From e543bd34c0b4884b5a27555f698f50af6a1c0b81 Mon Sep 17 00:00:00 2001 From: chriseth Date: Thu, 10 Nov 2016 18:16:21 +0100 Subject: Stored combined creation and runtime tags. Includes a change to Assembly to allow tags from sub-assemblies to be used. Sorry, this get a bit bigger than I thought. --- libevmasm/Assembly.cpp | 102 ++++++++++++++---------- libevmasm/Assembly.h | 6 +- libevmasm/AssemblyItem.cpp | 23 ++++++ libevmasm/AssemblyItem.h | 8 ++ libevmasm/BlockDeduplicator.cpp | 37 ++++++--- libevmasm/BlockDeduplicator.h | 16 ++++ libevmasm/ControlFlowGraph.h | 5 ++ liblll/CodeFragment.cpp | 4 +- libsolidity/codegen/Compiler.cpp | 11 ++- libsolidity/codegen/Compiler.h | 8 +- libsolidity/codegen/CompilerContext.cpp | 16 ++-- libsolidity/codegen/CompilerContext.h | 45 ++++++----- libsolidity/codegen/CompilerUtils.cpp | 13 +++ libsolidity/codegen/CompilerUtils.h | 4 + libsolidity/codegen/ContractCompiler.cpp | 43 +++++++--- libsolidity/codegen/ContractCompiler.h | 10 ++- libsolidity/codegen/ExpressionCompiler.cpp | 31 +++++-- libsolidity/interface/CompilerStack.cpp | 17 +++- test/libsolidity/SolidityEndToEndTest.cpp | 63 +++++++++++++++ test/libsolidity/SolidityExpressionCompiler.cpp | 2 +- 20 files changed, 347 insertions(+), 117 deletions(-) diff --git a/libevmasm/Assembly.cpp b/libevmasm/Assembly.cpp index e19b6b0d..23a8180b 100644 --- a/libevmasm/Assembly.cpp +++ b/libevmasm/Assembly.cpp @@ -308,9 +308,20 @@ void Assembly::injectStart(AssemblyItem const& _i) Assembly& Assembly::optimise(bool _enable, bool _isCreation, size_t _runs) { - if (!_enable) - return *this; + if (_enable) + optimiseInternal(_isCreation, _runs); + return *this; +} +map Assembly::optimiseInternal(bool _isCreation, size_t _runs) +{ + for (size_t subId = 0; subId < m_subs.size(); ++subId) + { + map subTagReplacements = m_subs[subId].optimiseInternal(false, _runs); + BlockDeduplicator::applyTagReplacement(m_items, subTagReplacements, subId); + } + + map tagReplacements; unsigned total = 0; for (unsigned count = 1; count > 0; total += count) { @@ -319,30 +330,39 @@ Assembly& Assembly::optimise(bool _enable, bool _isCreation, size_t _runs) // This only modifies PushTags, we have to run again to actually remove code. BlockDeduplicator dedup(m_items); if (dedup.deduplicate()) + { + tagReplacements.insert(dedup.replacedTags().begin(), dedup.replacedTags().end()); count++; + } { - // Control flow graph that resets knowledge at path joins. - ControlFlowGraph cfg(m_items, false); + // Control flow graph optimization has been here before but is disabled because it + // assumes we only jump to tags that are pushed. This is not the case anymore with + // function types that can be stored in storage. AssemblyItems optimisedItems; - for (BasicBlock const& block: cfg.optimisedBlocks()) + + auto iter = m_items.begin(); + while (iter != m_items.end()) { - // We used to start with the block's initial state but it caused - // too many inconsistencies. + auto end = iter; + while (end != m_items.end()) + if (SemanticInformation::altersControlFlow(*end++)) + break; + KnownState emptyState; CommonSubexpressionEliminator eliminator(emptyState); - auto iter = m_items.begin() + block.begin; - auto const end = m_items.begin() + block.end; - while (iter < end) + auto blockIter = iter; + auto const blockEnd = end; + while (blockIter < blockEnd) { - auto orig = iter; - iter = eliminator.feedItems(iter, end); + auto orig = blockIter; + blockIter = eliminator.feedItems(blockIter, blockEnd); bool shouldReplace = false; AssemblyItems optimisedChunk; try { optimisedChunk = eliminator.getOptimizedItems(); - shouldReplace = (optimisedChunk.size() < size_t(iter - orig)); + shouldReplace = (optimisedChunk.size() < size_t(blockIter - orig)); } catch (StackTooDeepException const&) { @@ -361,15 +381,11 @@ Assembly& Assembly::optimise(bool _enable, bool _isCreation, size_t _runs) optimisedItems += optimisedChunk; } else - copy(orig, iter, back_inserter(optimisedItems)); + copy(orig, blockIter, back_inserter(optimisedItems)); } + iter = end; } - if (optimisedItems.size() < m_items.size()) - { - m_items = move(optimisedItems); - count++; - } } } @@ -380,10 +396,7 @@ Assembly& Assembly::optimise(bool _enable, bool _isCreation, size_t _runs) m_items ); - for (auto& sub: m_subs) - sub.optimise(true, false, _runs); - - return *this; + return tagReplacements; } LinkerObject const& Assembly::assemble() const @@ -394,8 +407,8 @@ LinkerObject const& Assembly::assemble() const LinkerObject& ret = m_assembledObject; unsigned totalBytes = bytesRequired(); - vector tagPos(m_usedTags); - map tagRef; + m_tagPositionsInBytecode = vector(m_usedTags, -1); + map> tagRef; multimap dataRef; multimap subRef; vector sizeRef; ///< Pointers to code locations where the size of the program is inserted @@ -413,8 +426,8 @@ LinkerObject const& Assembly::assemble() const for (AssemblyItem const& i: m_items) { // store position of the invalid jump destination - if (i.type() != Tag && tagPos[0] == 