void UnitEmitter::commit(UnitOrigin unitOrigin) {
Repo& repo = Repo::get();
UnitRepoProxy& urp = repo.urp();
int repoId = Repo::get().repoIdForNewUnit(unitOrigin);
if (repoId == RepoIdInvalid) {
return;
}
m_repoId = repoId;
try {
RepoTxn txn(repo);
{
LineTable lines = createLineTable(m_sourceLocTab, m_bclen);
urp.insertUnit(repoId).insert(txn, m_sn, m_md5, m_bc, m_bclen,
m_bc_meta, m_bc_meta_len,
&m_mainReturn, lines);
}
int64 usn = m_sn;
for (unsigned i = 0; i < m_litstrs.size(); ++i) {
urp.insertUnitLitstr(repoId).insert(txn, usn, i, m_litstrs[i]);
}
for (unsigned i = 0; i < m_arrays.size(); ++i) {
urp.insertUnitArray(repoId).insert(txn, usn, i, m_arrays[i].serialized);
}
for (size_t i = 0; i < m_preConsts.size(); ++i) {
urp.insertUnitPreConst(repoId).insert(txn, usn, m_preConsts[i], i);
}
for (FeVec::const_iterator it = m_fes.begin(); it != m_fes.end(); ++it) {
(*it)->commit(txn);
}
for (PceVec::const_iterator it = m_pceVec.begin(); it != m_pceVec.end();
++it) {
(*it)->commit(txn);
}
if (RuntimeOption::RepoDebugInfo) {
for (size_t i = 0; i < m_sourceLocTab.size(); ++i) {
SourceLoc& e = m_sourceLocTab[i].second;
Offset endOff = i < m_sourceLocTab.size() - 1
? m_sourceLocTab[i + 1].first
: m_bclen;
urp.insertUnitSourceLoc(repoId)
.insert(txn, usn, endOff, e.line0, e.char0, e.line1, e.char1);
}
}
txn.commit();
} catch (RepoExc& re) {
TRACE(3, "Failed to commit '%s' (0x%016llx%016llx) to '%s': %s\n",
m_filepath->data(), m_md5.q[0], m_md5.q[1],
repo.repoName(repoId).c_str(), re.msg().c_str());
}
}
Unit* UnitEmitter::create() {
Unit* u = new Unit();
u->m_repoId = m_repoId;
u->m_sn = m_sn;
u->m_bc = (uchar*)malloc(m_bclen);
memcpy(u->m_bc, m_bc, m_bclen);
u->m_bclen = m_bclen;
if (m_bc_meta_len) {
u->m_bc_meta = (uchar*)malloc(m_bc_meta_len);
memcpy(u->m_bc_meta, m_bc_meta, m_bc_meta_len);
u->m_bc_meta_len = m_bc_meta_len;
}
u->m_filepath = m_filepath;
u->m_mainReturn = m_mainReturn;
{
const std::string& dirname = Util::safe_dirname(m_filepath->data(),
m_filepath->size());
u->m_dirpath = StringData::GetStaticString(dirname);
}
u->m_md5 = m_md5;
for (unsigned i = 0; i < m_litstrs.size(); ++i) {
NamedEntityPair np;
np.first = m_litstrs[i];
np.second = NULL;
u->m_namedInfo.push_back(np);
}
u->m_array2id = m_array2id;
for (unsigned i = 0; i < m_arrays.size(); ++i) {
u->m_arrays.push_back(m_arrays[i].array);
}
for (PceVec::const_iterator it = m_pceVec.begin(); it != m_pceVec.end();
++it) {
u->m_preClasses.push_back(PreClassPtr((*it)->create(*u)));
}
size_t mergeablesSize = m_fes.size() + m_hoistablePceVec.size();
if (m_mainReturn._count && !m_allClassesHoistable) {
mergeablesSize += m_remainingPceVec.size();
}
u->m_mergeables.reserve(mergeablesSize);
u->m_firstHoistableFunc = 0;
for (FeVec::const_iterator it = m_fes.begin(); it != m_fes.end(); ++it) {
Func* func = (*it)->create(*u);
if (func->top()) {
if (!u->m_firstHoistableFunc) {
u->m_firstHoistableFunc = u->m_mergeables.size();
}
} else {
ASSERT(!u->m_firstHoistableFunc);
}
u->m_mergeables.push_back(func);
}
ASSERT(u->getMain()->isPseudoMain());
if (!u->m_firstHoistableFunc) {
