1 //===-- ClangExpressionParser.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 "clang/AST/ASTContext.h"
10 #include "clang/AST/ASTDiagnostic.h"
11 #include "clang/AST/ExternalASTSource.h"
12 #include "clang/AST/PrettyPrinter.h"
13 #include "clang/Basic/Builtins.h"
14 #include "clang/Basic/DiagnosticIDs.h"
15 #include "clang/Basic/SourceLocation.h"
16 #include "clang/Basic/TargetInfo.h"
17 #include "clang/Basic/Version.h"
18 #include "clang/CodeGen/CodeGenAction.h"
19 #include "clang/CodeGen/ModuleBuilder.h"
20 #include "clang/Edit/Commit.h"
21 #include "clang/Edit/EditedSource.h"
22 #include "clang/Edit/EditsReceiver.h"
23 #include "clang/Frontend/CompilerInstance.h"
24 #include "clang/Frontend/CompilerInvocation.h"
25 #include "clang/Frontend/FrontendActions.h"
26 #include "clang/Frontend/FrontendDiagnostic.h"
27 #include "clang/Frontend/FrontendPluginRegistry.h"
28 #include "clang/Frontend/TextDiagnosticBuffer.h"
29 #include "clang/Frontend/TextDiagnosticPrinter.h"
30 #include "clang/Lex/Preprocessor.h"
31 #include "clang/Parse/ParseAST.h"
32 #include "clang/Rewrite/Core/Rewriter.h"
33 #include "clang/Rewrite/Frontend/FrontendActions.h"
34 #include "clang/Sema/CodeCompleteConsumer.h"
35 #include "clang/Sema/Sema.h"
36 #include "clang/Sema/SemaConsumer.h"
37 
38 #include "llvm/ADT/StringRef.h"
39 #include "llvm/ExecutionEngine/ExecutionEngine.h"
40 #include "llvm/Support/CrashRecoveryContext.h"
41 #include "llvm/Support/Debug.h"
42 #include "llvm/Support/FileSystem.h"
43 #include "llvm/Support/TargetSelect.h"
44 
45 #include "llvm/IR/LLVMContext.h"
46 #include "llvm/IR/Module.h"
47 #include "llvm/Support/DynamicLibrary.h"
48 #include "llvm/Support/ErrorHandling.h"
49 #include "llvm/Support/Host.h"
50 #include "llvm/Support/MemoryBuffer.h"
51 #include "llvm/Support/Signals.h"
52 
53 #include "ClangDiagnostic.h"
54 #include "ClangExpressionParser.h"
55 #include "ClangUserExpression.h"
56 
57 #include "ASTUtils.h"
58 #include "ClangASTSource.h"
59 #include "ClangDiagnostic.h"
60 #include "ClangExpressionDeclMap.h"
61 #include "ClangExpressionHelper.h"
62 #include "ClangExpressionParser.h"
63 #include "ClangHost.h"
64 #include "ClangModulesDeclVendor.h"
65 #include "ClangPersistentVariables.h"
66 #include "IRDynamicChecks.h"
67 #include "IRForTarget.h"
68 #include "ModuleDependencyCollector.h"
69 
70 #include "Plugins/TypeSystem/Clang/TypeSystemClang.h"
71 #include "lldb/Core/Debugger.h"
72 #include "lldb/Core/Disassembler.h"
73 #include "lldb/Core/Module.h"
74 #include "lldb/Core/StreamFile.h"
75 #include "lldb/Expression/IRExecutionUnit.h"
76 #include "lldb/Expression/IRInterpreter.h"
77 #include "lldb/Host/File.h"
78 #include "lldb/Host/HostInfo.h"
79 #include "lldb/Symbol/SymbolVendor.h"
80 #include "lldb/Target/ExecutionContext.h"
81 #include "lldb/Target/Language.h"
82 #include "lldb/Target/Process.h"
83 #include "lldb/Target/Target.h"
84 #include "lldb/Target/ThreadPlanCallFunction.h"
85 #include "lldb/Utility/DataBufferHeap.h"
86 #include "lldb/Utility/LLDBAssert.h"
87 #include "lldb/Utility/LLDBLog.h"
88 #include "lldb/Utility/Log.h"
89 #include "lldb/Utility/ReproducerProvider.h"
90 #include "lldb/Utility/Stream.h"
91 #include "lldb/Utility/StreamString.h"
92 #include "lldb/Utility/StringList.h"
93 