0) - tagPos[0] = ret.bytecode.size(); + if (i.type() != Tag && m_tagPositionsInBytecode[0] == size_t(-1)) + m_tagPositionsInBytecode[0] = ret.bytecode.size(); switch (i.type()) { @@ -446,7 +459,7 @@ LinkerObject const& Assembly::assemble() const case PushTag: { ret.bytecode.push_back(tagPush); - tagRef[ret.bytecode.size()] = (unsigned)i.data(); + tagRef[ret.bytecode.size()] = i.splitForeignPushTag(); ret.bytecode.resize(ret.bytecode.size() + bytesPerTag); break; } @@ -484,26 +497,16 @@ LinkerObject const& Assembly::assemble() const ret.bytecode.resize(ret.bytecode.size() + 20); break; case Tag: - tagPos[(unsigned)i.data()] = ret.bytecode.size(); - assertThrow(ret.bytecode.size() < 0xffffffffL, AssemblyException, "Tag too large."); assertThrow(i.data() != 0, AssemblyException, ""); + assertThrow(i.splitForeignPushTag().first == size_t(-1), AssemblyException, "Foreign tag."); + assertThrow(ret.bytecode.size() < 0xffffffffL, AssemblyException, "Tag too large."); + m_tagPositionsInBytecode[size_t(i.data())] = ret.bytecode.size(); ret.bytecode.push_back((byte)Instruction::JUMPDEST); break; default: BOOST_THROW_EXCEPTION(InvalidOpcode()); } } - for (auto const& i: tagRef) - { - bytesRef r(ret.bytecode.data() + i.first, bytesPerTag); - auto tag = i.second; - if (tag >= tagPos.size()) - tag = 0; - if (tag == 0) - assertThrow(tagPos[tag] != 0, AssemblyException, ""); - - toBigEndian(tagPos[tag], r); - } if (!dataRef.empty() && !subRef.empty()) ret.bytecode.push_back(0); @@ -519,6 +522,23 @@ LinkerObject const& Assembly::assemble() const } ret.append(m_subs[i].assemble()); } + for (auto const& i: tagRef) + { + size_t subId; + size_t tagId; + tie(subId, tagId) = i.second; + assertThrow(subId == size_t(-1) || subId < m_subs.size(), AssemblyException, "Invalid sub id"); + std::vector const& tagPositions = + subId == size_t(-1) ? + m_tagPositionsInBytecode : + m_subs[subId].m_tagPositionsInBytecode; + assertThrow(tagId < tagPositions.size(), AssemblyException, "Reference to non-existing tag."); + size_t pos = tagPositions[tagId]; + assertThrow(pos != size_t(-1), AssemblyException, "Reference to tag without position."); + assertThrow(dev::bytesRequired(pos) <= bytesPerTag, AssemblyException, "Tag too large for reserved space."); + bytesRef r(ret.bytecode.data() + i.first, bytesPerTag); + toBigEndian(pos, r); + } for (auto const& dataItem: m_data) { auto references = dataRef.equal_range(dataItem.first); diff --git a/libevmasm/Assembly.h b/libevmasm/Assembly.h index dae1e1da..0ae81e1d 100644 --- a/libevmasm/Assembly.h +++ b/libevmasm/Assembly.h @@ -53,7 +53,6 @@ public: AssemblyItem newPushSubSize(u256 const& _subId) { return AssemblyItem(PushSubSize, _subId); } AssemblyItem newPushLibraryAddress(std::string const& _identifier); - AssemblyItem append() { return append(newTag()); } void append(Assembly const& _a); void append(Assembly const& _a, int _deposit); AssemblyItem const& append(AssemblyItem const& _i); @@ -110,6 +109,10 @@ public: ) const; protected: + /// Does the same operations as @a optimise, but should only be applied to a sub and + /// returns the replaced tags. + std::map optimiseInternal(bool _isCreation, size_t _runs); + std::string locationFromSources(StringMap const& _sourceCodes, SourceLocation const& _location) const; void donePath() { if (m_totalDeposit != INT_MAX && m_totalDeposit != m_deposit) BOOST_THROW_EXCEPTION(InvalidDeposit()); } unsigned bytesRequired() const; @@ -129,6 +132,7 @@ protected: std::map m_libraries; ///< Identifiers of libraries to be linked. mutable LinkerObject m_assembledObject; + mutable std::vector m_tagPositionsInBytecode; int m_deposit = 0; int m_baseDeposit = 0; diff --git a/libevmasm/AssemblyItem.cpp b/libevmasm/AssemblyItem.cpp index 599ded85..2eae20af 100644 --- a/libevmasm/AssemblyItem.cpp +++ b/libevmasm/AssemblyItem.cpp @@ -26,6 +26,29 @@ using namespace std; using namespace dev; using namespace dev::eth; +AssemblyItem AssemblyItem::toSubAssemblyTag(size_t _subId) const +{ + assertThrow(m_data < (u256(1) << 64), Exception, "Tag already has subassembly set."); + + assertThrow(m_type == PushTag || m_type == Tag, Exception, ""); + AssemblyItem r = *this; + r.m_type = PushTag; + r.setPushTagSubIdAndTag(_subId, size_t(m_data)); + return r; +} + +pair AssemblyItem::splitForeignPushTag() const +{ + assertThrow(m_type == PushTag || m_type == Tag, Exception, ""); + return make_pair(size_t(m_data / (u256(1) << 64)) - 1, size_t(m_data)); +} + +void AssemblyItem::setPushTagSubIdAndTag(size_t _subId, size_t _tag) +{ + assertThrow(m_type == PushTag || m_type == Tag, Exception, ""); + setData(_tag + ((u256(_subId) + 1) << 64)); +} + unsigned AssemblyItem::bytesRequired(unsigned _addressLength) const { switch (m_type) diff --git a/libevmasm/AssemblyItem.h b/libevmasm/AssemblyItem.h index 1c3d9789..1a2fb1e6 100644 --- a/libevmasm/AssemblyItem.h +++ b/libevmasm/AssemblyItem.h @@ -69,6 +69,14 @@ public: AssemblyItem tag() const { assertThrow(m_type == PushTag || m_type == Tag, Exception, ""); return