1 //===--- CompileCommands.cpp ----------------------------------------------===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 
9 #include "CompileCommands.h"
10 #include "Config.h"
11 #include "support/Logger.h"
12 #include "support/Trace.h"
13 #include "clang/Driver/Driver.h"
14 #include "clang/Driver/Options.h"
15 #include "clang/Frontend/CompilerInvocation.h"
16 #include "clang/Tooling/ArgumentsAdjusters.h"
17 #include "clang/Tooling/CompilationDatabase.h"
18 #include "llvm/ADT/ArrayRef.h"
19 #include "llvm/ADT/STLExtras.h"
20 #include "llvm/ADT/SmallVector.h"
21 #include "llvm/ADT/StringRef.h"
22 #include "llvm/Option/ArgList.h"
23 #include "llvm/Option/Option.h"
24 #include "llvm/Support/Allocator.h"
25 #include "llvm/Support/Debug.h"
26 #include "llvm/Support/FileSystem.h"
27 #include "llvm/Support/FileUtilities.h"
28 #include "llvm/Support/MemoryBuffer.h"
29 #include "llvm/Support/Path.h"
30 #include "llvm/Support/Program.h"
31 #include <iterator>
32 #include <string>
33 #include <vector>
34 
35 namespace clang {
36 namespace clangd {
37 namespace {
38 
39 // Query apple's `xcrun` launcher, which is the source of truth for "how should"
40 // clang be invoked on this system.
queryXcrun(llvm::ArrayRef<llvm::StringRef> Argv)41 llvm::Optional<std::string> queryXcrun(llvm::ArrayRef<llvm::StringRef> Argv) {
42   auto Xcrun = llvm::sys::findProgramByName("xcrun");
43   if (!Xcrun) {
44     log("Couldn't find xcrun. Hopefully you have a non-apple toolchain...");
45     return llvm::None;
46   }
47   llvm::SmallString<64> OutFile;
48   llvm::sys::fs::createTemporaryFile("clangd-xcrun", "", OutFile);
49   llvm::FileRemover OutRemover(OutFile);
50   llvm::Optional<llvm::StringRef> Redirects[3] = {
51       /*stdin=*/{""}, /*stdout=*/{OutFile.str()}, /*stderr=*/{""}};
52   vlog("Invoking {0} to find clang installation", *Xcrun);
53   int Ret = llvm::sys::ExecuteAndWait(*Xcrun, Argv,
54                                       /*Env=*/llvm::None, Redirects,
55                                       /*SecondsToWait=*/10);
56   if (Ret != 0) {
57     log("xcrun exists but failed with code {0}. "
58         "If you have a non-apple toolchain, this is OK. "
59         "Otherwise, try xcode-select --install.",
60         Ret);
61     return llvm::None;
62   }
63 
64   auto Buf = llvm::MemoryBuffer::getFile(OutFile);
65   if (!Buf) {
66     log("Can't read xcrun output: {0}", Buf.getError().message());
67     return llvm::None;
68   }
69   StringRef Path = Buf->get()->getBuffer().trim();
70   if (Path.empty()) {
71     log("xcrun produced no output");
72     return llvm::None;
73   }
74   return Path.str();
75 }
76 
77 // Resolve symlinks if possible.
resolve(std::string Path)78 std::string resolve(std::string Path) {
79   llvm::SmallString<128> Resolved;
80   if (llvm::sys::fs::real_path(Path, Resolved)) {
81     log("Failed to resolve possible symlink {0}", Path);
82     return Path;
83   }
84   return std::string(Resolved.str());
85 }
86 
87 // Get a plausible full `clang` path.
88 // This is used in the fallback compile command, or when the CDB returns a
89 // generic driver with no path.
detectClangPath()90 std::string detectClangPath() {
91   // The driver and/or cc1 sometimes depend on the binary name to compute
92   // useful things like the standard library location.
93   // We need to emulate what clang on this system is likely to see.
94   // cc1 in particular looks at the "real path" of the running process, and
95   // so if /usr/bin/clang is a symlink, it sees the resolved path.
96   // clangd doesn't have that luxury, so we resolve symlinks ourselves.
97 
98   // On Mac, `which clang` is /usr/bin/clang. It runs `xcrun clang`, which knows
99   // where the real clang is kept. We need to do the same thing,
100   // because cc1 (not the driver!) will find libc++ relative to argv[0].