94 #include "Plugins/LanguageRuntime/ObjC/ObjCLanguageRuntime.h"
95 #include "Plugins/LanguageRuntime/RenderScript/RenderScriptRuntime/RenderScriptRuntime.h"
96 
97 #include <cctype>
98 #include <memory>
99 
100 using namespace clang;
101 using namespace llvm;
102 using namespace lldb_private;
103 
104 //===----------------------------------------------------------------------===//
105 // Utility Methods for Clang
106 //===----------------------------------------------------------------------===//
107 
108 class ClangExpressionParser::LLDBPreprocessorCallbacks : public PPCallbacks {
109   ClangModulesDeclVendor &m_decl_vendor;
110   ClangPersistentVariables &m_persistent_vars;
111   clang::SourceManager &m_source_mgr;
112   StreamString m_error_stream;
113   bool m_has_errors = false;
114 
115 public:
116   LLDBPreprocessorCallbacks(ClangModulesDeclVendor &decl_vendor,
117                             ClangPersistentVariables &persistent_vars,
118                             clang::SourceManager &source_mgr)
119       : m_decl_vendor(decl_vendor), m_persistent_vars(persistent_vars),
120         m_source_mgr(source_mgr) {}
121 
122   void moduleImport(SourceLocation import_location, clang::ModuleIdPath path,
123                     const clang::Module * /*null*/) override {
124     // Ignore modules that are imported in the wrapper code as these are not
125     // loaded by the user.
126     llvm::StringRef filename =
127         m_source_mgr.getPresumedLoc(import_location).getFilename();
128     if (filename == ClangExpressionSourceCode::g_prefix_file_name)
129       return;
130 
131     SourceModule module;
132 
133     for (const std::pair<IdentifierInfo *, SourceLocation> &component : path)
134       module.path.push_back(ConstString(component.first->getName()));
135 
136     StreamString error_stream;
137 
138     ClangModulesDeclVendor::ModuleVector exported_modules;
139     if (!m_decl_vendor.AddModule(module, &exported_modules, m_error_stream))
140       m_has_errors = true;
141 
142     for (ClangModulesDeclVendor::ModuleID module : exported_modules)
143       m_persistent_vars.AddHandLoadedClangModule(module);
144   }
145 
146   bool hasErrors() { return m_has_errors; }
147 
148   llvm::StringRef getErrorString() { return m_error_stream.GetString(); }
149 };
150 
151 static void AddAllFixIts(ClangDiagnostic *diag, const clang::Diagnostic &Info) {
152   for (auto &fix_it : Info.getFixItHints()) {
153     if (fix_it.isNull())
154       continue;
155     diag->AddFixitHint(fix_it);
156   }
157 }
158 
159 class ClangDiagnosticManagerAdapter : public clang::DiagnosticConsumer {
160 public:
161   ClangDiagnosticManagerAdapter(DiagnosticOptions &opts) {
162     DiagnosticOptions *options = new DiagnosticOptions(opts);
163     options->ShowPresumedLoc = true;
164     options->ShowLevel = false;
165     m_os = std::make_shared<llvm::raw_string_ostream>(m_output);
166     m_passthrough =
167         std::make_shared<clang::TextDiagnosticPrinter>(*m_os, options);
168   }
169 
170   void ResetManager(DiagnosticManager *manager = nullptr) {
171     m_manager = manager;
172   }
173 
174   /// Returns the last ClangDiagnostic message that the DiagnosticManager
175   /// received or a nullptr if the DiagnosticMangager hasn't seen any
176   /// Clang diagnostics yet.
