xref: /llvm-project/clang-tools-extra/clangd/CompileCommands.cpp (revision dd647e3e608ed0b2bac7c588d5859b80ef4a5976)
199768b24SSam McCall //===--- CompileCommands.cpp ----------------------------------------------===//
299768b24SSam McCall //
399768b24SSam McCall // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
499768b24SSam McCall // See https://llvm.org/LICENSE.txt for license information.
599768b24SSam McCall // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
699768b24SSam McCall //
799768b24SSam McCall //===----------------------------------------------------------------------===//
899768b24SSam McCall 
999768b24SSam McCall #include "CompileCommands.h"
10c5a6ee16SSam McCall #include "Config.h"
11ad97ccf6SSam McCall #include "support/Logger.h"
12ab714ba0SKadir Cetinkaya #include "support/Trace.h"
13ab714ba0SKadir Cetinkaya #include "clang/Driver/Driver.h"
148eb8c92eSSam McCall #include "clang/Driver/Options.h"
1599768b24SSam McCall #include "clang/Frontend/CompilerInvocation.h"
164d006520SSam McCall #include "clang/Tooling/CompilationDatabase.h"
17d60d3455SDmitry Polukhin #include "clang/Tooling/Tooling.h"
18ab714ba0SKadir Cetinkaya #include "llvm/ADT/ArrayRef.h"
190a3c7960SKadir Cetinkaya #include "llvm/ADT/STLExtras.h"
20ab714ba0SKadir Cetinkaya #include "llvm/ADT/SmallVector.h"
210a3c7960SKadir Cetinkaya #include "llvm/ADT/StringRef.h"
22ab714ba0SKadir Cetinkaya #include "llvm/Option/ArgList.h"
238eb8c92eSSam McCall #include "llvm/Option/Option.h"
248eb8c92eSSam McCall #include "llvm/Support/Allocator.h"
258eb8c92eSSam McCall #include "llvm/Support/Debug.h"
2699768b24SSam McCall #include "llvm/Support/FileSystem.h"
2799768b24SSam McCall #include "llvm/Support/FileUtilities.h"
2899768b24SSam McCall #include "llvm/Support/MemoryBuffer.h"
2999768b24SSam McCall #include "llvm/Support/Path.h"
3099768b24SSam McCall #include "llvm/Support/Program.h"
310a3c7960SKadir Cetinkaya #include <iterator>
32f885ae23SMatt Arsenault #include <optional>
33b6626515SKadir Cetinkaya #include <string>
34b6626515SKadir Cetinkaya #include <vector>
3599768b24SSam McCall 
3699768b24SSam McCall namespace clang {
3799768b24SSam McCall namespace clangd {
3899768b24SSam McCall namespace {
3999768b24SSam McCall 
4099768b24SSam McCall // Query apple's `xcrun` launcher, which is the source of truth for "how should"
4199768b24SSam McCall // clang be invoked on this system.
42f71ffd3bSKazu Hirata std::optional<std::string> queryXcrun(llvm::ArrayRef<llvm::StringRef> Argv) {
4399768b24SSam McCall   auto Xcrun = llvm::sys::findProgramByName("xcrun");
4499768b24SSam McCall   if (!Xcrun) {
4599768b24SSam McCall     log("Couldn't find xcrun. Hopefully you have a non-apple toolchain...");
46059a23c0SKazu Hirata     return std::nullopt;
4799768b24SSam McCall   }
4899768b24SSam McCall   llvm::SmallString<64> OutFile;
4999768b24SSam McCall   llvm::sys::fs::createTemporaryFile("clangd-xcrun", "", OutFile);
5099768b24SSam McCall   llvm::FileRemover OutRemover(OutFile);
51f4fb2b30SFangrui Song   std::optional<llvm::StringRef> Redirects[3] = {
529878e176SAaron Ballman       /*stdin=*/{""}, /*stdout=*/{OutFile.str()}, /*stderr=*/{""}};
5399768b24SSam McCall   vlog("Invoking {0} to find clang installation", *Xcrun);
5499768b24SSam McCall   int Ret = llvm::sys::ExecuteAndWait(*Xcrun, Argv,
55f4fb2b30SFangrui Song                                       /*Env=*/std::nullopt, Redirects,
5699768b24SSam McCall                                       /*SecondsToWait=*/10);
5799768b24SSam McCall   if (Ret != 0) {
5899768b24SSam McCall     log("xcrun exists but failed with code {0}. "
5999768b24SSam McCall         "If you have a non-apple toolchain, this is OK. "
6099768b24SSam McCall         "Otherwise, try xcode-select --install.",
6199768b24SSam McCall         Ret);
62059a23c0SKazu Hirata     return std::nullopt;
6399768b24SSam McCall   }
6499768b24SSam McCall 
6599768b24SSam McCall   auto Buf = llvm::MemoryBuffer::getFile(OutFile);
6699768b24SSam McCall   if (!Buf) {
6799768b24SSam McCall     log("Can't read xcrun output: {0}", Buf.getError().message());
68059a23c0SKazu Hirata     return std::nullopt;
6999768b24SSam McCall   }
7099768b24SSam McCall   StringRef Path = Buf->get()->getBuffer().trim();
7199768b24SSam McCall   if (Path.empty()) {
7299768b24SSam McCall     log("xcrun produced no output");
73059a23c0SKazu Hirata     return std::nullopt;
7499768b24SSam McCall   }
7599768b24SSam McCall   return Path.str();
7699768b24SSam McCall }
7799768b24SSam McCall 
7899768b24SSam McCall // Resolve symlinks if possible.
7999768b24SSam McCall std::string resolve(std::string Path) {
8099768b24SSam McCall   llvm::SmallString<128> Resolved;
8199768b24SSam McCall   if (llvm::sys::fs::real_path(Path, Resolved)) {
8299768b24SSam McCall     log("Failed to resolve possible symlink {0}", Path);
8399768b24SSam McCall     return Path;
8499768b24SSam McCall   }
855b2772e1SKazu Hirata   return std::string(Resolved);
8699768b24SSam McCall }
8799768b24SSam McCall 
8899768b24SSam McCall // Get a plausible full `clang` path.
8999768b24SSam McCall // This is used in the fallback compile command, or when the CDB returns a
9099768b24SSam McCall // generic driver with no path.