u->m_firstHoistableFunc = u->m_mergeables.size();
}
u->m_firstHoistablePreClass = u->m_mergeables.size();
ASSERT(m_fes.size());
for (PceVec::const_iterator it = m_hoistablePceVec.begin();
it != m_hoistablePceVec.end(); ++it) {
u->m_mergeables.push_back(u->m_preClasses[(*it)->id()].get());
}
u->m_firstMergablePreClass = u->m_mergeables.size();
if (m_mainReturn._count && !m_allClassesHoistable) {
for (PceVec::const_iterator it = m_remainingPceVec.begin();
it != m_remainingPceVec.end(); ++it) {
u->m_mergeables.push_back(u->m_preClasses[(*it)->id()].get());
}
}
u->m_lineTable = createLineTable(m_sourceLocTab, m_bclen);
for (size_t i = 0; i < m_feTab.size(); ++i) {
ASSERT(m_feTab[i].second->past() == m_feTab[i].first);
ASSERT(m_fMap.find(m_feTab[i].second) != m_fMap.end());
u->m_funcTable.push_back(
FuncEntry(m_feTab[i].first, m_fMap.find(m_feTab[i].second)->second));
}
// Funcs can be recorded out of order when loading them from the
// repo currently. So sort 'em here.
std::sort(u->m_funcTable.begin(), u->m_funcTable.end());
m_fMap.clear();
u->m_preConsts = m_preConsts;
for (PreConstVec::iterator i = u->m_preConsts.begin();
i != u->m_preConsts.end(); ++i) {
i->owner = u;
}
if (RuntimeOption::EvalDumpBytecode) {
// Dump human-readable bytecode.
std::cout << u->toString();
}
static const bool kAlwaysVerify = getenv("HPHP_ALWAYS_VERIFY");
if (kAlwaysVerify) {
Verifier::checkUnit(u);
}
return u;
}
bool UnitEmitter::insert(UnitOrigin unitOrigin, RepoTxn& txn) {
Repo& repo = Repo::get();
UnitRepoProxy& urp = repo.urp();
int repoId = Repo::get().repoIdForNewUnit(unitOrigin);
if (repoId == RepoIdInvalid) {
return true;
}
m_repoId = repoId;
try {
{
m_lineTable = createLineTable(m_sourceLocTab, m_bclen);
urp.insertUnit[repoId].insert(*this, txn, m_sn, m_md5, m_bc,
m_bclen);
}
int64_t usn = m_sn;
urp.insertUnitLineTable(repoId, txn, usn, m_lineTable);
for (unsigned i = 0; i < m_litstrs.size(); ++i) {
urp.insertUnitLitstr[repoId].insert(txn, usn, i, m_litstrs[i]);
}
for (unsigned i = 0; i < m_arrays.size(); ++i) {
VariableSerializer vs(VariableSerializer::Type::Serialize);
urp.insertUnitArray[repoId].insert(
txn, usn, i,
vs.serializeValue(VarNR(m_arrays[i]), false /* limit */).toCppString()
);
}
for (auto& fe : m_fes) {
fe->commit(txn);
}
for (auto& pce : m_pceVec) {
pce->commit(txn);
}
for (int i = 0, n = m_mergeableStmts.size(); i < n; i++) {
switch (m_mergeableStmts[i].first) {
case MergeKind::Done:
case MergeKind::UniqueDefinedClass:
not_reached();
case MergeKind::Class: break;
case MergeKind::TypeAlias:
case MergeKind::ReqDoc: {
urp.insertUnitMergeable[repoId].insert(
txn, usn, i,
m_mergeableStmts[i].first, m_mergeableStmts[i].second, nullptr);
break;
}
case MergeKind::Define:
case MergeKind::PersistentDefine:
case MergeKind::Global: {
int ix = m_mergeableStmts[i].second;
urp.insertUnitMergeable[repoId].insert(
txn, usn, i,
m_mergeableStmts[i].first,
m_mergeableValues[ix].first, &m_mergeableValues[ix].second);
break;
}
}
}
if (RuntimeOption::RepoDebugInfo) {
for (size_t i = 0; i < m_sourceLocTab.size(); ++i) {