177   ClangDiagnostic *MaybeGetLastClangDiag() const {
178     if (m_manager->Diagnostics().empty())
179       return nullptr;
180     lldb_private::Diagnostic *diag = m_manager->Diagnostics().back().get();
181     ClangDiagnostic *clang_diag = dyn_cast<ClangDiagnostic>(diag);
182     return clang_diag;
183   }
184 
185   void HandleDiagnostic(DiagnosticsEngine::Level DiagLevel,
186                         const clang::Diagnostic &Info) override {
187     if (!m_manager) {
188       // We have no DiagnosticManager before/after parsing but we still could
189       // receive diagnostics (e.g., by the ASTImporter failing to copy decls
190       // when we move the expression result ot the ScratchASTContext). Let's at
191       // least log these diagnostics until we find a way to properly render
192       // them and display them to the user.
193       Log *log = GetLog(LLDBLog::Expressions);
194       if (log) {
195         llvm::SmallVector<char, 32> diag_str;
196         Info.FormatDiagnostic(diag_str);
197         diag_str.push_back('\0');
198         const char *plain_diag = diag_str.data();
199         LLDB_LOG(log, "Received diagnostic outside parsing: {0}", plain_diag);
200       }
201       return;
202     }
203 
204     // Update error/warning counters.
205     DiagnosticConsumer::HandleDiagnostic(DiagLevel, Info);
206 
207     // Render diagnostic message to m_output.
208     m_output.clear();
209     m_passthrough->HandleDiagnostic(DiagLevel, Info);
210     m_os->flush();
211 
212     lldb_private::DiagnosticSeverity severity;
213     bool make_new_diagnostic = true;
214 
215     switch (DiagLevel) {
216     case DiagnosticsEngine::Level::Fatal:
217     case DiagnosticsEngine::Level::Error:
218       severity = eDiagnosticSeverityError;
219       break;
220     case DiagnosticsEngine::Level::Warning:
221       severity = eDiagnosticSeverityWarning;
222       break;
223     case DiagnosticsEngine::Level::Remark:
224     case DiagnosticsEngine::Level::Ignored:
225       severity = eDiagnosticSeverityRemark;
226       break;
227     case DiagnosticsEngine::Level::Note:
228       m_manager->AppendMessageToDiagnostic(m_output);
229       make_new_diagnostic = false;
230 
231       // 'note:' diagnostics for errors and warnings can also contain Fix-Its.
232       // We add these Fix-Its to the last error diagnostic to make sure
233       // that we later have all Fix-Its related to an 'error' diagnostic when
234       // we apply them to the user expression.
235       auto *clang_diag = MaybeGetLastClangDiag();
236       // If we don't have a previous diagnostic there is nothing to do.
237       // If the previous diagnostic already has its own Fix-Its, assume that
238       // the 'note:' Fix-It is just an alternative way to solve the issue and
239       // ignore these Fix-Its.
240       if (!clang_diag || clang_diag->HasFixIts())
241         break;
242       // Ignore all Fix-Its that are not associated with an error.
243       if (clang_diag->GetSeverity() != eDiagnosticSeverityError)
244         break;
245       AddAllFixIts(clang_diag, Info);
246       break;
247     }
248     if (make_new_diagnostic) {
249       // ClangDiagnostic messages are expected to have no whitespace/newlines
250       // around them.
251       std::string stripped_output =
252           std::string(llvm::StringRef(m_output).trim());
253 
254       auto new_diagnostic = std::make_unique<ClangDiagnostic>(
255           stripped_output, severity, Info.getID());
256 
257       // Don't store away warning fixits, since the compiler doesn't have
258       // enough context in an expression for the warning to be useful.
259       // FIXME: Should we try to filter out FixIts that apply to our generated
260       // code, and not the user's expression?
261       if (severity == eDiagnosticSeverityError)
262         AddAllFixIts(new_diagnostic.get(), Info);
263 
264       m_manager->AddDiagnostic(std::move(new_diagnostic));
265     }
266   }
267 
268   void BeginSourceFile(const LangOptions &LO, const Preprocessor *PP) override {
269     m_passthrough->BeginSourceFile(LO, PP);
270   }
271 
272   void EndSourceFile() override { m_passthrough->EndSourceFile(); }
273 
274 private:
275   DiagnosticManager *m_manager = nullptr;
276   std::shared_ptr<clang::TextDiagnosticPrinter> m_passthrough;
277   /// Output stream of m_passthrough.