AssemblyItem(Tag, m_data); } AssemblyItem pushTag() const { assertThrow(m_type == PushTag || m_type == Tag, Exception, ""); return AssemblyItem(PushTag, m_data); } + /// Converts the tag to a subassembly tag. This has to be called in order to move a tag across assemblies. + /// @param _subId the identifier of the subassembly the tag is taken from. + AssemblyItem toSubAssemblyTag(size_t _subId) const; + /// @returns splits the data of the push tag into sub assembly id and actual tag id. + /// The sub assembly id of non-foreign push tags is -1. + std::pair splitForeignPushTag() const; + /// Sets sub-assembly part and tag for a push tag. + void setPushTagSubIdAndTag(size_t _subId, size_t _tag); AssemblyItemType type() const { return m_type; } u256 const& data() const { return m_data; } diff --git a/libevmasm/BlockDeduplicator.cpp b/libevmasm/BlockDeduplicator.cpp index 3bb7a797..18b595cd 100644 --- a/libevmasm/BlockDeduplicator.cpp +++ b/libevmasm/BlockDeduplicator.cpp @@ -77,7 +77,6 @@ bool BlockDeduplicator::deduplicate() { //@todo this should probably be optimized. set> blocksSeen(comparator); - map tagReplacement; for (size_t i = 0; i < m_items.size(); ++i) { if (m_items.at(i).type() != Tag) @@ -86,22 +85,40 @@ bool BlockDeduplicator::deduplicate() if (it == blocksSeen.end()) blocksSeen.insert(i); else - tagReplacement[m_items.at(i).data()] = m_items.at(*it).data(); + m_replacedTags[m_items.at(i).data()] = m_items.at(*it).data(); } - bool changed = false; - for (AssemblyItem& item: m_items) - if (item.type() == PushTag && tagReplacement.count(item.data())) - { - changed = true; - item.setData(tagReplacement.at(item.data())); - } - if (!changed) + if (!applyTagReplacement(m_items, m_replacedTags)) break; } return iterations > 0; } +bool BlockDeduplicator::applyTagReplacement( + AssemblyItems& _items, + map const& _replacements, + size_t _subId +) +{ + bool changed = false; + for (AssemblyItem& item: _items) + if (item.type() == PushTag) + { + size_t subId; + size_t tagId; + tie(subId, tagId) = item.splitForeignPushTag(); + if (subId != _subId) + continue; + auto it = _replacements.find(tagId); + if (it != _replacements.end()) + { + changed = true; + item.setPushTagSubIdAndTag(subId, size_t(it->second)); + } + } + return changed; +} + BlockDeduplicator::BlockIterator& BlockDeduplicator::BlockIterator::operator++() { if (it == end) diff --git a/libevmasm/BlockDeduplicator.h b/libevmasm/BlockDeduplicator.h index c48835fd..5ecab7f3 100644 --- a/libevmasm/BlockDeduplicator.h +++ b/libevmasm/BlockDeduplicator.h @@ -23,9 +23,12 @@ #pragma once +#include + #include #include #include +#include namespace dev { @@ -45,6 +48,18 @@ public: BlockDeduplicator(AssemblyItems& _items): m_items(_items) {} /// @returns true if something was changed bool deduplicate(); + /// @returns the tags that were replaced. + std::map const& replacedTags() const { return m_replacedTags; } + + /// Replaces all PushTag operations insied @a _items that match a key in + /// @a _replacements by the respective value. If @a _subID is not -1, only + /// apply the replacement for foreign tags from this sub id. + /// @returns true iff a replacement was performed. + static bool applyTagReplacement( + AssemblyItems& _items, + std::map const& _replacements, + size_t _subID = size_t(-1) + ); private: /// Iterator that skips tags and skips to the end if (all branches of) the control @@ -70,6 +85,7 @@ private: AssemblyItem const* replaceWith; }; + std::map m_replacedTags; AssemblyItems& m_items; }; diff --git a/libevmasm/ControlFlowGraph.h b/libevmasm/ControlFlowGraph.h index 03a1f717..78998262 100644 --- a/libevmasm/ControlFlowGraph.h +++ b/libevmasm/ControlFlowGraph.h @@ -89,6 +89,11 @@ struct BasicBlock using BasicBlocks = std::vector; +/** + * Control flow graph optimizer. + * ASSUMES THAT WE ONLY JUMP TO TAGS THAT WERE PREVIOUSLY PUSHED. THIS IS NOT TRUE ANYMORE + * NOW THAT FUNCTION TAGS CAN BE STORED IN STORAGE. + */ class ControlFlowGraph { public: diff --git a/liblll/CodeFragment.cpp b/liblll/CodeFragment.cpp index 39b6376c..73a09aad 100644 --- a/liblll/CodeFragment.cpp +++ b/liblll/CodeFragment.cpp @@ -474,7 +474,7 @@ void CodeFragment::constructOperation(sp::utree const& _t, CompilerState& _s) requireSize(2); requireDeposit(0, 1); - auto begin = m_asm.append(); + auto begin = m_asm.append(m_asm.newTag()); m_asm.append(code[0].m_asm); if (us == "WHILE") m_asm.append(Instruction::ISZERO); @@ -489,7 +489,7 @@ void CodeFragment::constructOperation(sp::utree const& _t, CompilerState& _s) requireDeposit(1, 1); m_asm.append(code[0].m_asm, 0); - auto begin = m_asm.append(); + auto begin = m_asm.append(m_asm.newTag()); m_asm.append(code[1].m_asm); m_asm.append(Instruction::ISZERO); auto end = m_asm.appendJumpI(); diff --git a/libsolidity/codegen/Compiler.cpp b/libsolidity/codegen/Compiler.cpp index bb8211ad..eefa50c5 100644 --- a/libsolidity/codegen/Compiler.cpp +++ b/libsolidity/codegen/Compiler.cpp @@ -33,11 +33,13 @@ void Compiler::compileContract( std::map const& _contracts ) { - ContractCompiler runtimeCompiler(CompilationMode::Runtime, nullptr, m_runtimeContext, m_optimize); + ContractCompiler runtimeCompiler(nullptr, m_runtimeContext, m_optimize); runtimeCompiler.compileContract(_contract, _contracts); - ContractCompiler creationCompiler(CompilationMode::Creation, &m_runtimeContext, m_context, m_optimize); - m_runtimeSub = creationCompiler.compileConstructor(m_runtimeContext, _contract, _contracts); + // This might modify m_runtimeContext because it can access runtime functions at + // creation time. + ContractCompiler creationCompiler(&runtimeCompiler, m_context, m_optimize); + m_runtimeSub = creationCompiler.compileConstructor(_contract, _contracts); if (m_optimize) m_context.optimise(m_optimizeRuns); @@ -54,7 +56,8 @@ void Compiler::compileClone( map const& _contracts ) { - ContractCompiler cloneCompiler(CompilationMode::Creation, &m_runtimeContext, m_context, m_optimize); + ContractCompiler runtimeCompiler(nullptr, m_runtimeContext, m_optimize); + ContractCompiler cloneCompiler(&runtimeCompiler, m_context, m_optimize); m_runtimeSub = cloneCompiler.compileClone(_contract, _contracts); if (m_optimize) diff --git a/libsolidity/codegen/Compiler.h b/libsolidity/codegen/Compiler.h index 56849ea0..4a87de0e 100644 --- a/libsolidity/codegen/Compiler.h +++ b/libsolidity/codegen/Compiler.h @@ -36,8 +36,8 @@ public: explicit Compiler(bool _optimize = false, unsigned _runs = 200): m_optimize(_optimize), m_optimizeRuns(_runs), - m_context(CompilationMode::Creation, &m_runtimeContext), - m_runtimeContext(CompilationMode::Runtime) + m_runtimeContext(), + m_context(&m_runtimeContext) { } void compileContract( @@ -71,9 +71,9 @@ public: private: bool const m_optimize; unsigned const m_optimizeRuns; - CompilerContext m_context; - size_t m_runtimeSub = size_t(-1); ///< Identifier of the runtime sub-assembly, if present. CompilerContext m_runtimeContext; + size_t m_runtimeSub = size_t(-1); ///< Identifier of the runtime sub-assembly, if present. + CompilerContext m_context; }; } diff --git a/libsolidity/codegen/CompilerContext.cpp b/libsolidity/codegen/CompilerContext.cpp index 3ac5bd3c..f1d306ce 100644 --- a/libsolidity/codegen/CompilerContext.cpp +++ b/libsolidity/codegen/CompilerContext.cpp @@ -92,22 +92,22 @@ eth::AssemblyItem CompilerContext::functionEntryLabelIfExists(Declaration const& return m_functionCompilationQueue.entryLabelIfExists(_declaration); } -eth::AssemblyItem CompilerContext::virtualFunctionEntryLabel(FunctionDefinition const& _function) +FunctionDefinition const& CompilerContext::resolveVirtualFunction(FunctionDefinition const& _function) { // Libraries do not allow inheritance and their functions can be inlined, so we should not // search the inheritance hierarchy (which will be the wrong one in case the function // is inlined). if (auto scope = dynamic_cast(_function.scope())) if (scope->isLibrary()) - return functionEntryLabel(_function); + return _function; solAssert(!m_inheritanceHierarchy.empty(), "No inheritance hierarchy set."); - return virtualFunctionEntryLabel(_function, m_inheritanceHierarchy.begin()); + return resolveVirtualFunction(_function, m_inheritanceHierarchy.begin()); } -eth::AssemblyItem CompilerContext::superFunctionEntryLabel(FunctionDefinition const& _function, ContractDefinition const& _base) +FunctionDefinition const& CompilerContext::superFunction(FunctionDefinition const& _function, ContractDefinition const& _base) { solAssert(!m_inheritanceHierarchy.empty(), "No inheritance hierarchy set."); - return virtualFunctionEntryLabel(_function, superContract(_base)); + return resolveVirtualFunction(_function, superContract(_base)); } FunctionDefinition const* CompilerContext::nextConstructor(ContractDefinition const& _contract) const @@ -227,7 +227,7 @@ void CompilerContext::injectVersionStampIntoSub(size_t _subIndex) sub.injectStart(fromBigEndian(binaryVersion())); } -eth::AssemblyItem CompilerContext::virtualFunctionEntryLabel( +FunctionDefinition const& CompilerContext::resolveVirtualFunction( FunctionDefinition const& _function, vector::const_iterator _searchStart ) @@ -242,9 +242,9 @@ eth::AssemblyItem CompilerContext::virtualFunctionEntryLabel( !function->isConstructor() && FunctionType(*function).hasEqualArgumentTypes(functionType) ) - return functionEntryLabel(*function); + return *function; solAssert(false, "Super function " + name + " not found."); - return m_asm.newTag(); // not reached + return _function; // not reached } vector::const_iterator CompilerContext::superContract(ContractDefinition const& _contract) const diff --git a/libsolidity/codegen/CompilerContext.h b/libsolidity/codegen/CompilerContext.h index 8b95c9f5..c4724ee0 100644 --- a/libsolidity/codegen/CompilerContext.h +++ b/libsolidity/codegen/CompilerContext.h @@ -37,10 +37,6 @@ namespace dev { namespace solidity { -/// Depending on the compilation is on the runtime code or the creation code, -/// the interpretation of internal function values differ. -enum class CompilationMode { Runtime, Creation }; - /** * Context to be shared by all units that compile the same contract. * It stores the generated bytecode and the position of identifiers in memory and on the stack. @@ -48,15 +44,13 @@ enum class CompilationMode { Runtime, Creation }; class CompilerContext { public: - CompilerContext(CompilationMode _mode, CompilerContext* _runtimeContext = nullptr) : - m_mode(_mode), m_runtimeContext(_runtimeContext) + CompilerContext(CompilerContext* _runtimeContext = nullptr) : + m_runtimeContext(_runtimeContext) { - solAssert(m_mode != CompilationMode::Runtime || !m_runtimeContext, "runtime but another runtime context provided"); - solAssert(m_mode != CompilationMode::Creation || m_runtimeContext, "creation but no runtime context provided"); + if (m_runtimeContext) + m_runtimeSub = registerSubroutine(m_runtimeContext->assembly()); } - bool isCreationPhase() const { return m_mode == CompilationMode::Creation; } - void addMagicGlobal(MagicVariableDeclaration const& _declaration); void addStateVariable(VariableDeclaration const& _declaration, u256 const& _storageOffset, unsigned _byteOffset); void addVariable(VariableDeclaration const& _declaration, unsigned _offsetToCurrent = 0); @@ -80,10 +74,10 @@ public: eth::AssemblyItem functionEntryLabelIfExists(Declaration const& _declaration) const; void setInheritanceHierarchy(std::vector const& _hierarchy) { m_inheritanceHierarchy = _hierarchy; } /// @returns the entry label of the given function and takes overrides into account. - eth::AssemblyItem virtualFunctionEntryLabel(FunctionDefinition const& _function); - /// @returns the entry label of a function that overrides the given declaration from the most derived class just + FunctionDefinition const& resolveVirtualFunction(FunctionDefinition const& _function); + /// @returns the function that overrides the given declaration from the most derived class just /// above _base in the current inheritance hierarchy. - eth::AssemblyItem superFunctionEntryLabel(FunctionDefinition const& _function, ContractDefinition const& _base); + FunctionDefinition const& superFunction(FunctionDefinition const& _function, ContractDefinition const& _base); FunctionDefinition const* nextConstructor(ContractDefinition const& _contract) const; /// @returns the next function in the queue of functions that are still to be compiled @@ -123,11 +117,15 @@ public: eth::AssemblyItem pushNewTag() { return m_asm.append(m_asm.newPushTag()).tag(); } /// @returns a new tag without pushing any opcodes or data eth::AssemblyItem newTag() { return m_asm.newTag(); } + /// Adds a subroutine to the code (in the data section) + /// @returns the assembly item corresponding to the pushed subroutine, i.e. its offset in the list. + size_t registerSubroutine(eth::Assembly const& _assembly) { return size_t(m_asm.newSub(_assembly).data()); } /// Adds a subroutine to the code (in the data section) and pushes its size (via a tag) - /// on the stack. @returns the assembly item corresponding to the pushed subroutine, i.e. its offset. - eth::AssemblyItem addSubroutine(eth::Assembly const& _assembly) { return m_asm.appendSubSize(_assembly); } + /// on the stack. @returns the pushsub assembly item. + eth::AssemblyItem addSubroutine(eth::Assembly const& _assembly) { auto sub = m_asm.newSub(_assembly); m_asm.append(m_asm.newPushSubSize(size_t(sub.data()))); return sub; } + void appendSubroutineSize(size_t const& _subRoutine) { m_asm.append(m_asm.newPushSubSize(_subRoutine)); } /// Pushes the size of the final program - void appendProgramSize() { return m_asm.appendProgramSize(); } + void appendProgramSize() { m_asm.appendProgramSize(); } /// Adds data to the data section, pushes a reference to the stack eth::AssemblyItem appendData(bytes const& _data) { return m_asm.append(_data); } /// Appends the address (virtual, will be filled in by linker) of a library. @@ -159,6 +157,11 @@ public: void optimise(unsigned _runs = 200) { m_asm.optimise(true, true, _runs); } + /// @returns the runtime context if in creation mode and runtime context is set, nullptr otherwise. + CompilerContext* runtimeContext() { return m_runtimeContext; } + /// @returns the identifier of the runtime subroutine. + size_t runtimeSub() const { return m_runtimeSub; } + eth::Assembly const& assembly() const { return m_asm; } /// @returns non-const reference to the underlying assembly. Should be avoided in favour of /// wrappers in this class. @@ -185,9 +188,9 @@ public: }; private: - /// @returns the entry label of the given function - searches the inheritance hierarchy - /// startig from the given point towards the base. - eth::AssemblyItem virtualFunctionEntryLabel( + /// Searches the inheritance hierarchy towards the base starting from @a _searchStart and returns + /// the first function definition that is overwritten by _function. + FunctionDefinition const& resolveVirtualFunction( FunctionDefinition const& _function, std::vector::const_iterator _searchStart ); @@ -241,10 +244,10 @@ private: std::vector