101 #ifdef __APPLE__
102   if (auto MacClang = queryXcrun({"xcrun", "--find", "clang"}))
103     return resolve(std::move(*MacClang));
104 #endif
105   // On other platforms, just look for compilers on the PATH.
106   for (const char *Name : {"clang", "gcc", "cc"})
107     if (auto PathCC = llvm::sys::findProgramByName(Name))
108       return resolve(std::move(*PathCC));
109   // Fallback: a nonexistent 'clang' binary next to clangd.
110   static int StaticForMainAddr;
111   std::string ClangdExecutable =
112       llvm::sys::fs::getMainExecutable("clangd", (void *)&StaticForMainAddr);
113   SmallString<128> ClangPath;
114   ClangPath = llvm::sys::path::parent_path(ClangdExecutable);
115   llvm::sys::path::append(ClangPath, "clang");
116   return std::string(ClangPath.str());
117 }
118 
119 // On mac, /usr/bin/clang sets SDKROOT and then invokes the real clang.
120 // The effect of this is to set -isysroot correctly. We do the same.
detectSysroot()121 const llvm::Optional<std::string> detectSysroot() {
122 #ifndef __APPLE__
123   return llvm::None;
124 #endif
125 
126   // SDKROOT overridden in environment, respect it. Driver will set isysroot.
127   if (::getenv("SDKROOT"))
128     return llvm::None;
129   return queryXcrun({"xcrun", "--show-sdk-path"});
130 }
131 
detectStandardResourceDir()132 std::string detectStandardResourceDir() {
133   static int StaticForMainAddr; // Just an address in this process.
134   return CompilerInvocation::GetResourcesPath("clangd",
135                                               (void *)&StaticForMainAddr);
136 }
137 
138 // The path passed to argv[0] is important:
139 //  - its parent directory is Driver::Dir, used for library discovery
140 //  - its basename affects CLI parsing (clang-cl) and other settings
141 // Where possible it should be an absolute path with sensible directory, but
142 // with the original basename.
resolveDriver(llvm::StringRef Driver,bool FollowSymlink,llvm::Optional<std::string> ClangPath)143 static std::string resolveDriver(llvm::StringRef Driver, bool FollowSymlink,
144                                  llvm::Optional<std::string> ClangPath) {
145   auto SiblingOf = [&](llvm::StringRef AbsPath) {
146     llvm::SmallString<128> Result = llvm::sys::path::parent_path(AbsPath);
147     llvm::sys::path::append(Result, llvm::sys::path::filename(Driver));
148     return Result.str().str();
149   };
150 
151   // First, eliminate relative paths.
152   std::string Storage;
153   if (!llvm::sys::path::is_absolute(Driver)) {
154     // If it's working-dir relative like bin/clang, we can't resolve it.
155     // FIXME: we could if we had the working directory here.
156     // Let's hope it's not a symlink.
157     if (llvm::any_of(Driver,
158                      [](char C) { return llvm::sys::path::is_separator(C); }))
159       return Driver.str();
160     // If the driver is a generic like "g++" with no path, add clang dir.
161     if (ClangPath &&
162         (Driver == "clang" || Driver == "clang++" || Driver == "gcc" ||
163          Driver == "g++" || Driver == "cc" || Driver == "c++")) {
164       return SiblingOf(*ClangPath);
165     }
166     // Otherwise try to look it up on PATH. This won't change basename.
167     auto Absolute = llvm::sys::findProgramByName(Driver);
168     if (Absolute && llvm::sys::path::is_absolute(*Absolute))
169       Driver = Storage = std::move(*Absolute);
170     else if (ClangPath) // If we don't find it, use clang dir again.
171       return SiblingOf(*ClangPath);
172     else // Nothing to do: can't find the command and no detected dir.
173       return Driver.str();
174   }
175 
176   // Now we have an absolute path, but it may be a symlink.
177   assert(llvm::sys::path::is_absolute(Driver));
178   if (FollowSymlink) {
179     llvm::SmallString<256> Resolved;
180     if (!llvm::sys::fs::real_path(Driver, Resolved))
181       return SiblingOf(Resolved);
182   }
183   return Driver.str();
184 }
185 
186 } // namespace
187 
detect()188 CommandMangler CommandMangler::detect() {
189   CommandMangler Result;
190   Result.ClangPath = detectClangPath();
191   Result.ResourceDir = detectStandardResourceDir();
192   Result.Sysroot = detectSysroot();
193   return Result;
194 }
195 
forTests()196 CommandMangler CommandMangler::forTests() { return CommandMangler(); }
197 
adjust(std::vector<std::string> & Cmd,llvm::StringRef File) const198 void CommandMangler::adjust(std::vector<std::string> &Cmd,
199                             llvm::StringRef File) const {
200   trace::Span S("AdjustCompileFlags");
201   // Most of the modifications below assumes the Cmd starts with a driver name.