278   std::shared_ptr<llvm::raw_string_ostream> m_os;
279   /// Output string filled by m_os.
280   std::string m_output;
281 };
282 
283 static void SetupModuleHeaderPaths(CompilerInstance *compiler,
284                                    std::vector<std::string> include_directories,
285                                    lldb::TargetSP target_sp) {
286   Log *log = GetLog(LLDBLog::Expressions);
287 
288   HeaderSearchOptions &search_opts = compiler->getHeaderSearchOpts();
289 
290   for (const std::string &dir : include_directories) {
291     search_opts.AddPath(dir, frontend::System, false, true);
292     LLDB_LOG(log, "Added user include dir: {0}", dir);
293   }
294 
295   llvm::SmallString<128> module_cache;
296   const auto &props = ModuleList::GetGlobalModuleListProperties();
297   props.GetClangModulesCachePath().GetPath(module_cache);
298   search_opts.ModuleCachePath = std::string(module_cache.str());
299   LLDB_LOG(log, "Using module cache path: {0}", module_cache.c_str());
300 
301   search_opts.ResourceDir = GetClangResourceDir().GetPath();
302 
303   search_opts.ImplicitModuleMaps = true;
304 }
305 
306 /// Iff the given identifier is a C++ keyword, remove it from the
307 /// identifier table (i.e., make the token a normal identifier).
308 static void RemoveCppKeyword(IdentifierTable &idents, llvm::StringRef token) {
309   // FIXME: 'using' is used by LLDB for local variables, so we can't remove
310   // this keyword without breaking this functionality.
311   if (token == "using")
312     return;
313   // GCC's '__null' is used by LLDB to define NULL/Nil/nil.
314   if (token == "__null")
315     return;
316 
317   LangOptions cpp_lang_opts;
318   cpp_lang_opts.CPlusPlus = true;
319   cpp_lang_opts.CPlusPlus11 = true;
320   cpp_lang_opts.CPlusPlus20 = true;
321 
322   clang::IdentifierInfo &ii = idents.get(token);
323   // The identifier has to be a C++-exclusive keyword. if not, then there is
324   // nothing to do.
325   if (!ii.isCPlusPlusKeyword(cpp_lang_opts))
326     return;
327   // If the token is already an identifier, then there is nothing to do.
328   if (ii.getTokenID() == clang::tok::identifier)
329     return;
330   // Otherwise the token is a C++ keyword, so turn it back into a normal
331   // identifier.
332   ii.revertTokenIDToIdentifier();
333 }
334 
335 /// Remove all C++ keywords from the given identifier table.
336 static void RemoveAllCppKeywords(IdentifierTable &idents) {
337 #define KEYWORD(NAME, FLAGS) RemoveCppKeyword(idents, llvm::StringRef(#NAME));
338 #include "clang/Basic/TokenKinds.def"
339 }
340 
341 /// Configures Clang diagnostics for the expression parser.
342 static void SetupDefaultClangDiagnostics(CompilerInstance &compiler) {
343   // List of Clang warning groups that are not useful when parsing expressions.
344   const std::vector<const char *> groupsToIgnore = {
345       "unused-value",
346       "odr",
347       "unused-getter-return-value",
348   };
349   for (const char *group : groupsToIgnore) {
350     compiler.getDiagnostics().setSeverityForGroup(
351         clang::diag::Flavor::WarningOrError, group,
352         clang::diag::Severity::Ignored, SourceLocation());
353   }
354 }
355 
356 //===----------------------------------------------------------------------===//
357 // Implementation of ClangExpressionParser
358 //===----------------------------------------------------------------------===//
359 
360 ClangExpressionParser::ClangExpressionParser(
361     ExecutionContextScope *exe_scope, Expression &expr,
362     bool generate_debug_info, std::vector<std::string> include_directories,
363     std::string filename)
364     : ExpressionParser(exe_scope, expr, generate_debug_info), m_compiler(),
365       m_pp_callbacks(nullptr),
366       m_include_directories(std::move(include_directories)),
367       m_filename(std::move(filename)) {
368   Log *log = GetLog(LLDBLog::Expressions);
369 
370   // We can't compile expressions without a target.  So if the exe_scope is
371   // null or doesn't have a target, then we just need to get out of here.  I'll
372   // lldbassert and not make any of the compiler objects since
373   // I can't return errors directly from the constructor.  Further calls will
374   // check if the compiler was made and
375   // bag out if it wasn't.