9199768b24SSam McCall std::string detectClangPath() {
9299768b24SSam McCall   // The driver and/or cc1 sometimes depend on the binary name to compute
9399768b24SSam McCall   // useful things like the standard library location.
9499768b24SSam McCall   // We need to emulate what clang on this system is likely to see.
9599768b24SSam McCall   // cc1 in particular looks at the "real path" of the running process, and
9699768b24SSam McCall   // so if /usr/bin/clang is a symlink, it sees the resolved path.
9799768b24SSam McCall   // clangd doesn't have that luxury, so we resolve symlinks ourselves.
9899768b24SSam McCall 
9999768b24SSam McCall   // On Mac, `which clang` is /usr/bin/clang. It runs `xcrun clang`, which knows
10099768b24SSam McCall   // where the real clang is kept. We need to do the same thing,
10199768b24SSam McCall   // because cc1 (not the driver!) will find libc++ relative to argv[0].
10299768b24SSam McCall #ifdef __APPLE__
10399768b24SSam McCall   if (auto MacClang = queryXcrun({"xcrun", "--find", "clang"}))
10499768b24SSam McCall     return resolve(std::move(*MacClang));
10599768b24SSam McCall #endif
10699768b24SSam McCall   // On other platforms, just look for compilers on the PATH.
10799768b24SSam McCall   for (const char *Name : {"clang", "gcc", "cc"})
10899768b24SSam McCall     if (auto PathCC = llvm::sys::findProgramByName(Name))
10999768b24SSam McCall       return resolve(std::move(*PathCC));
11099768b24SSam McCall   // Fallback: a nonexistent 'clang' binary next to clangd.
111f71404c3SKadir Cetinkaya   static int StaticForMainAddr;
11299768b24SSam McCall   std::string ClangdExecutable =
113f71404c3SKadir Cetinkaya       llvm::sys::fs::getMainExecutable("clangd", (void *)&StaticForMainAddr);
11499768b24SSam McCall   SmallString<128> ClangPath;
11599768b24SSam McCall   ClangPath = llvm::sys::path::parent_path(ClangdExecutable);
11699768b24SSam McCall   llvm::sys::path::append(ClangPath, "clang");
1175b2772e1SKazu Hirata   return std::string(ClangPath);
11899768b24SSam McCall }
11999768b24SSam McCall 
12099768b24SSam McCall // On mac, /usr/bin/clang sets SDKROOT and then invokes the real clang.
12199768b24SSam McCall // The effect of this is to set -isysroot correctly. We do the same.
122f71ffd3bSKazu Hirata std::optional<std::string> detectSysroot() {
12399768b24SSam McCall #ifndef __APPLE__
124059a23c0SKazu Hirata   return std::nullopt;
12599768b24SSam McCall #endif
12699768b24SSam McCall 
12799768b24SSam McCall   // SDKROOT overridden in environment, respect it. Driver will set isysroot.
12899768b24SSam McCall   if (::getenv("SDKROOT"))
129059a23c0SKazu Hirata     return std::nullopt;
13099768b24SSam McCall   return queryXcrun({"xcrun", "--show-sdk-path"});
13199768b24SSam McCall }
13299768b24SSam McCall 
13399768b24SSam McCall std::string detectStandardResourceDir() {
134f71404c3SKadir Cetinkaya   static int StaticForMainAddr; // Just an address in this process.
135f71404c3SKadir Cetinkaya   return CompilerInvocation::GetResourcesPath("clangd",
136f71404c3SKadir Cetinkaya                                               (void *)&StaticForMainAddr);
13799768b24SSam McCall }
13899768b24SSam McCall 
1392a3ac01bSSam McCall // The path passed to argv[0] is important:
1402a3ac01bSSam McCall //  - its parent directory is Driver::Dir, used for library discovery
1412a3ac01bSSam McCall //  - its basename affects CLI parsing (clang-cl) and other settings
1422a3ac01bSSam McCall // Where possible it should be an absolute path with sensible directory, but
1432a3ac01bSSam McCall // with the original basename.
1442a3ac01bSSam McCall static std::string resolveDriver(llvm::StringRef Driver, bool FollowSymlink,
145f71ffd3bSKazu Hirata                                  std::optional<std::string> ClangPath) {
1462a3ac01bSSam McCall   auto SiblingOf = [&](llvm::StringRef AbsPath) {
1472a3ac01bSSam McCall     llvm::SmallString<128> Result = llvm::sys::path::parent_path(AbsPath);
1482a3ac01bSSam McCall     llvm::sys::path::append(Result, llvm::sys::path::filename(Driver));
1492a3ac01bSSam McCall     return Result.str().str();
1502a3ac01bSSam McCall   };
1512a3ac01bSSam McCall 
1522a3ac01bSSam McCall   // First, eliminate relative paths.
1532a3ac01bSSam McCall   std::string Storage;
1542a3ac01bSSam McCall   if (!llvm::sys::path::is_absolute(Driver)) {
155a3b79340SSam McCall     // If it's working-dir relative like bin/clang, we can't resolve it.
156a3b79340SSam McCall     // FIXME: we could if we had the working directory here.
157a3b79340SSam McCall     // Let's hope it's not a symlink.
158a3b79340SSam McCall     if (llvm::any_of(Driver,
159a3b79340SSam McCall                      [](char C) { return llvm::sys::path::is_separator(C); }))
160a3b79340SSam McCall       return Driver.str();
1612a3ac01bSSam McCall     // If the driver is a generic like "g++" with no path, add clang dir.
1622a3ac01bSSam McCall     if (ClangPath &&
1632a3ac01bSSam McCall         (Driver == "clang" || Driver == "clang++" || Driver == "gcc" ||
1642a3ac01bSSam McCall          Driver == "g++" || Driver == "cc" || Driver == "c++")) {
1652a3ac01bSSam McCall       return SiblingOf(*ClangPath);
1662a3ac01bSSam McCall     }
1672a3ac01bSSam McCall     // Otherwise try to look it up on PATH. This won't change basename.