202   // We might consider injecting a generic driver name like "cc" or "c++", but
203   // a Cmd missing the driver is probably rare enough in practice and errnous.
204   if (Cmd.empty())
205     return;
206   auto &OptTable = clang::driver::getDriverOptTable();
207   // OriginalArgs needs to outlive ArgList.
208   llvm::SmallVector<const char *, 16> OriginalArgs;
209   OriginalArgs.reserve(Cmd.size());
210   for (const auto &S : Cmd)
211     OriginalArgs.push_back(S.c_str());
212   bool IsCLMode = driver::IsClangCL(driver::getDriverMode(
213       OriginalArgs[0], llvm::makeArrayRef(OriginalArgs).slice(1)));
214   // ParseArgs propagates missig arg/opt counts on error, but preserves
215   // everything it could parse in ArgList. So we just ignore those counts.
216   unsigned IgnoredCount;
217   // Drop the executable name, as ParseArgs doesn't expect it. This means
218   // indices are actually of by one between ArgList and OriginalArgs.
219   llvm::opt::InputArgList ArgList;
220   ArgList = OptTable.ParseArgs(
221       llvm::makeArrayRef(OriginalArgs).drop_front(), IgnoredCount, IgnoredCount,
222       /*FlagsToInclude=*/
223       IsCLMode ? (driver::options::CLOption | driver::options::CoreOption |
224                   driver::options::CLDXCOption)
225                : /*everything*/ 0,
226       /*FlagsToExclude=*/driver::options::NoDriverOption |
227           (IsCLMode
228                ? 0
229                : (driver::options::CLOption | driver::options::CLDXCOption)));
230 
231   llvm::SmallVector<unsigned, 1> IndicesToDrop;
232   // Having multiple architecture options (e.g. when building fat binaries)
233   // results in multiple compiler jobs, which clangd cannot handle. In such
234   // cases strip all the `-arch` options and fallback to default architecture.
235   // As there are no signals to figure out which one user actually wants. They
236   // can explicitly specify one through `CompileFlags.Add` if need be.
237   unsigned ArchOptCount = 0;
238   for (auto *Input : ArgList.filtered(driver::options::OPT_arch)) {
239     ++ArchOptCount;
240     for (auto I = 0U; I <= Input->getNumValues(); ++I)
241       IndicesToDrop.push_back(Input->getIndex() + I);
242   }
243   // If there is a single `-arch` option, keep it.
244   if (ArchOptCount < 2)
245     IndicesToDrop.clear();
246 
247   // In some cases people may try to reuse the command from another file, e.g.
248   //   { File: "foo.h", CommandLine: "clang foo.cpp" }.
249   // We assume the intent is to parse foo.h the same way as foo.cpp, or as if
250   // it were being included from foo.cpp.
251   //
252   // We're going to rewrite the command to refer to foo.h, and this may change
253   // its semantics (e.g. by parsing the file as C). If we do this, we should
254   // use transferCompileCommand to adjust the argv.
255   // In practice only the extension of the file matters, so do this only when
256   // it differs.
257   llvm::StringRef FileExtension = llvm::sys::path::extension(File);
258   llvm::Optional<std::string> TransferFrom;
259   auto SawInput = [&](llvm::StringRef Input) {
260     if (llvm::sys::path::extension(Input) != FileExtension)
261       TransferFrom.emplace(Input);
262   };
263 
264   // Strip all the inputs and `--`. We'll put the input for the requested file
265   // explicitly at the end of the flags. This ensures modifications done in the
266   // following steps apply in more cases (like setting -x, which only affects
267   // inputs that come after it).
268   for (auto *Input : ArgList.filtered(driver::options::OPT_INPUT)) {
269     SawInput(Input->getValue(0));
270     IndicesToDrop.push_back(Input->getIndex());
271   }
272   // Anything after `--` is also treated as input, drop them as well.