376 
377   if (!exe_scope) {
378     lldbassert(exe_scope &&
379                "Can't make an expression parser with a null scope.");
380     return;
381   }
382 
383   lldb::TargetSP target_sp;
384   target_sp = exe_scope->CalculateTarget();
385   if (!target_sp) {
386     lldbassert(target_sp.get() &&
387                "Can't make an expression parser with a null target.");
388     return;
389   }
390 
391   // 1. Create a new compiler instance.
392   m_compiler = std::make_unique<CompilerInstance>();
393 
394   // When capturing a reproducer, hook up the file collector with clang to
395   // collector modules and headers.
396   if (repro::Generator *g = repro::Reproducer::Instance().GetGenerator()) {
397     repro::FileProvider &fp = g->GetOrCreate<repro::FileProvider>();
398     m_compiler->setModuleDepCollector(
399         std::make_shared<ModuleDependencyCollectorAdaptor>(
400             fp.GetFileCollector()));
401     DependencyOutputOptions &opts = m_compiler->getDependencyOutputOpts();
402     opts.IncludeSystemHeaders = true;
403     opts.IncludeModuleFiles = true;
404   }
405 
406   // Make sure clang uses the same VFS as LLDB.
407   m_compiler->createFileManager(FileSystem::Instance().GetVirtualFileSystem());
408 
409   lldb::LanguageType frame_lang =
410       expr.Language(); // defaults to lldb::eLanguageTypeUnknown
411   bool overridden_target_opts = false;
412   lldb_private::LanguageRuntime *lang_rt = nullptr;
413 
414   std::string abi;
415   ArchSpec target_arch;
416   target_arch = target_sp->GetArchitecture();
417 
418   const auto target_machine = target_arch.GetMachine();
419 
420   // If the expression is being evaluated in the context of an existing stack
421   // frame, we introspect to see if the language runtime is available.
422 
423   lldb::StackFrameSP frame_sp = exe_scope->CalculateStackFrame();
424   lldb::ProcessSP process_sp = exe_scope->CalculateProcess();
425 
426   // Make sure the user hasn't provided a preferred execution language with
427   // `expression --language X -- ...`
428   if (frame_sp && frame_lang == lldb::eLanguageTypeUnknown)
429     frame_lang = frame_sp->GetLanguage();
430 
431   if (process_sp && frame_lang != lldb::eLanguageTypeUnknown) {
432     lang_rt = process_sp->GetLanguageRuntime(frame_lang);
433     LLDB_LOGF(log, "Frame has language of type %s",
434               Language::GetNameForLanguageType(frame_lang));
435   }
436 
437   // 2. Configure the compiler with a set of default options that are
438   // appropriate for most situations.
439   if (target_arch.IsValid()) {
440     std::string triple = target_arch.GetTriple().str();
441     m_compiler->getTargetOpts().Triple = triple;
442     LLDB_LOGF(log, "Using %s as the target triple",
443               m_compiler->getTargetOpts().Triple.c_str());
444   } else {
445     // If we get here we don't have a valid target and just have to guess.
446     // Sometimes this will be ok to just use the host target triple (when we
447     // evaluate say "2+3", but other expressions like breakpoint conditions and
448     // other things that _are_ target specific really shouldn't just be using
449     // the host triple. In such a case the language runtime should expose an
450     // overridden options set (3), below.
