X-Git-Url: http://plrg.eecs.uci.edu/git/?a=blobdiff_plain;f=lib%2FLinker%2FLinkArchives.cpp;h=99f3a1b60d28497fcb3d6a7d3a50d85da48789f5;hb=12945acd4aa40b40b1517bcf785622ab9671893f;hp=0d0dc9fabaa2dc178096de7a085d8d02de5cff50;hpb=a58d2be76f90aaeeccd89e9794ccd053f370cf8e;p=oota-llvm.git diff --git a/lib/Linker/LinkArchives.cpp b/lib/Linker/LinkArchives.cpp index 0d0dc9fabaa..99f3a1b60d2 100644 --- a/lib/Linker/LinkArchives.cpp +++ b/lib/Linker/LinkArchives.cpp @@ -1,150 +1,35 @@ -//===- Linker.cpp - Link together LLVM objects and libraries --------------===// +//===- lib/Linker/LinkArchives.cpp - Link LLVM objects and libraries ------===// +// +// The LLVM Compiler Infrastructure +// +// This file was developed by the LLVM research group and is distributed under +// the University of Illinois Open Source License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// // // This file contains routines to handle linking together LLVM bytecode files, // and to handle annoying things like static libraries. // //===----------------------------------------------------------------------===// -#include "gccld.h" +#include "llvm/Linker.h" #include "llvm/Module.h" -#include "llvm/PassManager.h" +#include "llvm/ModuleProvider.h" +#include "llvm/ADT/SetOperations.h" #include "llvm/Bytecode/Reader.h" -#include "llvm/Bytecode/WriteBytecodePass.h" -#include "llvm/Target/TargetData.h" -#include "llvm/Transforms/IPO.h" -#include "llvm/Transforms/Scalar.h" -#include "llvm/Transforms/Utils/Linker.h" -#include "Support/CommandLine.h" -#include "Support/FileUtilities.h" -#include "Support/Signals.h" -#include "Support/SystemUtils.h" -#include "Config/stdlib.h" -#include -#include +#include "llvm/Bytecode/Archive.h" +#include "llvm/Config/config.h" #include #include +#include -// -// Function: FileExists () -// -// Description: -// Determine if the specified filename exists and is readable. -// -// Inputs: -// FN - The name of the file. -// -// Outputs: -// None. -// -// Return Value: -// TRUE - The file exists and is readable. -// FALSE - The file does not exist or is unreadable. -// -static inline bool FileExists(const std::string &FN) { - return access(FN.c_str(), R_OK | F_OK) != -1; -} - -// -// Function: IsArchive () -// -// Description: -// Determine if the specified file is an ar archive. It determines this by -// checking the magic string at the beginning of the file. -// -// Inputs: -// filename - A C++ string containing the name of the file. -// -// Outputs: -// None. -// -// Return value: -// TRUE - The file is an archive. -// FALSE - The file is not an archive. -// -static inline bool IsArchive(const std::string &filename) -{ - std::string ArchiveMagic("!\012"); - char buf[1 + ArchiveMagic.size()]; - std::ifstream f(filename.c_str()); - f.read(buf, ArchiveMagic.size()); - buf[ArchiveMagic.size()] = '\0'; - return ArchiveMagic == buf; -} - -// -// Function: FindLib () -// -// Description: -// This function locates a particular library. It will prepend and append -// various directories, prefixes, and suffixes until it can find the library. -// -// Inputs: -// Filename - Name of the file to find. -// Paths - List of directories to search. -// -// Outputs: -// None. -// -// Return value: -// The name of the file is returned. -// If the file is not found, an empty string is returned. -// -static std::string -FindLib(const std::string &Filename, const std::vector &Paths) -{ - // Determine if the pathname can be found as it stands. - if (FileExists(Filename)) - return Filename; - - // If that doesn't work, convert the name into a library name. - std::string LibName = "lib" + Filename; - - // Iterate over the directories in Paths to see if we can find the library - // there. - for (unsigned Index = 0; Index != Paths.size(); ++Index) { - std::string Directory = Paths[Index] + "/"; +using namespace llvm; - if (FileExists(Directory + LibName + ".bc")) - return Directory + LibName + ".bc"; - - if (FileExists(Directory + LibName + ".so")) - return Directory + LibName + ".so"; - - if (FileExists(Directory + LibName + ".a")) - return Directory + LibName + ".a"; - } - - // One last hope: Check LLVM_LIB_SEARCH_PATH. - char *SearchPath = getenv("LLVM_LIB_SEARCH_PATH"); - if (SearchPath == NULL) - return std::string(); - - LibName = std::string(SearchPath) + "/" + LibName; - if (FileExists(LibName)) - return LibName; - - return std::string(); -} - -// -// Function: GetAllDefinedSymbols () -// -// Description: -// Find all of the defined symbols in the specified module. -// -// Inputs: -// M - The module in which to find defined symbols. -// -// Outputs: -// DefinedSymbols - A set of C++ strings that will contain the name of all -// defined