1682a3ac01bSSam McCall     auto Absolute = llvm::sys::findProgramByName(Driver);
1692a3ac01bSSam McCall     if (Absolute && llvm::sys::path::is_absolute(*Absolute))
1702a3ac01bSSam McCall       Driver = Storage = std::move(*Absolute);
1712a3ac01bSSam McCall     else if (ClangPath) // If we don't find it, use clang dir again.
1722a3ac01bSSam McCall       return SiblingOf(*ClangPath);
1732a3ac01bSSam McCall     else // Nothing to do: can't find the command and no detected dir.
1742a3ac01bSSam McCall       return Driver.str();
1752a3ac01bSSam McCall   }
1762a3ac01bSSam McCall 
1772a3ac01bSSam McCall   // Now we have an absolute path, but it may be a symlink.
1782a3ac01bSSam McCall   assert(llvm::sys::path::is_absolute(Driver));
1792a3ac01bSSam McCall   if (FollowSymlink) {
1802a3ac01bSSam McCall     llvm::SmallString<256> Resolved;
1812a3ac01bSSam McCall     if (!llvm::sys::fs::real_path(Driver, Resolved))
1822a3ac01bSSam McCall       return SiblingOf(Resolved);
1832a3ac01bSSam McCall   }
1842a3ac01bSSam McCall   return Driver.str();
1852a3ac01bSSam McCall }
1862a3ac01bSSam McCall 
18799768b24SSam McCall } // namespace
18899768b24SSam McCall 
18999768b24SSam McCall CommandMangler CommandMangler::detect() {
19099768b24SSam McCall   CommandMangler Result;
19199768b24SSam McCall   Result.ClangPath = detectClangPath();
19299768b24SSam McCall   Result.ResourceDir = detectStandardResourceDir();
19399768b24SSam McCall   Result.Sysroot = detectSysroot();
19499768b24SSam McCall   return Result;
19599768b24SSam McCall }
19699768b24SSam McCall 
197ab714ba0SKadir Cetinkaya CommandMangler CommandMangler::forTests() { return CommandMangler(); }
19899768b24SSam McCall 
199afa22c56SNathan Ridge void CommandMangler::operator()(tooling::CompileCommand &Command,
200259e365dSKadir Cetinkaya                                 llvm::StringRef File) const {
201afa22c56SNathan Ridge   std::vector<std::string> &Cmd = Command.CommandLine;
202ab714ba0SKadir Cetinkaya   trace::Span S("AdjustCompileFlags");
203444a5f30SKadir Cetinkaya   // Most of the modifications below assumes the Cmd starts with a driver name.
204444a5f30SKadir Cetinkaya   // We might consider injecting a generic driver name like "cc" or "c++", but
205d60d3455SDmitry Polukhin   // a Cmd missing the driver is probably rare enough in practice and erroneous.
206444a5f30SKadir Cetinkaya   if (Cmd.empty())
207444a5f30SKadir Cetinkaya     return;
208d60d3455SDmitry Polukhin 
209ab714ba0SKadir Cetinkaya   auto &OptTable = clang::driver::getDriverOptTable();
210ab714ba0SKadir Cetinkaya   // OriginalArgs needs to outlive ArgList.
211ab714ba0SKadir Cetinkaya   llvm::SmallVector<const char *, 16> OriginalArgs;
212ab714ba0SKadir Cetinkaya   OriginalArgs.reserve(Cmd.size());
213ab714ba0SKadir Cetinkaya   for (const auto &S : Cmd)
214ab714ba0SKadir Cetinkaya     OriginalArgs.push_back(S.c_str());
215444a5f30SKadir Cetinkaya   bool IsCLMode = driver::IsClangCL(driver::getDriverMode(
216984b800aSserge-sans-paille       OriginalArgs[0], llvm::ArrayRef(OriginalArgs).slice(1)));
217d60d3455SDmitry Polukhin   // ParseArgs propagates missing arg/opt counts on error, but preserves
218ab714ba0SKadir Cetinkaya   // everything it could parse in ArgList. So we just ignore those counts.
219ab714ba0SKadir Cetinkaya   unsigned IgnoredCount;
220259e365dSKadir Cetinkaya   // Drop the executable name, as ParseArgs doesn't expect it. This means
221259e365dSKadir Cetinkaya   // indices are actually of by one between ArgList and OriginalArgs.
222444a5f30SKadir Cetinkaya   llvm::opt::InputArgList ArgList;
223444a5f30SKadir Cetinkaya   ArgList = OptTable.ParseArgs(
224984b800aSserge-sans-paille       llvm::ArrayRef(OriginalArgs).drop_front(), IgnoredCount, IgnoredCount,
2259478f661SJustin Bogner       llvm::opt::Visibility(IsCLMode ? driver::options::CLOption
2269478f661SJustin Bogner                                      : driver::options::ClangOption));
227ab714ba0SKadir Cetinkaya 
2283bf77980SKadir Cetinkaya   llvm::SmallVector<unsigned, 1> IndicesToDrop;
2293bf77980SKadir Cetinkaya   // Having multiple architecture options (e.g. when building fat binaries)
2303bf77980SKadir Cetinkaya   // results in multiple compiler jobs, which clangd cannot handle. In such
2313bf77980SKadir Cetinkaya   // cases strip all the `-arch` options and fallback to default architecture.
2323bf77980SKadir Cetinkaya   // As there are no signals to figure out which one user actually wants. They
2333bf77980SKadir Cetinkaya   // can explicitly specify one through `CompileFlags.Add` if need be.
2343bf77980SKadir Cetinkaya   unsigned ArchOptCount = 0;
2353bf77980SKadir Cetinkaya   for (auto *Input : ArgList.filtered(driver::options::OPT_arch)) {
2363bf77980SKadir Cetinkaya     ++ArchOptCount;
2373bf77980SKadir Cetinkaya     for (auto I = 0U; I <= Input->getNumValues(); ++I)
2383bf77980SKadir Cetinkaya       IndicesToDrop.push_back(Input->getIndex() + I);
2393bf77980SKadir Cetinkaya   }
2403bf77980SKadir Cetinkaya   // If there is a single `-arch` option, keep it.
2413bf77980SKadir Cetinkaya   if (ArchOptCount < 2)
2423bf77980SKadir Cetinkaya     IndicesToDrop.clear();
24379c2616dSKadir Cetinkaya 
2440683a1e5SSam McCall   // In some cases people may try to reuse the command from another file, e.g.