273   if (auto *DashDash =
274           ArgList.getLastArgNoClaim(driver::options::OPT__DASH_DASH)) {
275     auto DashDashIndex = DashDash->getIndex() + 1; // +1 accounts for Cmd[0]
276     for (unsigned I = DashDashIndex; I < Cmd.size(); ++I)
277       SawInput(Cmd[I]);
278     Cmd.resize(DashDashIndex);
279   }
280   llvm::sort(IndicesToDrop);
281   llvm::for_each(llvm::reverse(IndicesToDrop),
282                  // +1 to account for the executable name in Cmd[0] that
283                  // doesn't exist in ArgList.
284                  [&Cmd](unsigned Idx) { Cmd.erase(Cmd.begin() + Idx + 1); });
285   // All the inputs are stripped, append the name for the requested file. Rest
286   // of the modifications should respect `--`.
287   Cmd.push_back("--");
288   Cmd.push_back(File.str());
289 
290   if (TransferFrom) {
291     tooling::CompileCommand TransferCmd;
292     TransferCmd.Filename = std::move(*TransferFrom);
293     TransferCmd.CommandLine = std::move(Cmd);
294     TransferCmd = transferCompileCommand(std::move(TransferCmd), File);
295     Cmd = std::move(TransferCmd.CommandLine);
296     assert(Cmd.size() >= 2 && Cmd.back() == File &&
297            Cmd[Cmd.size() - 2] == "--" &&
298            "TransferCommand should produce a command ending in -- filename");
299   }
300 
301   for (auto &Edit : Config::current().CompileFlags.Edits)
302     Edit(Cmd);
303 
304   // Check whether the flag exists, either as -flag or -flag=*
305   auto Has = [&](llvm::StringRef Flag) {
306     for (llvm::StringRef Arg : Cmd) {
307       if (Arg.consume_front(Flag) && (Arg.empty() || Arg[0] == '='))
308         return true;
309     }
310     return false;
311   };
312 
313   llvm::erase_if(Cmd, [](llvm::StringRef Elem) {
314     return Elem.startswith("--save-temps") || Elem.startswith("-save-temps");
315   });
316 
317   std::vector<std::string> ToAppend;
318   if (ResourceDir && !Has("-resource-dir"))
319     ToAppend.push_back(("-resource-dir=" + *ResourceDir));
320 
321   // Don't set `-isysroot` if it is already set or if `--sysroot` is set.
322   // `--sysroot` is a superset of the `-isysroot` argument.
323   if (Sysroot && !Has("-isysroot") && !Has("--sysroot")) {
324     ToAppend.push_back("-isysroot");
325     ToAppend.push_back(*Sysroot);
326   }
327 
328   if (!ToAppend.empty()) {
329     Cmd.insert(llvm::find(Cmd, "--"), std::make_move_iterator(ToAppend.begin()),
330                std::make_move_iterator(ToAppend.end()));
331   }
332 
333   if (!Cmd.empty()) {
334     bool FollowSymlink = !Has("-no-canonical-prefixes");
335     Cmd.front() =
336         (FollowSymlink ? ResolvedDrivers : ResolvedDriversNoFollow)
337             .get(Cmd.front(), [&, this] {
338               return resolveDriver(Cmd.front(), FollowSymlink, ClangPath);
339             });
340   }
341 }
342 
operator clang::tooling::ArgumentsAdjuster()343 CommandMangler::operator clang::tooling::ArgumentsAdjuster() && {
344   // ArgumentsAdjuster is a std::function and so must be copyable.
345   return [Mangler = std::make_shared<CommandMangler>(std::move(*this))](
346              const std::vector<std::string> &Args, llvm::StringRef File) {
347     auto Result = Args;
348     Mangler->adjust(Result, File);
349     return Result;
350   };
351 }
352 
353 // ArgStripper implementation
354 namespace {
355 
356 // Determine total number of args consumed by this option.
357 // Return answers for {Exact, Prefix} match. 0 means not allowed.
getArgCount(const llvm::opt::Option & Opt)358 std::pair<unsigned, unsigned> getArgCount(const llvm::opt::Option &Opt) {
359   constexpr static unsigned Rest = 10000; // Should be all the rest!