SourceLoc& e = m_sourceLocTab[i].second;
Offset endOff = i < m_sourceLocTab.size() - 1
? m_sourceLocTab[i + 1].first
: m_bclen;
urp.insertUnitSourceLoc[repoId]
.insert(txn, usn, endOff, e.line0, e.char0, e.line1, e.char1);
}
}
return false;
} catch (RepoExc& re) {
TRACE(3, "Failed to commit '%s' (0x%016" PRIx64 "%016" PRIx64 ") to '%s': %s\n",
m_filepath->data(), m_md5.q[0], m_md5.q[1],
repo.repoName(repoId).c_str(), re.msg().c_str());
return true;
}
}
std::unique_ptr<Unit> UnitEmitter::create() {
auto u = folly::make_unique<Unit>();
u->m_repoId = m_repoId;
u->m_sn = m_sn;
u->m_bc = allocateBCRegion(m_bc, m_bclen);
u->m_bclen = m_bclen;
u->m_filepath = m_filepath;
u->m_mainReturn = m_mainReturn;
u->m_mergeOnly = m_mergeOnly;
u->m_isHHFile = m_isHHFile;
u->m_useStrictTypes = m_useStrictTypes;
{
const std::string& dirname = FileUtil::safe_dirname(m_filepath->data(),
m_filepath->size());
u->m_dirpath = makeStaticString(dirname);
}
u->m_md5 = m_md5;
for (unsigned i = 0; i < m_litstrs.size(); ++i) {
NamedEntityPair np;
np.first = m_litstrs[i];
np.second = nullptr;
u->m_namedInfo.push_back(np);
}
u->m_arrays = [&]() -> std::vector<const ArrayData*> {
auto ret = std::vector<const ArrayData*>{};
for (unsigned i = 0; i < m_arrays.size(); ++i) {
ret.push_back(m_arrays[i]);
}
return ret;
}();
for (auto const& pce : m_pceVec) {
u->m_preClasses.push_back(PreClassPtr(pce->create(*u)));
}
u->m_typeAliases = m_typeAliases;
size_t ix = m_fes.size() + m_hoistablePceIdList.size();
if (m_mergeOnly && !m_allClassesHoistable) {
size_t extra = 0;
for (auto& mergeable : m_mergeableStmts) {
extra++;
if (!RuntimeOption::RepoAuthoritative && SystemLib::s_inited) {
if (mergeable.first != MergeKind::Class) {
extra = 0;
u->m_mergeOnly = false;
break;
}
} else {
switch (mergeable.first) {
case MergeKind::PersistentDefine:
case MergeKind::Define:
case MergeKind::Global:
extra += sizeof(TypedValueAux) / sizeof(void*);
break;
default:
break;
}
}
}
ix += extra;
}
Unit::MergeInfo *mi = Unit::MergeInfo::alloc(ix);
u->m_mergeInfo = mi;
ix = 0;
for (auto& fe : m_fes) {
Func* func = fe->create(*u);
if (func->top()) {
if (!mi->m_firstHoistableFunc) {
mi->m_firstHoistableFunc = ix;
}
} else {
assert(!mi->m_firstHoistableFunc);
}
mi->mergeableObj(ix++) = func;
}
assert(u->getMain()->isPseudoMain());
if (!mi->m_firstHoistableFunc) {
mi->m_firstHoistableFunc = ix;
}
mi->m_firstHoistablePreClass = ix;
assert(m_fes.size());
for (auto& id : m_hoistablePceIdList) {
mi->mergeableObj(ix++) = u->m_preClasses[id].get();
}
mi->m_firstMergeablePreClass = ix;
if (u->m_mergeOnly && !m_allClassesHoistable) {
for (auto& mergeable : m_mergeableStmts) {
switch (mergeable.first) {
case MergeKind::Class:
mi->mergeableObj(ix++) = u->m_preClasses[mergeable.second].get();
break;
case MergeKind::TypeAlias:
mi->mergeableObj(ix++) =
(void*)((intptr_t(mergeable.second) << 3) + (int)mergeable.first);
break;
case MergeKind::ReqDoc: {
assert(RuntimeOption::RepoAuthoritative);
void* name = u->lookupLitstrId(mergeable.second);
mi->mergeableObj(ix++) = (char*)name + (int)mergeable.first;
break;
}
case MergeKind::Define:
case MergeKind::Global:
assert(RuntimeOption::RepoAuthoritative);
case MergeKind::PersistentDefine: {
void* name = u->lookupLitstrId
(m_mergeableValues[mergeable.second].first);
mi->mergeableObj(ix++) = (char*)name + (int)mergeable.first;
auto& tv = m_mergeableValues[mergeable.second].second;
auto* tva = (TypedValueAux*)mi->mergeableData(ix);
tva->m_data = tv.m_data;
tva->m_type = tv.m_type;
// leave tva->m_aux uninitialized
ix += sizeof(*tva) / sizeof(void*);
assert(sizeof(*tva) % sizeof(void*) == 0);
break;
}
case MergeKind::Done:
case MergeKind::UniqueDefinedClass:
not_reached();
}
}
}
assert(ix == mi->m_mergeablesSize);
mi->mergeableObj(ix) = (void*)MergeKind::Done;
/*
* What's going on is we're going to have a m_lineTable if this UnitEmitter
* was loaded from the repo, and no m_sourceLocTab (it's demand-loaded by
* unit.cpp because it's only used for the debugger). Don't bother creating
* the line table here, because we can retrieve it from the repo later.
*
* On the other hand, if this unit was just created by parsing a php file (or
* whatnot) which was not committed to the repo, we'll have a m_sourceLocTab.
* In this case we should populate m_lineTable (otherwise we might lose line
* info altogether, since it may not be backed by a repo).
*/
if (m_sourceLocTab.size() != 0) {
stashLineTable(u.get(), createLineTable(m_sourceLocTab, m_bclen));
}
for (size_t i = 0; i < m_feTab.size(); ++i) {
assert(m_feTab[i].second->past == m_feTab[i].first);
assert(m_fMap.find(m_feTab[i].second) != m_fMap.end());
u->m_funcTable.push_back(
FuncEntry(m_feTab[i].first, m_fMap.find(m_feTab[i].second)->second));
}
// Funcs can be recorded out of order when loading them from the
// repo currently. So sort 'em here.
std::sort(u->m_funcTable.begin(), u->m_funcTable.end());
m_fMap.clear();
if (RuntimeOption::EvalDumpBytecode) {
// Dump human-readable bytecode.
Trace::traceRelease("%s", u->toString().c_str());
}
if (RuntimeOption::EvalDumpHhas && SystemLib::s_inited) {
std::printf("%s", disassemble(u.get()).c_str());
std::fflush(stdout);
_Exit(0);
}
static const bool kVerify = debug || getenv("HHVM_VERIFY");
static const bool kVerifyVerboseSystem =
getenv("HHVM_VERIFY_VERBOSE_SYSTEM");
static const bool kVerifyVerbose =
kVerifyVerboseSystem || getenv("HHVM_VERIFY_VERBOSE");
const bool isSystemLib = u->filepath()->empty() ||
boost::contains(u->filepath()->data(), "systemlib");
const bool doVerify =
kVerify || boost::ends_with(u->filepath()->data(), "hhas");
if (doVerify) {
auto const verbose = isSystemLib ? kVerifyVerboseSystem : kVerifyVerbose;
auto const ok = Verifier::checkUnit(u.get(), verbose);
if (!ok && !verbose) {
std::cerr << folly::format(
"Verification failed for unit {}. Re-run with HHVM_VERIFY_VERBOSE{}=1 "
"to see more details.\n",
u->filepath()->data(), isSystemLib ? "_SYSTEM" : ""
);
}
}
return u;
}
std::unique_ptr<Unit> UnitEmitter::create(bool saveLineTable) {
INC_TPC(unit_load);
std::unique_ptr<Unit> u {
RuntimeOption::RepoAuthoritative && !RuntimeOption::SandboxMode &&
m_litstrs.empty() && m_arrayTypeTable.empty() ?