2450683a1e5SSam McCall   //   { File: "foo.h", CommandLine: "clang foo.cpp" }.
2460683a1e5SSam McCall   // We assume the intent is to parse foo.h the same way as foo.cpp, or as if
2470683a1e5SSam McCall   // it were being included from foo.cpp.
2480683a1e5SSam McCall   //
2490683a1e5SSam McCall   // We're going to rewrite the command to refer to foo.h, and this may change
2500683a1e5SSam McCall   // its semantics (e.g. by parsing the file as C). If we do this, we should
2510683a1e5SSam McCall   // use transferCompileCommand to adjust the argv.
2520683a1e5SSam McCall   // In practice only the extension of the file matters, so do this only when
2530683a1e5SSam McCall   // it differs.
2540683a1e5SSam McCall   llvm::StringRef FileExtension = llvm::sys::path::extension(File);
255f71ffd3bSKazu Hirata   std::optional<std::string> TransferFrom;
2560683a1e5SSam McCall   auto SawInput = [&](llvm::StringRef Input) {
2570683a1e5SSam McCall     if (llvm::sys::path::extension(Input) != FileExtension)
2580683a1e5SSam McCall       TransferFrom.emplace(Input);
2590683a1e5SSam McCall   };
2600683a1e5SSam McCall 
26179c2616dSKadir Cetinkaya   // Strip all the inputs and `--`. We'll put the input for the requested file
26279c2616dSKadir Cetinkaya   // explicitly at the end of the flags. This ensures modifications done in the
26379c2616dSKadir Cetinkaya   // following steps apply in more cases (like setting -x, which only affects
26479c2616dSKadir Cetinkaya   // inputs that come after it).
2650683a1e5SSam McCall   for (auto *Input : ArgList.filtered(driver::options::OPT_INPUT)) {
2660683a1e5SSam McCall     SawInput(Input->getValue(0));
267259e365dSKadir Cetinkaya     IndicesToDrop.push_back(Input->getIndex());
2680683a1e5SSam McCall   }
26979c2616dSKadir Cetinkaya   // Anything after `--` is also treated as input, drop them as well.
27079c2616dSKadir Cetinkaya   if (auto *DashDash =
27179c2616dSKadir Cetinkaya           ArgList.getLastArgNoClaim(driver::options::OPT__DASH_DASH)) {
2720683a1e5SSam McCall     auto DashDashIndex = DashDash->getIndex() + 1; // +1 accounts for Cmd[0]
2730683a1e5SSam McCall     for (unsigned I = DashDashIndex; I < Cmd.size(); ++I)
2740683a1e5SSam McCall       SawInput(Cmd[I]);
2750683a1e5SSam McCall     Cmd.resize(DashDashIndex);
2763bf77980SKadir Cetinkaya   }
277259e365dSKadir Cetinkaya   llvm::sort(IndicesToDrop);
278840a9682SKazu Hirata   for (unsigned Idx : llvm::reverse(IndicesToDrop))
279259e365dSKadir Cetinkaya     // +1 to account for the executable name in Cmd[0] that
280259e365dSKadir Cetinkaya     // doesn't exist in ArgList.
281840a9682SKazu Hirata     Cmd.erase(Cmd.begin() + Idx + 1);
28279c2616dSKadir Cetinkaya   // All the inputs are stripped, append the name for the requested file. Rest
28379c2616dSKadir Cetinkaya   // of the modifications should respect `--`.
28479c2616dSKadir Cetinkaya   Cmd.push_back("--");
28579c2616dSKadir Cetinkaya   Cmd.push_back(File.str());
286ab714ba0SKadir Cetinkaya 
2870683a1e5SSam McCall   if (TransferFrom) {
2880683a1e5SSam McCall     tooling::CompileCommand TransferCmd;
2890683a1e5SSam McCall     TransferCmd.Filename = std::move(*TransferFrom);
2900683a1e5SSam McCall     TransferCmd.CommandLine = std::move(Cmd);
2910683a1e5SSam McCall     TransferCmd = transferCompileCommand(std::move(TransferCmd), File);
2920683a1e5SSam McCall     Cmd = std::move(TransferCmd.CommandLine);
2934258d68dSSam McCall     assert(Cmd.size() >= 2 && Cmd.back() == File &&
2944258d68dSSam McCall            Cmd[Cmd.size() - 2] == "--" &&
2954258d68dSSam McCall            "TransferCommand should produce a command ending in -- filename");
2960683a1e5SSam McCall   }
2970683a1e5SSam McCall 
2989d3e9a3eSSam McCall   for (auto &Edit : Config::current().CompileFlags.Edits)
299c5a6ee16SSam McCall     Edit(Cmd);
300c5a6ee16SSam McCall 
30168e230aaSNathan Ridge   // The system include extractor needs to run:
30268e230aaSNathan Ridge   //  - AFTER transferCompileCommand(), because the -x flag it adds may be
30368e230aaSNathan Ridge   //    necessary for the system include extractor to identify the file type
30468e230aaSNathan Ridge   //  - AFTER applying CompileFlags.Edits, because the name of the compiler
30568e230aaSNathan Ridge   //    that needs to be invoked may come from the CompileFlags->Compiler key
306d60d3455SDmitry Polukhin   //  - BEFORE addTargetAndModeForProgramName(), because gcc doesn't support
307d60d3455SDmitry Polukhin   //    the target flag that might be added.
30868e230aaSNathan Ridge   //  - BEFORE resolveDriver() because that can mess up the driver path,
30968e230aaSNathan Ridge   //    e.g. changing gcc to /path/to/clang/bin/gcc
31068e230aaSNathan Ridge   if (SystemIncludeExtractor) {
31168e230aaSNathan Ridge     SystemIncludeExtractor(Command, File);
31268e230aaSNathan Ridge   }
31368e230aaSNathan Ridge 
314d60d3455SDmitry Polukhin   tooling::addTargetAndModeForProgramName(Cmd, Cmd.front());
315d60d3455SDmitry Polukhin 
316f489fb3dSKon   // Check whether the flag exists in the command.