360   // Reference is llvm::opt::Option::acceptInternal()
361   using llvm::opt::Option;
362   switch (Opt.getKind()) {
363   case Option::FlagClass:
364     return {1, 0};
365   case Option::JoinedClass:
366   case Option::CommaJoinedClass:
367     return {1, 1};
368   case Option::GroupClass:
369   case Option::InputClass:
370   case Option::UnknownClass:
371   case Option::ValuesClass:
372     return {1, 0};
373   case Option::JoinedAndSeparateClass:
374     return {2, 2};
375   case Option::SeparateClass:
376     return {2, 0};
377   case Option::MultiArgClass:
378     return {1 + Opt.getNumArgs(), 0};
379   case Option::JoinedOrSeparateClass:
380     return {2, 1};
381   case Option::RemainingArgsClass:
382     return {Rest, 0};
383   case Option::RemainingArgsJoinedClass:
384     return {Rest, Rest};
385   }
386   llvm_unreachable("Unhandled option kind");
387 }
388 
389 // Flag-parsing mode, which affects which flags are available.
390 enum DriverMode : unsigned char {
391   DM_None = 0,
392   DM_GCC = 1, // Default mode e.g. when invoked as 'clang'
393   DM_CL = 2,  // MS CL.exe compatible mode e.g. when invoked as 'clang-cl'
394   DM_CC1 = 4, // When invoked as 'clang -cc1' or after '-Xclang'
395   DM_All = 7
396 };
397 
398 // Examine args list to determine if we're in GCC, CL-compatible, or cc1 mode.
getDriverMode(const std::vector<std::string> & Args)399 DriverMode getDriverMode(const std::vector<std::string> &Args) {
400   DriverMode Mode = DM_GCC;
401   llvm::StringRef Argv0 = Args.front();
402   if (Argv0.endswith_insensitive(".exe"))
403     Argv0 = Argv0.drop_back(strlen(".exe"));
404   if (Argv0.endswith_insensitive("cl"))
405     Mode = DM_CL;
406   for (const llvm::StringRef Arg : Args) {
407     if (Arg == "--driver-mode=cl") {
408       Mode = DM_CL;
409       break;
410     }
411     if (Arg == "-cc1") {
412       Mode = DM_CC1;
413       break;
414     }
415   }
416   return Mode;
417 }
418 
419 // Returns the set of DriverModes where an option may be used.
getModes(const llvm::opt::Option & Opt)420 unsigned char getModes(const llvm::opt::Option &Opt) {
421   // Why is this so complicated?!
422   // Reference is clang::driver::Driver::getIncludeExcludeOptionFlagMasks()
423   unsigned char Result = DM_None;
424   if (Opt.hasFlag(driver::options::CC1Option))
425     Result |= DM_CC1;
426   if (!Opt.hasFlag(driver::options::NoDriverOption)) {
427     if (Opt.hasFlag(driver::options::CLOption)) {
428       Result |= DM_CL;
429     } else if (Opt.hasFlag(driver::options::CLDXCOption)) {
430       Result |= DM_CL;
431     } else {
432       Result |= DM_GCC;
433       if (Opt.hasFlag(driver::options::CoreOption)) {
434         Result |= DM_CL;
435       }
436     }
437   }
438   return Result;
439 }
440 
441 } // namespace
442 
rulesFor(llvm::StringRef Arg)443 llvm::ArrayRef<ArgStripper::Rule> ArgStripper::rulesFor(llvm::StringRef Arg) {
444   // All the hard work is done once in a static initializer.
445   // We compute a table containing strings to look for and #args to skip.
446   // e.g. "-x" => {-x 2 args, -x* 1 arg, --language 2 args, --language=* 1 arg}
447   using TableTy =
448       llvm::StringMap<llvm::SmallVector<Rule, 4>, llvm::BumpPtrAllocator>;
449   static TableTy *Table = [] {
450     auto &DriverTable = driver::getDriverOptTable();
451     using DriverID = clang::driver::options::ID;
452 
453     // Collect sets of aliases, so we can treat -foo and -foo= as synonyms.
454     // Conceptually a double-linked list: PrevAlias[I] -> I -> NextAlias[I].
455     // If PrevAlias[I] is INVALID, then I is canonical.
456     DriverID PrevAlias[DriverID::LastOption] = {DriverID::OPT_INVALID};
457     DriverID NextAlias[DriverID::LastOption] = {DriverID::OPT_INVALID};
458     auto AddAlias = [&](DriverID Self, DriverID T) {
459       if (NextAlias[T]) {
460         PrevAlias[NextAlias[T]] = Self;
461         NextAlias[Self] = NextAlias[T];
462       }
463       PrevAlias[Self] = T;
464       NextAlias[T] = Self;
465     };
466     // Also grab prefixes for each option, these are not fully exposed.