451     m_compiler->getTargetOpts().Triple = llvm::sys::getDefaultTargetTriple();
452     LLDB_LOGF(log, "Using default target triple of %s",
453               m_compiler->getTargetOpts().Triple.c_str());
454   }
455   // Now add some special fixes for known architectures: Any arm32 iOS
456   // environment, but not on arm64
457   if (m_compiler->getTargetOpts().Triple.find("arm64") == std::string::npos &&
458       m_compiler->getTargetOpts().Triple.find("arm") != std::string::npos &&
459       m_compiler->getTargetOpts().Triple.find("ios") != std::string::npos) {
460     m_compiler->getTargetOpts().ABI = "apcs-gnu";
461   }
462   // Supported subsets of x86
463   if (target_machine == llvm::Triple::x86 ||
464       target_machine == llvm::Triple::x86_64) {
465     m_compiler->getTargetOpts().Features.push_back("+sse");
466     m_compiler->getTargetOpts().Features.push_back("+sse2");
467   }
468 
469   // Set the target CPU to generate code for. This will be empty for any CPU
470   // that doesn't really need to make a special
471   // CPU string.
472   m_compiler->getTargetOpts().CPU = target_arch.GetClangTargetCPU();
473 
474   // Set the target ABI
475   abi = GetClangTargetABI(target_arch);
476   if (!abi.empty())
477     m_compiler->getTargetOpts().ABI = abi;
478 
479   // 3. Now allow the runtime to provide custom configuration options for the
480   // target. In this case, a specialized language runtime is available and we
481   // can query it for extra options. For 99% of use cases, this will not be
482   // needed and should be provided when basic platform detection is not enough.
483   // FIXME: Generalize this. Only RenderScriptRuntime currently supports this
484   // currently. Hardcoding this isn't ideal but it's better than LanguageRuntime
485   // having knowledge of clang::TargetOpts.
486   if (auto *renderscript_rt =
487           llvm::dyn_cast_or_null<RenderScriptRuntime>(lang_rt))
488     overridden_target_opts =
489         renderscript_rt->GetOverrideExprOptions(m_compiler->getTargetOpts());
490 
491   if (overridden_target_opts)
492     if (log && log->GetVerbose()) {
493       LLDB_LOGV(
494           log, "Using overridden target options for the expression evaluation");
495 
496       auto opts = m_compiler->getTargetOpts();
497       LLDB_LOGV(log, "Triple: '{0}'", opts.Triple);
498       LLDB_LOGV(log, "CPU: '{0}'", opts.CPU);
499       LLDB_LOGV(log, "FPMath: '{0}'", opts.FPMath);
500       LLDB_LOGV(log, "ABI: '{0}'", opts.ABI);
501       LLDB_LOGV(log, "LinkerVersion: '{0}'", opts.LinkerVersion);
502       StringList::LogDump(log, opts.FeaturesAsWritten, "FeaturesAsWritten");
503       StringList::LogDump(log, opts.Features, "Features");
504     }
505 
506   // 4. Create and install the target on the compiler.
507   m_compiler->createDiagnostics();
508   // Limit the number of error diagnostics we emit.
509   // A value of 0 means no limit for both LLDB and Clang.
510   m_compiler->getDiagnostics().setErrorLimit(target_sp->GetExprErrorLimit());
511 
512   auto target_info = TargetInfo::CreateTargetInfo(
513       m_compiler->getDiagnostics(), m_compiler->getInvocation().TargetOpts);
514   if (log) {
515     LLDB_LOGF(log, "Using SIMD alignment: %d",
516               target_info->getSimdDefaultAlign());
517     LLDB_LOGF(log, "Target datalayout string: '%s'",
518               target_info->getDataLayoutString());
519     LLDB_LOGF(log, "Target ABI: '%s'", target_info->getABI().str().c_str());
520     LLDB_LOGF(log, "Target vector alignment: %d",
521               target_info->getMaxVectorAlign());
522   }
523   m_compiler->setTarget(target_info);
524 
525   assert(m_compiler->hasTarget());
526 
527   // 5. Set language options.
528   lldb::LanguageType language = expr.Language();
529   LangOptions &lang_opts = m_compiler->getLangOpts();
530 
531   switch (language) {
532   case lldb::eLanguageTypeC:
533   case lldb::eLanguageTypeC89:
534   case lldb::eLanguageTypeC99:
535   case lldb::eLanguageTypeC11:
536     // FIXME: the following language option is a temporary workaround,
537     // to "ask for C, get C++."
538     // For now, the expression parser must use C++ anytime the language is a C
539     // family language, because the expression parser uses features of C++ to
540     // capture values.