317f489fb3dSKon   auto HasExact = [&](llvm::StringRef Flag) {
318f489fb3dSKon     return llvm::any_of(Cmd, [&](llvm::StringRef Arg) { return Arg == Flag; });
319f489fb3dSKon   };
320f489fb3dSKon 
321f489fb3dSKon   // Check whether the flag appears in the command as a prefix.
322f489fb3dSKon   auto HasPrefix = [&](llvm::StringRef Flag) {
323f489fb3dSKon     return llvm::any_of(
324dabc9018SJie Fu         Cmd, [&](llvm::StringRef Arg) { return Arg.starts_with(Flag); });
32599768b24SSam McCall   };
32699768b24SSam McCall 
3270a3c7960SKadir Cetinkaya   llvm::erase_if(Cmd, [](llvm::StringRef Elem) {
328d5953e3eSKazu Hirata     return Elem.starts_with("--save-temps") || Elem.starts_with("-save-temps");
3290a3c7960SKadir Cetinkaya   });
33099768b24SSam McCall 
331b6626515SKadir Cetinkaya   std::vector<std::string> ToAppend;
332f489fb3dSKon   if (ResourceDir && !HasExact("-resource-dir") && !HasPrefix("-resource-dir="))
333b6626515SKadir Cetinkaya     ToAppend.push_back(("-resource-dir=" + *ResourceDir));
33499768b24SSam McCall 
335ea9888b8SDavid Goldman   // Don't set `-isysroot` if it is already set or if `--sysroot` is set.
336ea9888b8SDavid Goldman   // `--sysroot` is a superset of the `-isysroot` argument.
337f489fb3dSKon   if (Sysroot && !HasPrefix("-isysroot") && !HasExact("--sysroot") &&
338f489fb3dSKon       !HasPrefix("--sysroot=")) {
339b6626515SKadir Cetinkaya     ToAppend.push_back("-isysroot");
340b6626515SKadir Cetinkaya     ToAppend.push_back(*Sysroot);
341b6626515SKadir Cetinkaya   }
342b6626515SKadir Cetinkaya 
343b6626515SKadir Cetinkaya   if (!ToAppend.empty()) {
3440a3c7960SKadir Cetinkaya     Cmd.insert(llvm::find(Cmd, "--"), std::make_move_iterator(ToAppend.begin()),
3450a3c7960SKadir Cetinkaya                std::make_move_iterator(ToAppend.end()));
34699768b24SSam McCall   }
34799768b24SSam McCall 
3482a3ac01bSSam McCall   if (!Cmd.empty()) {
349f489fb3dSKon     bool FollowSymlink = !HasExact("-no-canonical-prefixes");
3502a3ac01bSSam McCall     Cmd.front() =
3512a3ac01bSSam McCall         (FollowSymlink ? ResolvedDrivers : ResolvedDriversNoFollow)
3522a3ac01bSSam McCall             .get(Cmd.front(), [&, this] {
3532a3ac01bSSam McCall               return resolveDriver(Cmd.front(), FollowSymlink, ClangPath);
3542a3ac01bSSam McCall             });
35599768b24SSam McCall   }
35699768b24SSam McCall }
35799768b24SSam McCall 
3588eb8c92eSSam McCall // ArgStripper implementation
3598eb8c92eSSam McCall namespace {
3608eb8c92eSSam McCall 
3618eb8c92eSSam McCall // Determine total number of args consumed by this option.
3628eb8c92eSSam McCall // Return answers for {Exact, Prefix} match. 0 means not allowed.
3638eb8c92eSSam McCall std::pair<unsigned, unsigned> getArgCount(const llvm::opt::Option &Opt) {
3648eb8c92eSSam McCall   constexpr static unsigned Rest = 10000; // Should be all the rest!
3658eb8c92eSSam McCall   // Reference is llvm::opt::Option::acceptInternal()
3668eb8c92eSSam McCall   using llvm::opt::Option;
3678eb8c92eSSam McCall   switch (Opt.getKind()) {
3688eb8c92eSSam McCall   case Option::FlagClass:
3698eb8c92eSSam McCall     return {1, 0};
3708eb8c92eSSam McCall   case Option::JoinedClass:
3718eb8c92eSSam McCall   case Option::CommaJoinedClass:
3728eb8c92eSSam McCall     return {1, 1};
3738eb8c92eSSam McCall   case Option::GroupClass:
3748eb8c92eSSam McCall   case Option::InputClass:
3758eb8c92eSSam McCall   case Option::UnknownClass:
3768eb8c92eSSam McCall   case Option::ValuesClass:
3778eb8c92eSSam McCall     return {1, 0};
3788eb8c92eSSam McCall   case Option::JoinedAndSeparateClass:
3798eb8c92eSSam McCall     return {2, 2};
3808eb8c92eSSam McCall   case Option::SeparateClass:
3818eb8c92eSSam McCall     return {2, 0};
3828eb8c92eSSam McCall   case Option::MultiArgClass:
3838eb8c92eSSam McCall     return {1 + Opt.getNumArgs(), 0};
3848eb8c92eSSam McCall   case Option::JoinedOrSeparateClass:
3858eb8c92eSSam McCall     return {2, 1};
3868eb8c92eSSam McCall   case Option::RemainingArgsClass:
3878eb8c92eSSam McCall     return {Rest, 0};
3888eb8c92eSSam McCall   case Option::RemainingArgsJoinedClass:
3898eb8c92eSSam McCall     return {Rest, Rest};
3908eb8c92eSSam McCall   }
39127433228SMikael Holmen   llvm_unreachable("Unhandled option kind");
3928eb8c92eSSam McCall }
3938eb8c92eSSam McCall 
3948eb8c92eSSam McCall // Flag-parsing mode, which affects which flags are available.
3958eb8c92eSSam McCall enum DriverMode : unsigned char {
3968eb8c92eSSam McCall   DM_None = 0,
3978eb8c92eSSam McCall   DM_GCC = 1, // Default mode e.g. when invoked as 'clang'
3988eb8c92eSSam McCall   DM_CL = 2,  // MS CL.exe compatible mode e.g. when invoked as 'clang-cl'
3998eb8c92eSSam McCall   DM_CC1 = 4, // When invoked as 'clang -cc1' or after '-Xclang'
4008eb8c92eSSam McCall   DM_All = 7
4018eb8c92eSSam McCall };
4028eb8c92eSSam McCall 
4038eb8c92eSSam McCall // Examine args list to determine if we're in GCC, CL-compatible, or cc1 mode.