new Unit : new UnitExtended
};
u->m_repoId = saveLineTable ? RepoIdInvalid : m_repoId;
u->m_sn = m_sn;
u->m_bc = allocateBCRegion(m_bc, m_bclen);
u->m_bclen = m_bclen;
u->m_filepath = m_filepath;
u->m_mainReturn = m_mainReturn;
u->m_mergeOnly = m_mergeOnly;
u->m_isHHFile = m_isHHFile;
u->m_useStrictTypes = m_useStrictTypes;
u->m_useStrictTypesForBuiltins = m_useStrictTypesForBuiltins;
u->m_dirpath = makeStaticString(FileUtil::dirname(StrNR{m_filepath}));
u->m_md5 = m_md5;
u->m_arrays = m_arrays;
for (auto const& pce : m_pceVec) {
u->m_preClasses.push_back(PreClassPtr(pce->create(*u)));
}
u->m_typeAliases = m_typeAliases;
u->m_metaData = m_metaData;
size_t ix = m_fes.size() + m_hoistablePceIdList.size();
if (m_mergeOnly && !m_allClassesHoistable) {
size_t extra = 0;
for (auto& mergeable : m_mergeableStmts) {
extra++;
if (!RuntimeOption::RepoAuthoritative && SystemLib::s_inited) {
if (mergeable.first != MergeKind::Class) {
extra = 0;
u->m_mergeOnly = false;
break;
}
} else {
switch (mergeable.first) {
case MergeKind::PersistentDefine:
case MergeKind::Define:
case MergeKind::Global:
extra += sizeof(TypedValueAux) / sizeof(void*);
break;
default:
break;
}
}
}
ix += extra;
}
Unit::MergeInfo *mi = Unit::MergeInfo::alloc(ix);
u->m_mergeInfo.store(mi, std::memory_order_relaxed);
ix = 0;
for (auto& fe : m_fes) {
Func* func = fe->create(*u);
if (func->top()) {
if (!mi->m_firstHoistableFunc) {
mi->m_firstHoistableFunc = ix;
}
} else {
assertx(!mi->m_firstHoistableFunc);
}
assertx(ix == fe->id());
mi->mergeableObj(ix++) = func;
}
assertx(u->getMain(nullptr)->isPseudoMain());
if (!mi->m_firstHoistableFunc) {
mi->m_firstHoistableFunc = ix;
}
mi->m_firstHoistablePreClass = ix;
assertx(m_fes.size());
for (auto& id : m_hoistablePceIdList) {
mi->mergeableObj(ix++) = u->m_preClasses[id].get();
}
mi->m_firstMergeablePreClass = ix;
if (u->m_mergeOnly && !m_allClassesHoistable) {
for (auto& mergeable : m_mergeableStmts) {
switch (mergeable.first) {
case MergeKind::Class:
mi->mergeableObj(ix++) = u->m_preClasses[mergeable.second].get();
break;
case MergeKind::TypeAlias:
mi->mergeableObj(ix++) =
(void*)((intptr_t(mergeable.second) << 3) + (int)mergeable.first);
break;
case MergeKind::ReqDoc: {
assertx(RuntimeOption::RepoAuthoritative);
void* name = u->lookupLitstrId(mergeable.second);
mi->mergeableObj(ix++) = (char*)name + (int)mergeable.first;
break;
}
case MergeKind::Define:
case MergeKind::Global:
assertx(RuntimeOption::RepoAuthoritative);
case MergeKind::PersistentDefine: {
void* name = u->lookupLitstrId
(m_mergeableValues[mergeable.second].first);
mi->mergeableObj(ix++) = (char*)name + (int)mergeable.first;
auto& tv = m_mergeableValues[mergeable.second].second;
auto* tva = (TypedValueAux*)mi->mergeableData(ix);
tva->m_data = tv.m_data;
tva->m_type = tv.m_type;
// leave tva->m_aux uninitialized
ix += sizeof(*tva) / sizeof(void*);
assertx(sizeof(*tva) % sizeof(void*) == 0);
break;
}
case MergeKind::Done:
case MergeKind::UniqueDefinedClass:
not_reached();
}
}
}
assertx(ix == mi->m_mergeablesSize);
mi->mergeableObj(ix) = (void*)MergeKind::Done;
/*
* What's going on is we're going to have a m_lineTable if this UnitEmitter
* was loaded from the repo, and no m_sourceLocTab (it's demand-loaded by
* unit.cpp because it's only used for the debugger). Don't bother creating
* the line table here, because we can retrieve it from the repo later.
