
isAllExhausted(rose, scratch->core_info.exhaustionVector)) {
DEBUG_PRINTF("stream exhausted\n");
scratch->core_info.status |= STATUS_EXHAUSTED;
}
}
/**
* \brief Execute a boundary report program.
*
* Returns MO_HALT_MATCHING if the stream is exhausted or the user has
* instructed us to halt, or MO_CONTINUE_MATCHING otherwise.
*/
int roseRunBoundaryProgram(const struct RoseEngine *rose, u32 program,
u64a stream_offset, struct hs_scratch *scratch) {
DEBUG_PRINTF("running boundary program at offset %u\n", program);
if (can_stop_matching(scratch)) {
DEBUG_PRINTF("can stop matching\n");
return MO_HALT_MATCHING;
}
if (rose->hasSom && scratch->deduper.current_report_offset == ~0ULL) {
/* we cannot delay the initialization of the som deduper logs any longer
* as we are reporting matches. This is done explicitly as we are
* shortcutting the som handling in the vacuous repeats as we know they
* all come from non-som patterns. */
fatbit_clear(scratch->deduper.som_log[0]);
fatbit_clear(scratch->deduper.som_log[1]);
scratch->deduper.som_log_dirty = 0;
}
// Keep assertions in program report path happy. At offset zero, there can
// have been no earlier reports. At EOD, all earlier reports should have
// been handled and we will have been caught up to the stream offset by the
// time we are running boundary report programs.
scratch->tctxt.minMatchOffset = stream_offset;
const u64a som = 0;
const size_t match_len = 0;
const u8 flags = 0;
hwlmcb_rv_t rv = roseRunProgram(rose, scratch, program, som, stream_offset,
match_len, flags);
if (rv == HWLM_TERMINATE_MATCHING) {
return MO_HALT_MATCHING;
}
return MO_CONTINUE_MATCHING;
}
void roseStreamExec(const struct RoseEngine *t, struct hs_scratch *scratch) {
DEBUG_PRINTF("OH HAI [%llu, %llu)\n", scratch->core_info.buf_offset,
scratch->core_info.buf_offset + (u64a)scratch->core_info.len);
assert(t);
assert(scratch->core_info.hbuf);
assert(scratch->core_info.buf);
// We should not have been called if we've already been told to terminate
// matching.
assert(!told_to_stop_matching(scratch));
assert(mmbit_sparse_iter_state_size(t->rolesWithStateCount)
< MAX_SPARSE_ITER_STATES);
size_t length = scratch->core_info.len;
u64a offset = scratch->core_info.buf_offset;
// We may have a maximum width (for engines constructed entirely
// of bi-anchored patterns). If this write would result in us progressing
// beyond this point, we cannot possibly match.
if (t->maxBiAnchoredWidth != ROSE_BOUND_INF
&& offset + length > t->maxBiAnchoredWidth) {
DEBUG_PRINTF("bailing, write would progress beyond maxBAWidth\n");
return;
}
char *state = scratch->core_info.state;
struct RoseContext *tctxt = &scratch->tctxt;
tctxt->mpv_inactive = 0;
tctxt->groups = loadGroups(t, state);
tctxt->lit_offset_adjust = offset + 1; // index after last byte
tctxt->delayLastEndOffset = offset;
tctxt->lastEndOffset = offset;
tctxt->filledDelayedSlots = 0;
tctxt->lastMatchOffset = 0;
tctxt->minMatchOffset = offset;
tctxt->minNonMpvMatchOffset = offset;
tctxt->next_mpv_offset = 0;
DEBUG_PRINTF("BEGIN: history len=%zu, buffer len=%zu groups=%016llx\n",
scratch->core_info.hlen, scratch->core_info.len, tctxt->groups);
fatbit_clear(scratch->aqa);
scratch->al_log_sum = 0;
scratch->catchup_pq.qm_size = 0;
if (t->outfixBeginQueue != t->outfixEndQueue) {
streamInitSufPQ(t, state, scratch);
}
runEagerPrefixesStream(t, scratch);
u32 alen = t->anchoredDistance > offset ?
MIN(length + offset, t->anchoredDistance) - offset : 0;
const struct anchored_matcher_info *atable = getALiteralMatcher(t);
if (atable && alen) {
DEBUG_PRINTF("BEGIN ANCHORED %zu/%u\n", scratch->core_info.hlen, alen);
runAnchoredTableStream(t, atable, alen, offset, scratch);
if (can_stop_matching(scratch)) {
goto exit;
}
}
const struct HWLM *ftable = getFLiteralMatcher(t);
if (ftable) {
if (t->noFloatingRoots && !roseHasInFlightMatches(t, state, scratch)) {
DEBUG_PRINTF("skip FLOATING: no inflight matches\n");
goto flush_delay_and_exit;
}
size_t flen = length;
if (t->floatingDistance != ROSE_BOUND_INF) {
flen = t->floatingDistance > offset ?