4048eb8c92eSSam McCall DriverMode getDriverMode(const std::vector<std::string> &Args) {
4058eb8c92eSSam McCall   DriverMode Mode = DM_GCC;
4068eb8c92eSSam McCall   llvm::StringRef Argv0 = Args.front();
407ed1539c6SKazu Hirata   if (Argv0.ends_with_insensitive(".exe"))
4088eb8c92eSSam McCall     Argv0 = Argv0.drop_back(strlen(".exe"));
409ed1539c6SKazu Hirata   if (Argv0.ends_with_insensitive("cl"))
4108eb8c92eSSam McCall     Mode = DM_CL;
4118eb8c92eSSam McCall   for (const llvm::StringRef Arg : Args) {
4128eb8c92eSSam McCall     if (Arg == "--driver-mode=cl") {
4138eb8c92eSSam McCall       Mode = DM_CL;
4148eb8c92eSSam McCall       break;
4158eb8c92eSSam McCall     }
4168eb8c92eSSam McCall     if (Arg == "-cc1") {
4178eb8c92eSSam McCall       Mode = DM_CC1;
4188eb8c92eSSam McCall       break;
4198eb8c92eSSam McCall     }
4208eb8c92eSSam McCall   }
4218eb8c92eSSam McCall   return Mode;
4228eb8c92eSSam McCall }
4238eb8c92eSSam McCall 
4248eb8c92eSSam McCall // Returns the set of DriverModes where an option may be used.
4258eb8c92eSSam McCall unsigned char getModes(const llvm::opt::Option &Opt) {
4268eb8c92eSSam McCall   unsigned char Result = DM_None;
4279478f661SJustin Bogner   if (Opt.hasVisibilityFlag(driver::options::ClangOption))
4288eb8c92eSSam McCall     Result |= DM_GCC;
4299478f661SJustin Bogner   if (Opt.hasVisibilityFlag(driver::options::CC1Option))
4309478f661SJustin Bogner     Result |= DM_CC1;
4319478f661SJustin Bogner   if (Opt.hasVisibilityFlag(driver::options::CLOption))
4328eb8c92eSSam McCall     Result |= DM_CL;
4338eb8c92eSSam McCall   return Result;
43427433228SMikael Holmen }
4358eb8c92eSSam McCall 
4368eb8c92eSSam McCall } // namespace
4378eb8c92eSSam McCall 
4388eb8c92eSSam McCall llvm::ArrayRef<ArgStripper::Rule> ArgStripper::rulesFor(llvm::StringRef Arg) {
4398eb8c92eSSam McCall   // All the hard work is done once in a static initializer.
4408eb8c92eSSam McCall   // We compute a table containing strings to look for and #args to skip.
4418eb8c92eSSam McCall   // e.g. "-x" => {-x 2 args, -x* 1 arg, --language 2 args, --language=* 1 arg}
4428eb8c92eSSam McCall   using TableTy =
4438eb8c92eSSam McCall       llvm::StringMap<llvm::SmallVector<Rule, 4>, llvm::BumpPtrAllocator>;
4448eb8c92eSSam McCall   static TableTy *Table = [] {
4458eb8c92eSSam McCall     auto &DriverTable = driver::getDriverOptTable();
4468eb8c92eSSam McCall     using DriverID = clang::driver::options::ID;
4478eb8c92eSSam McCall 
4488eb8c92eSSam McCall     // Collect sets of aliases, so we can treat -foo and -foo= as synonyms.
4498eb8c92eSSam McCall     // Conceptually a double-linked list: PrevAlias[I] -> I -> NextAlias[I].
4508eb8c92eSSam McCall     // If PrevAlias[I] is INVALID, then I is canonical.
4518eb8c92eSSam McCall     DriverID PrevAlias[DriverID::LastOption] = {DriverID::OPT_INVALID};
4528eb8c92eSSam McCall     DriverID NextAlias[DriverID::LastOption] = {DriverID::OPT_INVALID};
4538eb8c92eSSam McCall     auto AddAlias = [&](DriverID Self, DriverID T) {
4548eb8c92eSSam McCall       if (NextAlias[T]) {
4558eb8c92eSSam McCall         PrevAlias[NextAlias[T]] = Self;
4568eb8c92eSSam McCall         NextAlias[Self] = NextAlias[T];
4578eb8c92eSSam McCall       }
4588eb8c92eSSam McCall       PrevAlias[Self] = T;
4598eb8c92eSSam McCall       NextAlias[T] = Self;
4608eb8c92eSSam McCall     };
4618eb8c92eSSam McCall 
4623b64ab57SYuanfang Chen     struct {
4633b64ab57SYuanfang Chen       DriverID ID;
4643b64ab57SYuanfang Chen       DriverID AliasID;
4654f30a526SYuanfang Chen       const void *AliasArgs;
4663b64ab57SYuanfang Chen     } AliasTable[] = {
467501f92d3SJan Svoboda #define OPTION(PREFIX, PREFIXED_NAME, ID, KIND, GROUP, ALIAS, ALIASARGS,       \
468aff197ffSDavid Spickett                FLAGS, VISIBILITY, PARAM, HELPTEXT, HELPTEXTSFORVARIANTS,       \
469aff197ffSDavid Spickett                METAVAR, VALUES)                                                \
4704f30a526SYuanfang Chen   {DriverID::OPT_##ID, DriverID::OPT_##ALIAS, ALIASARGS},
4713b64ab57SYuanfang Chen #include "clang/Driver/Options.inc"
4723b64ab57SYuanfang Chen #undef OPTION
4733b64ab57SYuanfang Chen     };
4743b64ab57SYuanfang Chen     for (auto &E : AliasTable)
4753b64ab57SYuanfang Chen       if (E.AliasID != DriverID::OPT_INVALID && E.AliasArgs == nullptr)
4763b64ab57SYuanfang Chen         AddAlias(E.ID, E.AliasID);
4773b64ab57SYuanfang Chen 
4788eb8c92eSSam McCall     auto Result = std::make_unique<TableTy>();
4798eb8c92eSSam McCall     // Iterate over distinct options (represented by the canonical alias).