symbols. -// -// Return value: -// None. -// -void -GetAllDefinedSymbols(Module *M, std::set &DefinedSymbols) -{ +/// GetAllDefinedSymbols - Modifies its parameter DefinedSymbols to contain the +/// name of each externally-visible symbol defined in M. +/// +static void +GetAllDefinedSymbols(Module *M, std::set &DefinedSymbols) { for (Module::iterator I = M->begin(), E = M->end(); I != E; ++I) if (I->hasName() && !I->isExternal() && !I->hasInternalLinkage()) DefinedSymbols.insert(I->getName()); @@ -153,30 +38,22 @@ GetAllDefinedSymbols(Module *M, std::set &DefinedSymbols) DefinedSymbols.insert(I->getName()); } -// -// Function: GetAllUndefinedSymbols () -// -// Description: -// This calculates the set of undefined symbols that still exist in an LLVM -// module. This is a bit tricky because there may be two symbols with the -// same name but different LLVM types that will be resolved to each other but -// aren't currently (thus we need to treat it as resolved). -// -// Inputs: -// M - The module in which to find undefined symbols. -// -// Outputs: -// UndefinedSymbols - A set of C++ strings containing the name of all -// undefined symbols. -// -// Return value: -// None. -// -void -GetAllUndefinedSymbols(Module *M, std::set &UndefinedSymbols) -{ +/// GetAllUndefinedSymbols - calculates the set of undefined symbols that still +/// exist in an LLVM module. This is a bit tricky because there may be two +/// symbols with the same name but different LLVM types that will be resolved to +/// each other but aren't currently (thus we need to treat it as resolved). +/// +/// Inputs: +/// M - The module in which to find undefined symbols. +/// +/// Outputs: +/// UndefinedSymbols - A set of C++ strings containing the name of all +/// undefined symbols. +/// +static void +GetAllUndefinedSymbols(Module *M, std::set &UndefinedSymbols) { std::set DefinedSymbols; - UndefinedSymbols.clear(); // Start out empty + UndefinedSymbols.clear(); for (Module::iterator I = M->begin(), E = M->end(); I != E; ++I) if (I->hasName()) { @@ -202,309 +79,102 @@ GetAllUndefinedSymbols(Module *M, std::set &UndefinedSymbols) ++I; // Keep this symbol in the undefined symbols list } +/// LinkInArchive - opens an archive library and link in all objects which +/// provide symbols that are currently undefined. +/// +/// Inputs: +/// Filename - The pathname of the archive. +/// +/// Return Value: +/// TRUE - An error occurred. +/// FALSE - No errors. +bool +Linker::LinkInArchive(const sys::Path &Filename) { + + // Make sure this is an archive file we're dealing with + if (!Filename.isArchive()) + return error("File '" + Filename.toString() + "' is not an archive."); + + // Open the archive file + verbose("Linking archive file '" + Filename.toString() + "'"); -// -// Function: LoadObject () -// -// Description: -// Read the specified bytecode object file. -// -// Inputs: -// FN - The name of the file to load. -// -// Outputs: -// OutErrorMessage - The error message to give back to the caller. -// -// Return Value: -// A pointer to a module represening the bytecode file is returned. -// If an error occurs, the pointer is 0. -// -std::auto_ptr -LoadObject(const std::string & FN, std::string &OutErrorMessage) { - std::string ErrorMessage; - Module *Result = ParseBytecodeFile(FN, &ErrorMessage); - if (Result) return std::auto_ptr(Result); - OutErrorMessage = "Bytecode file '" + FN + "' corrupt!"; - if (ErrorMessage.size()) OutErrorMessage += ": " + ErrorMessage; - return std::auto_ptr(); -} - -// -// Function: LinkInArchive () -// -// Description: -// This function will open an archive library and link in all objects which -// provide symbols that are currently undefined. -// -// Inputs: -// M - The module in which to link the archives. -// Filename - The pathname of the archive. -// Verbose - Flags whether verbose messages should be printed. -// -// Outputs: -// ErrorMessage - A C++ string detailing what error occurred, if any. -// -// Return Value: -// TRUE - An error occurred. -// FALSE - No errors. -// -static bool LinkInArchive(Module *M, - const std::string &Filename, - std::string &ErrorMessage, - bool Verbose) -{ - // // Find all of the symbols currently undefined in the bytecode program. // If all the symbols are defined, the program is complete, and there is // no reason to link in any archive files. - // std::set UndefinedSymbols; - GetAllUndefinedSymbols(M, UndefinedSymbols); + GetAllUndefinedSymbols(Composite, UndefinedSymbols); + if (UndefinedSymbols.empty()) { - if (Verbose) std::cerr << " No symbols undefined, don't link library!