541     lang_opts.CPlusPlus = true;
542     break;
543   case lldb::eLanguageTypeObjC:
544     lang_opts.ObjC = true;
545     // FIXME: the following language option is a temporary workaround,
546     // to "ask for ObjC, get ObjC++" (see comment above).
547     lang_opts.CPlusPlus = true;
548 
549     // Clang now sets as default C++14 as the default standard (with
550     // GNU extensions), so we do the same here to avoid mismatches that
551     // cause compiler error when evaluating expressions (e.g. nullptr not found
552     // as it's a C++11 feature). Currently lldb evaluates C++14 as C++11 (see
553     // two lines below) so we decide to be consistent with that, but this could
554     // be re-evaluated in the future.
555     lang_opts.CPlusPlus11 = true;
556     break;
557   case lldb::eLanguageTypeC_plus_plus:
558   case lldb::eLanguageTypeC_plus_plus_11:
559   case lldb::eLanguageTypeC_plus_plus_14:
560     lang_opts.CPlusPlus11 = true;
561     m_compiler->getHeaderSearchOpts().UseLibcxx = true;
562     LLVM_FALLTHROUGH;
563   case lldb::eLanguageTypeC_plus_plus_03:
564     lang_opts.CPlusPlus = true;
565     if (process_sp)
566       lang_opts.ObjC =
567           process_sp->GetLanguageRuntime(lldb::eLanguageTypeObjC) != nullptr;
568     break;
569   case lldb::eLanguageTypeObjC_plus_plus:
570   case lldb::eLanguageTypeUnknown:
571   default:
572     lang_opts.ObjC = true;
573     lang_opts.CPlusPlus = true;
574     lang_opts.CPlusPlus11 = true;
575     m_compiler->getHeaderSearchOpts().UseLibcxx = true;
576     break;
577   }
578 
579   lang_opts.Bool = true;
580   lang_opts.WChar = true;
581   lang_opts.Blocks = true;
582   lang_opts.DebuggerSupport =
583       true; // Features specifically for debugger clients
584   if (expr.DesiredResultType() == Expression::eResultTypeId)
585     lang_opts.DebuggerCastResultToId = true;
586 
587   lang_opts.CharIsSigned = ArchSpec(m_compiler->getTargetOpts().Triple.c_str())
588                                .CharIsSignedByDefault();
589 
590   // Spell checking is a nice feature, but it ends up completing a lot of types
591   // that we didn't strictly speaking need to complete. As a result, we spend a
592   // long time parsing and importing debug information.
593   lang_opts.SpellChecking = false;
594 
595   auto *clang_expr = dyn_cast<ClangUserExpression>(&m_expr);
596   if (clang_expr && clang_expr->DidImportCxxModules()) {
597     LLDB_LOG(log, "Adding lang options for importing C++ modules");
598 
599     lang_opts.Modules = true;
600     // We want to implicitly build modules.
601     lang_opts.ImplicitModules = true;
602     // To automatically import all submodules when we import 'std'.
603     lang_opts.ModulesLocalVisibility = false;
604 
605     // We use the @import statements, so we need this:
606     // FIXME: We could use the modules-ts, but that currently doesn't work.
607     lang_opts.ObjC = true;
608 
609     // Options we need to parse libc++ code successfully.
610     // FIXME: We should ask the driver for the appropriate default flags.
611     lang_opts.GNUMode = true;
612     lang_opts.GNUKeywords = true;
613     lang_opts.DoubleSquareBracketAttributes = true;
614     lang_opts.CPlusPlus11 = true;
615 
616     // The Darwin libc expects this macro to be set.