4808eb8c92eSSam McCall     // Every spelling of this option will get the same set of rules.
4818eb8c92eSSam McCall     for (unsigned ID = 1 /*Skip INVALID */; ID < DriverID::LastOption; ++ID) {
4828eb8c92eSSam McCall       if (PrevAlias[ID] || ID == DriverID::OPT_Xclang)
4838eb8c92eSSam McCall         continue; // Not canonical, or specially handled.
484ee02e20cSKirill Bobyrev       llvm::SmallVector<Rule> Rules;
4858eb8c92eSSam McCall       // Iterate over each alias, to add rules for parsing it.
4868eb8c92eSSam McCall       for (unsigned A = ID; A != DriverID::OPT_INVALID; A = NextAlias[A]) {
487*dd647e3eSChandler Carruth         llvm::SmallVector<llvm::StringRef, 4> Prefixes;
488*dd647e3eSChandler Carruth         DriverTable.appendOptionPrefixes(A, Prefixes);
489*dd647e3eSChandler Carruth         if (Prefixes.empty()) // option groups.
4908eb8c92eSSam McCall           continue;
4918eb8c92eSSam McCall         auto Opt = DriverTable.getOption(A);
4928eb8c92eSSam McCall         // Exclude - and -foo pseudo-options.
4938eb8c92eSSam McCall         if (Opt.getName().empty())
4948eb8c92eSSam McCall           continue;
4958eb8c92eSSam McCall         auto Modes = getModes(Opt);
4968eb8c92eSSam McCall         std::pair<unsigned, unsigned> ArgCount = getArgCount(Opt);
4978eb8c92eSSam McCall         // Iterate over each spelling of the alias, e.g. -foo vs --foo.
498*dd647e3eSChandler Carruth         for (StringRef Prefix : Prefixes) {
499d9ab3e82Sserge-sans-paille           llvm::SmallString<64> Buf(Prefix);
5008eb8c92eSSam McCall           Buf.append(Opt.getName());
5018eb8c92eSSam McCall           llvm::StringRef Spelling = Result->try_emplace(Buf).first->getKey();
5028eb8c92eSSam McCall           Rules.emplace_back();
5038eb8c92eSSam McCall           Rule &R = Rules.back();
5048eb8c92eSSam McCall           R.Text = Spelling;
5058eb8c92eSSam McCall           R.Modes = Modes;
5068eb8c92eSSam McCall           R.ExactArgs = ArgCount.first;
5078eb8c92eSSam McCall           R.PrefixArgs = ArgCount.second;
5088eb8c92eSSam McCall           // Concrete priority is the index into the option table.
5098eb8c92eSSam McCall           // Effectively, earlier entries take priority over later ones.
5108eb8c92eSSam McCall           assert(ID < std::numeric_limits<decltype(R.Priority)>::max() &&
5118eb8c92eSSam McCall                  "Rules::Priority overflowed by options table");
5128eb8c92eSSam McCall           R.Priority = ID;
5138eb8c92eSSam McCall         }
5148eb8c92eSSam McCall       }
5158eb8c92eSSam McCall       // Register the set of rules under each possible name.
5168eb8c92eSSam McCall       for (const auto &R : Rules)
5178eb8c92eSSam McCall         Result->find(R.Text)->second.append(Rules.begin(), Rules.end());
5188eb8c92eSSam McCall     }
5198eb8c92eSSam McCall #ifndef NDEBUG
5208eb8c92eSSam McCall     // Dump the table and various measures of its size.
5218eb8c92eSSam McCall     unsigned RuleCount = 0;
5228eb8c92eSSam McCall     dlog("ArgStripper Option spelling table");
5238eb8c92eSSam McCall     for (const auto &Entry : *Result) {
5248eb8c92eSSam McCall       dlog("{0}", Entry.first());
5258eb8c92eSSam McCall       RuleCount += Entry.second.size();
5268eb8c92eSSam McCall       for (const auto &R : Entry.second)
5278eb8c92eSSam McCall         dlog("  {0} #={1} *={2} Mode={3}", R.Text, R.ExactArgs, R.PrefixArgs,
5288eb8c92eSSam McCall              int(R.Modes));
5298eb8c92eSSam McCall     }
5308eb8c92eSSam McCall     dlog("Table spellings={0} rules={1} string-bytes={2}", Result->size(),
5318eb8c92eSSam McCall          RuleCount, Result->getAllocator().getBytesAllocated());
5328eb8c92eSSam McCall #endif
5338eb8c92eSSam McCall     // The static table will never be destroyed.
5348eb8c92eSSam McCall     return Result.release();
5358eb8c92eSSam McCall   }();
5368eb8c92eSSam McCall 
5378eb8c92eSSam McCall   auto It = Table->find(Arg);
5388eb8c92eSSam McCall   return (It == Table->end()) ? llvm::ArrayRef<Rule>() : It->second;
5398eb8c92eSSam McCall }
5408eb8c92eSSam McCall 
5418eb8c92eSSam McCall void ArgStripper::strip(llvm::StringRef Arg) {
5428eb8c92eSSam McCall   auto OptionRules = rulesFor(Arg);
5438eb8c92eSSam McCall   if (OptionRules.empty()) {
5448eb8c92eSSam McCall     // Not a recognized flag. Strip it literally.
5458eb8c92eSSam McCall     Storage.emplace_back(Arg);
5468eb8c92eSSam McCall     Rules.emplace_back();
5478eb8c92eSSam McCall     Rules.back().Text = Storage.back();
5488eb8c92eSSam McCall     Rules.back().ExactArgs = 1;
5498eb8c92eSSam McCall     if (Rules.back().Text.consume_back("*"))
5508eb8c92eSSam McCall       Rules.back().PrefixArgs = 1;
5518eb8c92eSSam McCall     Rules.back().Modes = DM_All;
5528eb8c92eSSam McCall     Rules.back().Priority = -1; // Max unsigned = lowest priority.