m_inheritanceHierarchy; /// Stack of current visited AST nodes, used for location attachment std::stack m_visitedNodes; - /// The current mode of the compilation - CompilationMode m_mode; /// The runtime context if in Creation mode, this is used for generating tags that would be stored into the storage and then used at runtime. CompilerContext *m_runtimeContext; + /// The index of the runtime subroutine. + size_t m_runtimeSub = -1; }; } diff --git a/libsolidity/codegen/CompilerUtils.cpp b/libsolidity/codegen/CompilerUtils.cpp index dedb53e7..1e21c020 100644 --- a/libsolidity/codegen/CompilerUtils.cpp +++ b/libsolidity/codegen/CompilerUtils.cpp @@ -340,6 +340,19 @@ void CompilerUtils::combineExternalFunctionType(bool _leftAligned) m_context << Instruction::OR; } +void CompilerUtils::pushCombinedFunctionEntryLabel(Declaration const& _function) +{ + m_context << m_context.functionEntryLabel(_function).pushTag(); + // If there is a runtime context, we have to merge both labels into the same + // stack slot in case we store it in storage. + if (CompilerContext* rtc = m_context.runtimeContext()) + m_context << + (u256(1) << 32) << + Instruction::MUL << + rtc->functionEntryLabel(_function).toSubAssemblyTag(m_context.runtimeSub()) << + Instruction::OR; +} + void CompilerUtils::convertType(Type const& _typeOnStack, Type const& _targetType, bool _cleanupNeeded) { // For a type extension, we need to remove all higher-order bits that we might have ignored in diff --git a/libsolidity/codegen/CompilerUtils.h b/libsolidity/codegen/CompilerUtils.h index 690452f9..2ebec81a 100644 --- a/libsolidity/codegen/CompilerUtils.h +++ b/libsolidity/codegen/CompilerUtils.h @@ -120,6 +120,10 @@ public: void splitExternalFunctionType(bool _rightAligned); /// Performs the opposite operation of splitExternalFunctionType(_rightAligned) void combineExternalFunctionType(bool _rightAligned); + /// Appends code that combines the construction-time (if available) and runtime function + /// entry label of the given function into a single stack slot. + /// Note: This might cause the compilation queue of the runtime context to be extended. + void pushCombinedFunctionEntryLabel(Declaration const& _function); /// Appends code for an implicit or explicit type conversion. This includes erasing higher /// order bits (@see appendHighBitCleanup) when widening integer but also copy to memory diff --git a/libsolidity/codegen/ContractCompiler.cpp b/libsolidity/codegen/ContractCompiler.cpp index 9cd893e8..79987af6 100644 --- a/libsolidity/codegen/ContractCompiler.cpp +++ b/libsolidity/codegen/ContractCompiler.cpp @@ -60,14 +60,13 @@ void ContractCompiler::compileContract( } size_t ContractCompiler::compileConstructor( - CompilerContext const& _runtimeContext, ContractDefinition const& _contract, std::map const& _contracts ) { CompilerContext::LocationSetter locationSetter(m_context, _contract); initializeContext(_contract, _contracts); - return packIntoContractCreator(_contract, _runtimeContext); + return packIntoContractCreator(_contract); } size_t ContractCompiler::compileClone( @@ -141,21 +140,31 @@ void ContractCompiler::appendInitAndConstructorCode(ContractDefinition const& _c appendBaseConstructor(*c); } -size_t ContractCompiler::packIntoContractCreator(ContractDefinition const& _contract, CompilerContext const& _runtimeContext) +size_t ContractCompiler::packIntoContractCreator(ContractDefinition const& _contract) { + solAssert(!!m_runtimeCompiler, ""); + appendInitAndConstructorCode(_contract); - eth::AssemblyItem runtimeSub = m_context.addSubroutine(_runtimeContext.assembly()); + // We jump to the deploy routine because we first have to append all missing functions, + // which can cause further functions to be added to the runtime context. + eth::AssemblyItem deployRoutine = m_context.appendJumpToNew(); + + // We have to include copies of functions in the construction time and runtime context + // because of absolute jumps. + appendMissingFunctions(); + m_runtimeCompiler->appendMissingFunctions(); + + m_context << deployRoutine; + + solAssert(m_context.runtimeSub() != size_t(-1), "Runtime sub not registered"); + m_context.appendSubroutineSize(m_context.runtimeSub()); // stack contains sub size - m_context << Instruction::DUP1 << runtimeSub << u256(0) << Instruction::CODECOPY; + m_context << Instruction::DUP1 << m_context.runtimeSub() << u256(0) << Instruction::CODECOPY; m_context << u256(0) << Instruction::RETURN; - // note that we have to include the functions again because of absolute jump labels - appendMissingFunctions(); - - solAssert(runtimeSub.data() < numeric_limits::max(), ""); - return size_t(runtimeSub.data()); + return m_context.runtimeSub(); } void ContractCompiler::appendBaseConstructor(FunctionDefinition const& _constructor) @@ -516,7 +525,19 @@ bool ContractCompiler::visit(InlineAssembly const& _inlineAssembly) { solAssert(!!decl->type(), "Type of declaration required but not yet determined."); if (FunctionDefinition const* functionDef = dynamic_cast(decl)) - _assembly.append(m_context.virtualFunctionEntryLabel(*functionDef).pushTag()); + { + functionDef = &m_context.resolveVirtualFunction(*functionDef); + _assembly.append(m_context.functionEntryLabel(*functionDef).pushTag()); + // If there is a runtime context, we have to merge both labels into the same + // stack slot in case we store it in storage. + if (CompilerContext* rtc = m_context.runtimeContext()) + { + _assembly.append(u256(1) << 32); + _assembly.append(Instruction::MUL); + _assembly.append(rtc->functionEntryLabel(*functionDef).toSubAssemblyTag(m_context.runtimeSub())); + _assembly.append(Instruction::OR); + } + } else if (auto variable = dynamic_cast(decl)) { solAssert(!variable->isConstant(), ""); diff --git a/libsolidity/codegen/ContractCompiler.h b/libsolidity/codegen/ContractCompiler.h index fecf6f5a..9e24523c 100644 --- a/libsolidity/codegen/ContractCompiler.h +++ b/libsolidity/codegen/ContractCompiler.h @@ -38,11 +38,12 @@ namespace solidity { class ContractCompiler: private ASTConstVisitor { public: - explicit ContractCompiler(CompilationMode _mode, CompilerContext* _runtimeContext, CompilerContext& _context, bool _optimise): + explicit ContractCompiler(ContractCompiler* _runtimeCompiler, CompilerContext& _context, bool _optimise): m_optimise(_optimise), + m_runtimeCompiler(_runtimeCompiler), m_context(_context) { - m_context = CompilerContext(_mode, _runtimeContext); + m_context = CompilerContext(_runtimeCompiler ? &_runtimeCompiler->m_context : nullptr); } void compileContract( @@ -52,7 +53,6 @@ public: /// Compiles the constructor part of the contract. /// @returns the identifier of the runtime sub-assembly. size_t compileConstructor( - CompilerContext const& _runtimeContext, ContractDefinition const& _contract, std::map const& _contracts ); @@ -74,7 +74,7 @@ private: /// Adds the code that is run at creation time. Should be run after exchanging the run-time context /// with a new and initialized context. Adds the constructor code. /// @returns the identifier of the runtime sub assembly - size_t packIntoContractCreator(ContractDefinition const& _contract, CompilerContext const& _runtimeContext); + size_t packIntoContractCreator(ContractDefinition const& _contract); /// Appends state variable initialisation and constructor code. void appendInitAndConstructorCode(ContractDefinition const& _contract); void appendBaseConstructor(FunctionDefinition const& _constructor); @@ -117,6 +117,8 @@ private: static eth::Assembly cloneRuntime(); bool const m_optimise; + /// Pointer to the runtime compiler in case this is a creation compiler. + ContractCompiler* m_runtimeCompiler = nullptr; CompilerContext& m_context; std::vector m_breakTags; ///< tag to jump to for a "break" statement std::vector m_continueTags; ///< tag to jump to for a "continue" statement diff --git a/libsolidity/codegen/ExpressionCompiler.cpp b/libsolidity/codegen/ExpressionCompiler.cpp index e3f05c21..83c3a2c4 100644 --- a/libsolidity/codegen/ExpressionCompiler.cpp +++ b/libsolidity/codegen/ExpressionCompiler.cpp @@ -488,6 +488,13 @@ bool ExpressionCompiler::visit(FunctionCall const& _functionCall) parameterSize += function.selfType()->sizeOnStack(); } + if (m_context.runtimeContext()) + // We have a runtime context, so we need the creation part. + m_context << (u256(1) << 32) << Instruction::SWAP1 << Instruction::DIV; + else + // Extract the runtime part. + m_context << ((u256(1) << 32) - 1) << Instruction::AND; + m_context.appendJump(eth::AssemblyItem::JumpType::IntoFunction); m_context << returnLabel; @@ -845,9 +852,8 @@ bool ExpressionCompiler::visit(MemberAccess const& _memberAccess) ); if (funType->location() == FunctionType::Location::Internal) { - m_context << m_context.functionEntryLabel( - dynamic_cast(funType->declaration()) - ).pushTag(); + FunctionDefinition const& funDef = dynamic_cast(funType->declaration()); + utils().pushCombinedFunctionEntryLabel(funDef); utils().moveIntoStack(funType->selfType()->sizeOnStack(), 1); } else @@ -883,7 +889,7 @@ bool ExpressionCompiler::visit(MemberAccess const& _memberAccess) // us to link against it although we actually do not need it. auto const* function = dynamic_cast(_memberAccess.annotation().referencedDeclaration); solAssert(!!function, "Function not found in member access"); - m_context << m_context.functionEntryLabel(*function).pushTag(); + utils().pushCombinedFunctionEntryLabel(*function); } } else if (dynamic_cast(_memberAccess.annotation().type.get())) @@ -915,10 +921,10 @@ bool ExpressionCompiler::visit(MemberAccess const& _memberAccess) if (type.isSuper()) { solAssert(!!