\n"; + verbose("No symbols undefined, skipping library '" + + Filename.toString() + "'"); return false; // No need to link anything in! } - // - // Load in the archive objects. - // - if (Verbose) std::cerr << " Loading '" << Filename << "'\n"; - std::vector Objects; - if (ReadArchiveFile(Filename, Objects, &ErrorMessage)) - return true; - - // - // Figure out which symbols are defined by all of the modules in the archive. - // - std::vector > DefinedSymbols; - DefinedSymbols.resize(Objects.size()); - for (unsigned i = 0; i != Objects.size(); ++i) { - GetAllDefinedSymbols(Objects[i], DefinedSymbols[i]); - } - - // While we are linking in object files, loop. - bool Linked = true; - while (Linked) { - Linked = false; + std::string ErrMsg; + std::auto_ptr AutoArch ( + Archive::OpenAndLoadSymbols(Filename,&ErrMsg)); - for (unsigned i = 0; i != Objects.size(); ++i) { - // Consider whether we need to link in this module... we only need to - // link it in if it defines some symbol which is so far undefined. - // - const std::set &DefSymbols = DefinedSymbols[i]; + Archive* arch = AutoArch.get(); - bool ObjectRequired = false; - for (std::set::iterator I = UndefinedSymbols.begin(), - E = UndefinedSymbols.end(); I != E; ++I) - if (DefSymbols.count(*I)) { - if (Verbose) - std::cerr << " Found object providing symbol '" << *I << "'...\n"; - ObjectRequired = true; - break; - } - - // We DO need to link this object into the program... - if (ObjectRequired) { - if (LinkModules(M, Objects[i], &ErrorMessage)) - return true; // Couldn't link in the right object file... - - // Since we have linked in this object, delete it from the list of - // objects to consider in this archive file. - std::swap(Objects[i], Objects.back()); - std::swap(DefinedSymbols[i], DefinedSymbols.back()); - Objects.pop_back(); - DefinedSymbols.pop_back(); - --i; // Do not skip an entry - - // The undefined symbols set should have shrunk. - GetAllUndefinedSymbols(M, UndefinedSymbols); - Linked = true; // We have linked something in! - } - } - } - - return false; -} + if (!arch) + return error("Cannot read archive '" + Filename.toString() + + "': " + ErrMsg); -// -// Function: LinkInFile () -// -// Description: -// This function will open an archive library and link in all objects which -// provide symbols that are currently undefined. -// -// Inputs: -// HeadModule - The module in which to link the archives. -// Filename - The pathname of the archive. -// Verbose - Flags whether verbose messages should be printed. -// -// Outputs: -// ErrorMessage - A C++ string detailing what error occurred, if any. -// -// Return Value: -// TRUE - An error occurred. -// FALSE - No errors. -// -static bool LinkInFile(Module *HeadModule, - const std::string &Filename, - std::string &ErrorMessage, - bool Verbose) -{ - std::auto_ptr M(LoadObject(Filename, ErrorMessage)); - if (M.get() == 0) return true; - if (Verbose) std::cerr << "Linking in '" << Filename << "'\n"; - return LinkModules(HeadModule, M.get(), &ErrorMessage); -} + // Save a set of symbols that are not defined by the archive. Since we're + // entering a loop, there's no point searching for these multiple times. This + // variable is used to "set_subtract" from the set of undefined symbols. + std::set NotDefinedByArchive; -// -// Function: LinkFiles () -// -// Description: -// This function takes a module and a list of files and links them all -// together. It locates the file either in the current directory, as it's -// absolute or relative pathname, or as a file somewhere in -// LLVM_LIB_SEARCH_PATH. -// -// Inputs: -// progname - The name of the program (infamous argv[0]). -// HeadModule - The module under which all files will be linked. -// Files - A vector of C++ strings indicating the LLVM bytecode filenames -// to be linked. The names can refer to a mixture of pure LLVM -// bytecode files and archive (ar) formatted files. -// Verbose - Flags whether verbose output should be printed while linking. -// -// Outputs: -// HeadModule - The module will have the specified LLVM bytecode files linked -// in. -// -// Return value: -// FALSE - No errors. -// TRUE - Some error occurred. -// -bool LinkFiles(const char *progname, - Module *HeadModule, - const std::vector &Files, - bool Verbose) -{ - // String in which to receive error messages. - std::string ErrorMessage; + // While we are linking in object files, loop. + while (true) { - // Full pathname of the file - std::string Pathname; + // Find the modules we need to link into the target module + std::set Modules; + arch->findModulesDefiningSymbols(UndefinedSymbols, Modules); - // Get the library search path from the environment - char *SearchPath = getenv("LLVM_LIB_SEARCH_PATH"); + // If we didn't find any more modules to link this time, we are done + // searching this