5538eb8c92eSSam McCall   } else {
5548eb8c92eSSam McCall     Rules.append(OptionRules.begin(), OptionRules.end());
5558eb8c92eSSam McCall   }
5568eb8c92eSSam McCall }
5578eb8c92eSSam McCall 
5588eb8c92eSSam McCall const ArgStripper::Rule *ArgStripper::matchingRule(llvm::StringRef Arg,
5598eb8c92eSSam McCall                                                    unsigned Mode,
5608eb8c92eSSam McCall                                                    unsigned &ArgCount) const {
5618eb8c92eSSam McCall   const ArgStripper::Rule *BestRule = nullptr;
5628eb8c92eSSam McCall   for (const Rule &R : Rules) {
5638eb8c92eSSam McCall     // Rule can fail to match if...
5648eb8c92eSSam McCall     if (!(R.Modes & Mode))
5658eb8c92eSSam McCall       continue; // not applicable to current driver mode
5668eb8c92eSSam McCall     if (BestRule && BestRule->Priority < R.Priority)
5678eb8c92eSSam McCall       continue; // lower-priority than best candidate.
568d5953e3eSKazu Hirata     if (!Arg.starts_with(R.Text))
5698eb8c92eSSam McCall       continue; // current arg doesn't match the prefix string
5708eb8c92eSSam McCall     bool PrefixMatch = Arg.size() > R.Text.size();
5718eb8c92eSSam McCall     // Can rule apply as an exact/prefix match?
5728eb8c92eSSam McCall     if (unsigned Count = PrefixMatch ? R.PrefixArgs : R.ExactArgs) {
5738eb8c92eSSam McCall       BestRule = &R;
5748eb8c92eSSam McCall       ArgCount = Count;
5758eb8c92eSSam McCall     }
5768eb8c92eSSam McCall     // Continue in case we find a higher-priority rule.
5778eb8c92eSSam McCall   }
5788eb8c92eSSam McCall   return BestRule;
5798eb8c92eSSam McCall }
5808eb8c92eSSam McCall 
5818eb8c92eSSam McCall void ArgStripper::process(std::vector<std::string> &Args) const {
5828eb8c92eSSam McCall   if (Args.empty())
5838eb8c92eSSam McCall     return;
5848eb8c92eSSam McCall 
5858eb8c92eSSam McCall   // We're parsing the args list in some mode (e.g. gcc-compatible) but may
5868eb8c92eSSam McCall   // temporarily switch to another mode with the -Xclang flag.
5878eb8c92eSSam McCall   DriverMode MainMode = getDriverMode(Args);
5888eb8c92eSSam McCall   DriverMode CurrentMode = MainMode;
5898eb8c92eSSam McCall 
5908eb8c92eSSam McCall   // Read and write heads for in-place deletion.
5918eb8c92eSSam McCall   unsigned Read = 0, Write = 0;
5928eb8c92eSSam McCall   bool WasXclang = false;
5938eb8c92eSSam McCall   while (Read < Args.size()) {
5948eb8c92eSSam McCall     unsigned ArgCount = 0;
59527433228SMikael Holmen     if (matchingRule(Args[Read], CurrentMode, ArgCount)) {
5968eb8c92eSSam McCall       // Delete it and its args.
5978eb8c92eSSam McCall       if (WasXclang) {
5988eb8c92eSSam McCall         assert(Write > 0);
5998eb8c92eSSam McCall         --Write; // Drop previous -Xclang arg
6008eb8c92eSSam McCall         CurrentMode = MainMode;
6018eb8c92eSSam McCall         WasXclang = false;
6028eb8c92eSSam McCall       }
6038eb8c92eSSam McCall       // Advance to last arg. An arg may be foo or -Xclang foo.
6048eb8c92eSSam McCall       for (unsigned I = 1; Read < Args.size() && I < ArgCount; ++I) {
6058eb8c92eSSam McCall         ++Read;
6068eb8c92eSSam McCall         if (Read < Args.size() && Args[Read] == "-Xclang")
6078eb8c92eSSam McCall           ++Read;
6088eb8c92eSSam McCall       }
6098eb8c92eSSam McCall     } else {
6108eb8c92eSSam McCall       // No match, just copy the arg through.
6118eb8c92eSSam McCall       WasXclang = Args[Read] == "-Xclang";
6128eb8c92eSSam McCall       CurrentMode = WasXclang ? DM_CC1 : MainMode;
6138eb8c92eSSam McCall       if (Write != Read)
6148eb8c92eSSam McCall         Args[Write] = std::move(Args[Read]);
6158eb8c92eSSam McCall       ++Write;
6168eb8c92eSSam McCall     }
6178eb8c92eSSam McCall     ++Read;
6188eb8c92eSSam McCall   }
6198eb8c92eSSam McCall   Args.resize(Write);
6208eb8c92eSSam McCall }
6218eb8c92eSSam McCall 
6227de711ecSSam McCall std::string printArgv(llvm::ArrayRef<llvm::StringRef> Args) {
6237de711ecSSam McCall   std::string Buf;
6247de711ecSSam McCall   llvm::raw_string_ostream OS(Buf);
6257de711ecSSam McCall   bool Sep = false;
6267de711ecSSam McCall   for (llvm::StringRef Arg : Args) {
6277de711ecSSam McCall     if (Sep)
6287de711ecSSam McCall       OS << ' ';
6297de711ecSSam McCall     Sep = true;
6307de711ecSSam McCall     if (llvm::all_of(Arg, llvm::isPrint) &&
6317de711ecSSam McCall         Arg.find_first_of(" \t\n\"\\") == llvm::StringRef::npos) {
6327de711ecSSam McCall       OS << Arg;
6337de711ecSSam McCall       continue;
6347de711ecSSam McCall     }
6357de711ecSSam McCall     OS << '"';
6367de711ecSSam McCall     OS.write_escaped(Arg, /*UseHexEscapes=*/true);
6377de711ecSSam McCall     OS << '"';
6387de711ecSSam McCall   }
6397de711ecSSam McCall   return std::move(OS.str());
6407de711ecSSam McCall }
6417de711ecSSam McCall 
6427de711ecSSam McCall std::string printArgv(llvm::ArrayRef<std::string> Args) {
6437de711ecSSam McCall   std::vector<llvm::StringRef> Refs(Args.size());
6447de711ecSSam McCall   llvm::copy(Args, Refs.begin());
6457de711ecSSam McCall   return printArgv(Refs);
6467de711ecSSam McCall }
6477de711ecSSam McCall 
64899768b24SSam McCall } // namespace clangd
64999768b24SSam McCall } // namespace clang
650