MIN(t->floatingDistance, length + offset) - offset : 0;
}
size_t hlength = scratch->core_info.hlen;
char rebuild = hlength &&
(scratch->core_info.status & STATUS_DELAY_DIRTY) &&
(t->maxFloatingDelayedMatch == ROSE_BOUND_INF ||
offset < t->maxFloatingDelayedMatch);
DEBUG_PRINTF("**rebuild %hhd status %hhu mfdm %u, offset %llu\n",
rebuild, scratch->core_info.status,
t->maxFloatingDelayedMatch, offset);
if (!flen) {
if (rebuild) { /* rebuild floating delayed match stuff */
do_rebuild(t, ftable, scratch);
}
goto flush_delay_and_exit;
}
if (rebuild) { /* rebuild floating delayed match stuff */
do_rebuild(t, ftable, scratch);
}
if (flen + offset <= t->floatingMinDistance) {
DEBUG_PRINTF("skip FLOATING: before floating min\n");
goto flush_delay_and_exit;
}
size_t start = 0;
if (offset < t->floatingMinDistance) {
// This scan crosses the floating min distance, so we can use that
// to set HWLM's "start" offset.
start = t->floatingMinDistance - offset;
}
DEBUG_PRINTF("start=%zu\n", start);
u8 *stream_state;
if (t->floatingStreamState) {
stream_state = getFloatingMatcherState(t, state);
} else {
stream_state = NULL;
}
DEBUG_PRINTF("BEGIN FLOATING (over %zu/%zu)\n", flen, length);
hwlmExecStreaming(ftable, scratch, flen, start, roseFloatingCallback,
scratch, tctxt->groups & t->floating_group_mask,
stream_state);
}
flush_delay_and_exit:
DEBUG_PRINTF("flushing floating\n");
if (cleanUpDelayed(t, scratch, length, offset) == HWLM_TERMINATE_MATCHING) {
return;
}
exit:
DEBUG_PRINTF("CLEAN UP TIME\n");
if (!can_stop_matching(scratch)) {
ensureStreamNeatAndTidy(t, state, scratch, length, offset);
}
DEBUG_PRINTF("DONE STREAMING SCAN, status = %u\n",
scratch->core_info.status);
return;
}
static really_inline
int roseSomAdaptor_i(u64a from_offset, u64a to_offset, ReportID id,
void *context, char is_simple) {
assert(id != MO_INVALID_IDX); // Should never get an invalid ID.
u32 flags = 0;
struct hs_scratch *scratch = (struct hs_scratch *)context;
struct core_info *ci = &scratch->core_info;
const struct RoseEngine *rose = ci->rose;
const struct internal_report *ri = getInternalReport(rose, id);
/* internal events should be handled by rose directly */
assert(ri->type == EXTERNAL_CALLBACK);
DEBUG_PRINTF("internal match at %llu: IID=%u type=%hhu RID=%u "
"offsetAdj=%d\n", to_offset, id, ri->type, ri->onmatch,
ri->offsetAdjust);
if (unlikely(can_stop_matching(scratch))) {
DEBUG_PRINTF("pre broken - halting\n");
return MO_HALT_MATCHING;
}
if (!is_simple && ri->hasBounds) {
assert(ri->minOffset || ri->minLength || ri->maxOffset < MAX_OFFSET);
if (to_offset < ri->minOffset || to_offset > ri->maxOffset) {
DEBUG_PRINTF("match fell outside valid range %llu !: [%llu,%llu]\n",
to_offset, ri->minOffset, ri->maxOffset);
return MO_CONTINUE_MATCHING;
}
}
int halt = 0;
if (!is_simple && unlikely(isExhausted(ci->exhaustionVector, ri->ekey))) {
DEBUG_PRINTF("ate exhausted match\n");
goto do_return;
}
#ifdef DEDUPE_MATCHES
u64a offset = to_offset;
#endif
to_offset += ri->offsetAdjust;
assert(from_offset == HS_OFFSET_PAST_HORIZON || from_offset <= to_offset);
if (!is_simple && ri->minLength) {
if (from_offset != HS_OFFSET_PAST_HORIZON &&
(to_offset - from_offset < ri->minLength)) {
return MO_CONTINUE_MATCHING;
}
if (ri->quashSom) {
from_offset = 0;
}
}
DEBUG_PRINTF(">> reporting match @[%llu,%llu] for sig %u ctxt %p <<\n",
from_offset, to_offset, ri->onmatch, ci->userContext);
#ifndef RELEASE_BUILD
if (ri->offsetAdjust != 0) {
// alert testing tools that we've got adjusted matches
flags |= HS_MATCH_FLAG_ADJUSTED;
}
#endif
#ifdef DEDUPE_MATCHES
u32 dkeyCount = rose->dkeyCount;
if (offset != scratch->deduper.current_report_offset) {
assert(scratch->deduper.current_report_offset == ~0ULL
|| scratch->deduper.current_report_offset < offset);
if (offset == scratch->deduper.current_report_offset + 1) {