467     const char *const *Prefixes[DriverID::LastOption] = {nullptr};
468 #define PREFIX(NAME, VALUE) static const char *const NAME[] = VALUE;
469 #define OPTION(PREFIX, NAME, ID, KIND, GROUP, ALIAS, ALIASARGS, FLAGS, PARAM,  \
470                HELP, METAVAR, VALUES)                                          \
471   Prefixes[DriverID::OPT_##ID] = PREFIX;
472 #include "clang/Driver/Options.inc"
473 #undef OPTION
474 #undef PREFIX
475 
476     struct {
477       DriverID ID;
478       DriverID AliasID;
479       const void *AliasArgs;
480     } AliasTable[] = {
481 #define OPTION(PREFIX, NAME, ID, KIND, GROUP, ALIAS, ALIASARGS, FLAGS, PARAM,  \
482                HELP, METAVAR, VALUES)                                          \
483   {DriverID::OPT_##ID, DriverID::OPT_##ALIAS, ALIASARGS},
484 #include "clang/Driver/Options.inc"
485 #undef OPTION
486     };
487     for (auto &E : AliasTable)
488       if (E.AliasID != DriverID::OPT_INVALID && E.AliasArgs == nullptr)
489         AddAlias(E.ID, E.AliasID);
490 
491     auto Result = std::make_unique<TableTy>();
492     // Iterate over distinct options (represented by the canonical alias).
493     // Every spelling of this option will get the same set of rules.
494     for (unsigned ID = 1 /*Skip INVALID */; ID < DriverID::LastOption; ++ID) {
495       if (PrevAlias[ID] || ID == DriverID::OPT_Xclang)
496         continue; // Not canonical, or specially handled.
497       llvm::SmallVector<Rule> Rules;
498       // Iterate over each alias, to add rules for parsing it.
499       for (unsigned A = ID; A != DriverID::OPT_INVALID; A = NextAlias[A]) {
500         if (Prefixes[A] == nullptr) // option groups.
501           continue;
502         auto Opt = DriverTable.getOption(A);
503         // Exclude - and -foo pseudo-options.
504         if (Opt.getName().empty())
505           continue;
506         auto Modes = getModes(Opt);
507         std::pair<unsigned, unsigned> ArgCount = getArgCount(Opt);
508         // Iterate over each spelling of the alias, e.g. -foo vs --foo.
509         for (auto *Prefix = Prefixes[A]; *Prefix != nullptr; ++Prefix) {
510           llvm::SmallString<64> Buf(*Prefix);
511           Buf.append(Opt.getName());
512           llvm::StringRef Spelling = Result->try_emplace(Buf).first->getKey();
513           Rules.emplace_back();
514           Rule &R = Rules.back();
515           R.Text = Spelling;
516           R.Modes = Modes;
517           R.ExactArgs = ArgCount.first;
518           R.PrefixArgs = ArgCount.second;
519           // Concrete priority is the index into the option table.
520           // Effectively, earlier entries take priority over later ones.
521           assert(ID < std::numeric_limits<decltype(R.Priority)>::max() &&
522                  "Rules::Priority overflowed by options table");
523           R.Priority = ID;
524         }
525       }
526       // Register the set of rules under each possible name.
527       for (const auto &R : Rules)
528         Result->find(R.Text)->second.append(Rules.begin(), Rules.end());
529     }
530 #ifndef NDEBUG
531     // Dump the table and various measures of its size.
532     unsigned RuleCount = 0;
533     dlog("ArgStripper Option spelling table");
534     for (const auto &Entry : *Result) {
535       dlog("{0}", Entry.first());
536       RuleCount += Entry.second.size();
537       for (const auto &R : Entry.second)
538         dlog("  {0} #={1} *={2} Mode={3}", R.Text, R.ExactArgs, R.PrefixArgs,
539              int(R.Modes));
540     }
541     dlog("Table spellings={0} rules={1} string-bytes={2}", Result->size(),
542          RuleCount, Result->getAllocator().getBytesAllocated());
543 #endif
544     // The static table will never be destroyed.
545     return Result.release();
546   }();
547 
548   auto It = Table->find(Arg);
549   return (It == Table->end()) ? llvm::ArrayRef<Rule>() : It->second;
550 }
551 
strip(llvm::StringRef Arg)552 void ArgStripper::strip(llvm::StringRef Arg) {
553   auto OptionRules = rulesFor(Arg);
554   if (OptionRules.empty()) {
555     // Not a recognized flag. Strip it literally.