*
* On the other hand, if this unit was just created by parsing a php file (or
* whatnot) which was not committed to the repo, we'll have a m_sourceLocTab.
* In this case we should populate m_lineTable (otherwise we might lose line
* info altogether, since it may not be backed by a repo).
*/
if (m_sourceLocTab.size() != 0) {
stashLineTable(u.get(), createLineTable(m_sourceLocTab, m_bclen));
// If we plan to dump hhas we will need the extended line table information
// in the output, and if we're not writing the repo, stashing it here is
// necessary for it to make it through.
if (RuntimeOption::RepoDebugInfo &&
RuntimeOption::EvalDumpHhas &&
SystemLib::s_inited) {
stashExtendedLineTable(u.get(), createSourceLocTable());
}
} else if (saveLineTable) {
stashLineTable(u.get(), m_lineTable);
}
if (u->m_extended) {
auto ux = u->getExtended();
for (auto s : m_litstrs) {
ux->m_namedInfo.push_back(s);
}
ux->m_arrayTypeTable = m_arrayTypeTable;
for (auto const fe : m_feTab) {
assertx(m_fMap.find(fe) != m_fMap.end());
auto const func = m_fMap.find(fe)->second;
ux->m_funcTable.push_back(func);
}
// Funcs can be recorded out of order when loading them from the
// repo currently. So sort 'em here.
std::sort(ux->m_funcTable.begin(), ux->m_funcTable.end(),
[] (const Func* a, const Func* b) {
return a->past() < b->past();
});
m_fMap.clear();
} else {
assertx(!m_litstrs.size());
assertx(m_arrayTypeTable.empty());
}
static const bool kVerify = debug || RuntimeOption::EvalVerify ||
RuntimeOption::EvalVerifyOnly;
static const bool kVerifyVerboseSystem =
getenv("HHVM_VERIFY_VERBOSE_SYSTEM");
static const bool kVerifyVerbose =
kVerifyVerboseSystem || getenv("HHVM_VERIFY_VERBOSE");
const bool isSystemLib = u->filepath()->empty() ||
boost::contains(u->filepath()->data(), "systemlib");
const bool doVerify =
kVerify || boost::ends_with(u->filepath()->data(), "hhas");
if (doVerify) {
auto const verbose = isSystemLib ? kVerifyVerboseSystem : kVerifyVerbose;
auto const ok = Verifier::checkUnit(
u.get(),
verbose ? Verifier::kVerbose : Verifier::kStderr
);
if (!ok && !verbose) {
std::cerr << folly::format(
"Verification failed for unit {}. Re-run with HHVM_VERIFY_VERBOSE{}=1 "
"to see more details.\n",
u->filepath()->data(), isSystemLib ? "_SYSTEM" : ""
);
}
}
if (RuntimeOption::EvalVerifyOnly) {
std::fflush(stdout);
_Exit(0);
}
if (RuntimeOption::EvalDumpHhas > 1 ||
(SystemLib::s_inited && RuntimeOption::EvalDumpHhas == 1)) {
auto const& hhaspath = RuntimeOption::EvalDumpHhasToFile;
if (!hhaspath.empty()) {
static std::atomic<bool> first_unit{true};
auto const flags = O_WRONLY | O_CREAT | (first_unit ? O_TRUNC : O_APPEND);
if (!folly::writeFile(disassemble(u.get()), hhaspath.c_str(), flags)) {
Logger::Error("Failed to write hhas to %s", hhaspath.c_str());
_Exit(1);
}
first_unit = false;
} else {
std::printf("%s", disassemble(u.get()).c_str());
std::fflush(stdout);
}
if (SystemLib::s_inited) {
_Exit(0);
}
}
if (RuntimeOption::EvalDumpBytecode) {
// Dump human-readable bytecode.
Trace::traceRelease("%s", u->toString().c_str());
}
return u;
}