archive. + if (Modules.empty()) + break; - for (unsigned i = 1; i < Files.size(); ++i) { - // Determine where this file lives. - if (FileExists(Files[i])) { - Pathname = Files[i]; - } else { - if (SearchPath == NULL) { - std::cerr << "Cannot find " << Files[i]; - return true; - } + // Any symbols remaining in UndefinedSymbols after + // findModulesDefiningSymbols are ones that the archive does not define. So + // we add them to the NotDefinedByArchive variable now. + NotDefinedByArchive.insert(UndefinedSymbols.begin(), + UndefinedSymbols.end()); - Pathname = std::string(SearchPath)+"/"+Files[i]; - if (!FileExists(Pathname)) { - std::cerr << "Cannot find " << Files[i]; - return true; - } - } + // Loop over all the ModuleProviders that we got back from the archive + for (std::set::iterator I=Modules.begin(), E=Modules.end(); + I != E; ++I) { - // A user may specify an ar archive without -l, perhaps because it - // is not installed as a library. Detect that and link the library. - if (IsArchive(Pathname)) { - if (Verbose) - std::cerr << "Linking archive '" << Files[i] << "'\n"; + // Get the module we must link in. + std::auto_ptr AutoModule( (*I)->releaseModule() ); + Module* aModule = AutoModule.get(); - if (LinkInArchive(HeadModule, Pathname, ErrorMessage, Verbose)) { - PrintAndReturn(progname, ErrorMessage, - ": Error linking in '" + Files[i] + "'"); - return true; - } - } else { - if (Verbose) - std::cerr << "Linking file '" << Files[i] << "'\n"; + verbose(" Linking in module: " + aModule->getModuleIdentifier()); - if (LinkInFile(HeadModule, Pathname, ErrorMessage, Verbose)) { - PrintAndReturn(progname, ErrorMessage, - ": error linking in '" + Files[i] + "'"); - return true; - } + // Link it in + if (LinkInModule(aModule)) + return error("Cannot link in module '" + + aModule->getModuleIdentifier() + "': " + Error); } - } - - return false; -} -// -// Function: LinkLibraries () -// -// Description: -// This function takes the specified library files and links them into the -// main bytecode object file. -// -// Inputs: -// progname - The name of the program (infamous argv[0]). -// HeadModule - The module into which all necessary libraries will be linked. -// Libraries - The list of libraries to link into the module. -// LibPaths - The list of library paths in which to find libraries. -// Verbose - Flags whether verbose messages should be printed. -// Native - Flags whether native code is being generated. -// -// Outputs: -// HeadModule - The module will have all necessary libraries linked in. -// -// Return value: -// FALSE - No error. -// TRUE - Error. -// -bool LinkLibraries(const char *progname, - Module *HeadModule, - const std::vector &Libraries, - const std::vector &LibPaths, - bool Verbose, - bool Native) -{ - // String in which to receive error messages. - std::string ErrorMessage; - - for (unsigned i = 1; i < Libraries.size(); ++i) { - // Determine where this library lives. - std::string Pathname = FindLib(Libraries[i], LibPaths); - if (Pathname.empty()) { - // If the pathname does not exist, then continue to the next one if - // we're doing a native link and give an error if we're doing a bytecode - // link. - if (!Native) { - PrintAndReturn(progname, "Cannot find " + Libraries[i]); - return true; - } - } - - // A user may specify an ar archive without -l, perhaps because it - // is not installed as a library. Detect that and link the library. - if (IsArchive(Pathname)) { - if (Verbose) - std::cerr << "Linking archive '" << Libraries[i] << "'\n"; - - if (LinkInArchive(HeadModule, Pathname, ErrorMessage, Verbose)) { - PrintAndReturn(progname, ErrorMessage, - ": Error linking in '" + Libraries[i] + "'"); - return true; - } - } else { - if (Verbose) - std::cerr << "Linking file '" << Libraries[i] << "'\n"; - - if (LinkInFile(HeadModule, Pathname, ErrorMessage, Verbose)) { - PrintAndReturn(progname, ErrorMessage, - ": error linking in '" + Libraries[i] + "'"); - return true; - } - } + // Get the undefined symbols from the aggregate module. This recomputes the + // symbols we still need after the new modules have been linked in. + GetAllUndefinedSymbols(Composite, UndefinedSymbols); + + // At this point we have two sets of undefined symbols: UndefinedSymbols + // which holds the undefined symbols from all the modules, and + // NotDefinedByArchive which holds symbols we know the archive doesn't + // define. There's no point searching for symbols that we won't find in the + // archive so we subtract these sets. + set_subtract(UndefinedSymbols, NotDefinedByArchive); + + // If there's no symbols left, no point in continuing to search the + // archive. + if (UndefinedSymbols.empty()) + break; } - + return false; }