556     Storage.emplace_back(Arg);
557     Rules.emplace_back();
558     Rules.back().Text = Storage.back();
559     Rules.back().ExactArgs = 1;
560     if (Rules.back().Text.consume_back("*"))
561       Rules.back().PrefixArgs = 1;
562     Rules.back().Modes = DM_All;
563     Rules.back().Priority = -1; // Max unsigned = lowest priority.
564   } else {
565     Rules.append(OptionRules.begin(), OptionRules.end());
566   }
567 }
568 
matchingRule(llvm::StringRef Arg,unsigned Mode,unsigned & ArgCount) const569 const ArgStripper::Rule *ArgStripper::matchingRule(llvm::StringRef Arg,
570                                                    unsigned Mode,
571                                                    unsigned &ArgCount) const {
572   const ArgStripper::Rule *BestRule = nullptr;
573   for (const Rule &R : Rules) {
574     // Rule can fail to match if...
575     if (!(R.Modes & Mode))
576       continue; // not applicable to current driver mode
577     if (BestRule && BestRule->Priority < R.Priority)
578       continue; // lower-priority than best candidate.
579     if (!Arg.startswith(R.Text))
580       continue; // current arg doesn't match the prefix string
581     bool PrefixMatch = Arg.size() > R.Text.size();
582     // Can rule apply as an exact/prefix match?
583     if (unsigned Count = PrefixMatch ? R.PrefixArgs : R.ExactArgs) {
584       BestRule = &R;
585       ArgCount = Count;
586     }
587     // Continue in case we find a higher-priority rule.
588   }
589   return BestRule;
590 }
591 
process(std::vector<std::string> & Args) const592 void ArgStripper::process(std::vector<std::string> &Args) const {
593   if (Args.empty())
594     return;
595 
596   // We're parsing the args list in some mode (e.g. gcc-compatible) but may
597   // temporarily switch to another mode with the -Xclang flag.
598   DriverMode MainMode = getDriverMode(Args);
599   DriverMode CurrentMode = MainMode;
600 
601   // Read and write heads for in-place deletion.
602   unsigned Read = 0, Write = 0;
603   bool WasXclang = false;
604   while (Read < Args.size()) {
605     unsigned ArgCount = 0;
606     if (matchingRule(Args[Read], CurrentMode, ArgCount)) {
607       // Delete it and its args.
608       if (WasXclang) {
609         assert(Write > 0);
610         --Write; // Drop previous -Xclang arg
611         CurrentMode = MainMode;
612         WasXclang = false;
613       }
614       // Advance to last arg. An arg may be foo or -Xclang foo.
615       for (unsigned I = 1; Read < Args.size() && I < ArgCount; ++I) {
616         ++Read;
617         if (Read < Args.size() && Args[Read] == "-Xclang")
618           ++Read;
619       }
620     } else {
621       // No match, just copy the arg through.
622       WasXclang = Args[Read] == "-Xclang";
623       CurrentMode = WasXclang ? DM_CC1 : MainMode;
624       if (Write != Read)
625         Args[Write] = std::move(Args[Read]);
626       ++Write;
627     }
628     ++Read;
629   }
630   Args.resize(Write);
631 }
632 
printArgv(llvm::ArrayRef<llvm::StringRef> Args)633 std::string printArgv(llvm::ArrayRef<llvm::StringRef> Args) {
634   std::string Buf;
635   llvm::raw_string_ostream OS(Buf);
636   bool Sep = false;
637   for (llvm::StringRef Arg : Args) {
638     if (Sep)
639       OS << ' ';
640     Sep = true;
641     if (llvm::all_of(Arg, llvm::isPrint) &&
642         Arg.find_first_of(" \t\n\"\\") == llvm::StringRef::npos) {
643       OS << Arg;
644       continue;
645     }
646     OS << '"';
647     OS.write_escaped(Arg, /*UseHexEscapes=*/true);
648     OS << '"';
649   }
650   return std::move(OS.str());
651 }
652 
printArgv(llvm::ArrayRef<std::string> Args)653 std::string printArgv(llvm::ArrayRef<std::string> Args) {
654   std::vector<llvm::StringRef> Refs(Args.size());
655   llvm::copy(Args, Refs.begin());
656   return printArgv(Refs);
657 }
658 
659 } // namespace clangd
660 } // namespace clang
661