_memberAccess.annotation().referencedDeclaration, "Referenced declaration not resolved."); - m_context << m_context.superFunctionEntryLabel( + utils().pushCombinedFunctionEntryLabel(m_context.superFunction( dynamic_cast(*_memberAccess.annotation().referencedDeclaration), type.contractDefinition() - ).pushTag(); + )); } else { @@ -1203,7 +1209,7 @@ void ExpressionCompiler::endVisit(Identifier const& _identifier) } } else if (FunctionDefinition const* functionDef = dynamic_cast(declaration)) - m_context << m_context.virtualFunctionEntryLabel(*functionDef).pushTag(); + utils().pushCombinedFunctionEntryLabel(m_context.resolveVirtualFunction(*functionDef)); else if (auto variable = dynamic_cast(declaration)) appendVariable(*variable, static_cast(_identifier)); else if (auto contract = dynamic_cast(declaration)) @@ -1266,6 +1272,17 @@ void ExpressionCompiler::appendCompareOperatorCode(Token::Value _operator, Type { if (_operator == Token::Equal || _operator == Token::NotEqual) { + if (FunctionType const* funType = dynamic_cast(&_type)) + { + if (funType->location() == FunctionType::Location::Internal) + { + // We have to remove the upper bits (construction time value) because they might + // be "unknown" in one of the operands and not in the other. + m_context << ((u256(1) << 32) - 1) << Instruction::AND; + m_context << Instruction::SWAP1; + m_context << ((u256(1) << 32) - 1) << Instruction::AND; + } + } m_context << Instruction::EQ; if (_operator == Token::NotEqual) m_context << Instruction::ISZERO; diff --git a/libsolidity/interface/CompilerStack.cpp b/libsolidity/interface/CompilerStack.cpp index 519027bc..c86de43f 100644 --- a/libsolidity/interface/CompilerStack.cpp +++ b/libsolidity/interface/CompilerStack.cpp @@ -37,6 +37,7 @@ #include #include +#include #include #include @@ -590,9 +591,19 @@ void CompilerStack::compileContract( compiledContract.runtimeObject = compiler->runtimeObject(); _compiledContracts[compiledContract.contract] = &compiler->assembly(); - Compiler cloneCompiler(_optimize, _runs); - cloneCompiler.compileClone(_contract, _compiledContracts); - compiledContract.cloneObject = cloneCompiler.assembledObject(); + try + { + Compiler cloneCompiler(_optimize, _runs); + cloneCompiler.compileClone(_contract, _compiledContracts); + compiledContract.cloneObject = cloneCompiler.assembledObject(); + } + catch (eth::AssemblyException const& _e) + { + // In some cases (if the constructor requests a runtime function), it is not + // possible to compile the clone. + + // TODO: Report error / warning + } } std::string CompilerStack::defaultContractName() const diff --git a/test/libsolidity/SolidityEndToEndTest.cpp b/test/libsolidity/SolidityEndToEndTest.cpp index b05316dd..8f49b9c4 100644 --- a/test/libsolidity/SolidityEndToEndTest.cpp +++ b/test/libsolidity/SolidityEndToEndTest.cpp @@ -7777,6 +7777,48 @@ BOOST_AUTO_TEST_CASE(store_function_in_constructor) BOOST_CHECK(callContractFunction("result_in_constructor()") == encodeArgs(u256(4))); } +// TODO: store bound internal library functions + +BOOST_AUTO_TEST_CASE(store_internal_unused_function_in_constructor) +{ + char const* sourceCode = R"( + contract C { + function () internal returns (uint) x; + function C () { + x = unused; + } + function unused() internal returns (uint) { + return 7; + } + function t() returns (uint) { + return x(); + } + } + )"; + + compileAndRun(sourceCode, 0, "C"); + BOOST_CHECK(callContractFunction("t()") == encodeArgs(u256(7))); +} + +BOOST_AUTO_TEST_CASE(store_internal_unused_library_function_in_constructor) +{ + char const* sourceCode = R"( + library L { function x() internal returns (uint) { return 7; } } + contract C { + function () internal returns (uint) x; + function C () { + x = L.x; + } + function t() returns (uint) { + return x(); + } + } + )"; + + compileAndRun(sourceCode, 0, "C"); + BOOST_CHECK(callContractFunction("t()") == encodeArgs(u256(7))); +} + BOOST_AUTO_TEST_CASE(same_function_in_construction_and_runtime) { char const* sourceCode = R"( @@ -7799,6 +7841,27 @@ BOOST_AUTO_TEST_CASE(same_function_in_construction_and_runtime) BOOST_CHECK(callContractFunction("initial()") == encodeArgs(u256(4))); } +BOOST_AUTO_TEST_CASE(same_function_in_construction_and_runtime_equality_check) +{ + char const* sourceCode = R"( + contract C { + function (uint) internal returns (uint) x; + function C() { + x = double; + } + function test() returns (bool) { + return x == double; + } + function double(uint _arg) returns (uint _ret) { + _ret = _arg * 2; + } + } + )"; + + compileAndRun(sourceCode, 0, "C"); + BOOST_CHECK(callContractFunction("test()") == encodeArgs(true)); +} + BOOST_AUTO_TEST_CASE(function_type_library_internal) { char const* sourceCode = R"( diff --git a/test/libsolidity/SolidityExpressionCompiler.cpp b/test/libsolidity/SolidityExpressionCompiler.cpp index 09d556a5..e9a05745 100644 --- a/test/libsolidity/SolidityExpressionCompiler.cpp +++ b/test/libsolidity/SolidityExpressionCompiler.cpp @@ -136,7 +136,7 @@ bytes compileFirstExpression( FirstExpressionExtractor extractor(*contract); BOOST_REQUIRE(extractor.expression() != nullptr); - CompilerContext context { CompilationMode::Runtime /* probably simpler */ }; + CompilerContext context; context.resetVisitedNodes(contract); context.setInheritanceHierarchy(inheritanceHierarchy); unsigned parametersSize = _localVariables.size(); // assume they are all one slot on the stack -- cgit