fatbit_clear(scratch->deduper.log[offset % 2]);
} else {
fatbit_clear(scratch->deduper.log[0]);
fatbit_clear(scratch->deduper.log[1]);
}
halt = flushStoredSomMatches(scratch, offset);
if (halt) {
goto do_return;
}
scratch->deduper.current_report_offset = offset;
}
u32 dkey = ri->dkey;
if (dkey != MO_INVALID_IDX) {
if (ri->quashSom) {
DEBUG_PRINTF("checking dkey %u at offset %llu\n", dkey, to_offset);
assert(ri->offsetAdjust == 0 || ri->offsetAdjust == -1);
if (fatbit_set(scratch->deduper.log[to_offset % 2], dkeyCount,
dkey)) {
/* we have already raised this report at this offset, squash
* dupe match. */
DEBUG_PRINTF("dedupe\n");
goto do_return;
}
} else {
/* SOM external event */
DEBUG_PRINTF("checking dkey %u at offset %llu\n", dkey, to_offset);
assert(ri->offsetAdjust == 0 || ri->offsetAdjust == -1);
u64a *starts = scratch->deduper.som_start_log[to_offset % 2];
if (fatbit_set(scratch->deduper.som_log[to_offset % 2], dkeyCount,
dkey)) {
starts[dkey] = MIN(starts[dkey], from_offset);
} else {
starts[dkey] = from_offset;
}
if (ri->offsetAdjust) {
scratch->deduper.som_log_dirty |= 1;
} else {
scratch->deduper.som_log_dirty |= 2;
}
goto do_return;
}
}
#endif
halt = ci->userCallback((unsigned int)ri->onmatch, from_offset, to_offset,
flags, ci->userContext);
if (!is_simple) {
markAsMatched(ci->exhaustionVector, ri->ekey);
}
do_return:
if (halt) {
DEBUG_PRINTF("callback requested to terminate matches\n");
setBroken(ci->state, BROKEN_FROM_USER);
ci->broken = BROKEN_FROM_USER;
return MO_HALT_MATCHING;
}
return MO_CONTINUE_MATCHING;
}
static inline
hs_error_t hs_scan_stream_internal(hs_stream_t *id, const char *data,
unsigned length, UNUSED unsigned flags,
hs_scratch_t *scratch,
match_event_handler onEvent, void *context) {
if (unlikely(!id || !scratch || !data || !validScratch(id->rose, scratch))) {
return HS_INVALID;
}
const struct RoseEngine *rose = id->rose;
char *state = getMultiState(id);
u8 broken = getBroken(state);
if (broken) {
DEBUG_PRINTF("stream is broken, halting scan\n");
if (broken == BROKEN_FROM_USER) {
return HS_SCAN_TERMINATED;
} else {
assert(broken == BROKEN_EXHAUSTED);
return HS_SUCCESS;
}
}
// We avoid doing any work if the user has given us zero bytes of data to
// scan. Arguably we should define some semantics for how we treat vacuous
// cases here.
if (unlikely(length == 0)) {
DEBUG_PRINTF("zero length block\n");
assert(getBroken(state) != BROKEN_FROM_USER);
return HS_SUCCESS;
}
u32 historyAmount = getHistoryAmount(rose, id->offset);
populateCoreInfo(scratch, rose, state, onEvent, context, data, length,
getHistory(state, rose, id->offset), historyAmount,
id->offset, flags);
assert(scratch->core_info.hlen <= id->offset
&& scratch->core_info.hlen <= rose->historyRequired);
prefetch_data(data, length);
if (rose->somLocationCount) {
loadSomFromStream(scratch, id->offset);
}
if (!id->offset && rose->boundary.reportZeroOffset) {
DEBUG_PRINTF("zero reports\n");
processReportList(rose, rose->boundary.reportZeroOffset, 0, scratch);
}
switch (rose->runtimeImpl) {
default:
assert(0);
case ROSE_RUNTIME_FULL_ROSE:
rawStreamExec(id, scratch);
break;
case ROSE_RUNTIME_PURE_LITERAL:
pureLiteralStreamExec(id, scratch);
break;
case ROSE_RUNTIME_SINGLE_OUTFIX:
soleOutfixStreamExec(id, scratch);
}
if (rose->hasSom && !told_to_stop_matching(scratch)) {
int halt = flushStoredSomMatches(scratch, ~0ULL);
if (halt) {
setBroken(state, BROKEN_FROM_USER);
scratch->core_info.broken = BROKEN_FROM_USER;
}
}
if (likely(!can_stop_matching(scratch))) {
maintainHistoryBuffer(id->rose, getMultiState(id), data, length);
id->offset += length; /* maintain offset */
if (rose->somLocationCount) {
storeSomToStream(scratch, id->offset);
}
} else if (told_to_stop_matching(scratch)) {
return HS_SCAN_TERMINATED;
} else { /* exhausted */
setBroken(state, BROKEN_EXHAUSTED);
}
return HS_SUCCESS;
}