617     lang_opts.GNUCVersion = 40201;
618 
619     SetupModuleHeaderPaths(m_compiler.get(), m_include_directories,
620                            target_sp);
621   }
622 
623   if (process_sp && lang_opts.ObjC) {
624     if (auto *runtime = ObjCLanguageRuntime::Get(*process_sp)) {
625       if (runtime->GetRuntimeVersion() ==
626           ObjCLanguageRuntime::ObjCRuntimeVersions::eAppleObjC_V2)
627         lang_opts.ObjCRuntime.set(ObjCRuntime::MacOSX, VersionTuple(10, 7));
628       else
629         lang_opts.ObjCRuntime.set(ObjCRuntime::FragileMacOSX,
630                                   VersionTuple(10, 7));
631 
632       if (runtime->HasNewLiteralsAndIndexing())
633         lang_opts.DebuggerObjCLiteral = true;
634     }
635   }
636 
637   lang_opts.ThreadsafeStatics = false;
638   lang_opts.AccessControl = false; // Debuggers get universal access
639   lang_opts.DollarIdents = true;   // $ indicates a persistent variable name
640   // We enable all builtin functions beside the builtins from libc/libm (e.g.
641   // 'fopen'). Those libc functions are already correctly handled by LLDB, and
642   // additionally enabling them as expandable builtins is breaking Clang.
643   lang_opts.NoBuiltin = true;
644 
645   // Set CodeGen options
646   m_compiler->getCodeGenOpts().EmitDeclMetadata = true;
647   m_compiler->getCodeGenOpts().InstrumentFunctions = false;
648   m_compiler->getCodeGenOpts().setFramePointer(
649                                     CodeGenOptions::FramePointerKind::All);
650   if (generate_debug_info)
651     m_compiler->getCodeGenOpts().setDebugInfo(codegenoptions::FullDebugInfo);
652   else
653     m_compiler->getCodeGenOpts().setDebugInfo(codegenoptions::NoDebugInfo);
654 
655   // Disable some warnings.
656   SetupDefaultClangDiagnostics(*m_compiler);
657 
658   // Inform the target of the language options
659   //
660   // FIXME: We shouldn't need to do this, the target should be immutable once
661   // created. This complexity should be lifted elsewhere.
662   m_compiler->getTarget().adjust(m_compiler->getDiagnostics(),
663 		                 m_compiler->getLangOpts());
664 
665   // 6. Set up the diagnostic buffer for reporting errors
666 
667   auto diag_mgr = new ClangDiagnosticManagerAdapter(
668       m_compiler->getDiagnostics().getDiagnosticOptions());
669   m_compiler->getDiagnostics().setClient(diag_mgr);
670 
671   // 7. Set up the source management objects inside the compiler
672   m_compiler->createFileManager();
673   if (!m_compiler->hasSourceManager())
674     m_compiler->createSourceManager(m_compiler->getFileManager());
675   m_compiler->createPreprocessor(TU_Complete);
676 
677   switch (language) {
678   case lldb::eLanguageTypeC:
679   case lldb::eLanguageTypeC89:
680   case lldb::eLanguageTypeC99:
681   case lldb::eLanguageTypeC11:
682   case lldb::eLanguageTypeObjC:
683     // This is not a C++ expression but we enabled C++ as explained above.
684     // Remove all C++ keywords from the PP so that the user can still use
685     // variables that have C++ keywords as names (e.g. 'int template;').
686     RemoveAllCppKeywords(m_compiler->getPreprocessor().getIdentifierTable());
687     break;
688   default:
689     break;
690   }
691 
692   if (auto *clang_persistent_vars = llvm::cast<ClangPersistentVariables>(
693           target_sp->GetPersistentExpressionStateForLanguage(
694               lldb::eLanguageTypeC))) {
695     if (std::shared_ptr<ClangModulesDeclVendor> decl_vendor =
696             clang_persistent_vars->GetClangModulesDeclVendor()) {
697       std::unique_ptr<PPCallbacks> pp_callbacks(
698           new LLDBPreprocessorCallbacks(*decl_vendor, *clang_persistent_vars,
699                                         m_compiler->getSourceManager()));
700       m_pp_callbacks =
701           static_cast<LLDBPreprocessorCallbacks *>(pp_callbacks.get());
702       m_compiler->getPreprocessor().addPPCallbacks(std::move(pp_callbacks));
703     }
704   }
705 
706   // 8. Most of this we get from the CompilerInstance, but we also want to give
707   // the context an ExternalASTSource.
708 
709   auto &PP = m_compiler->getPreprocessor();
710   auto &builtin_context = PP.getBuiltinInfo();
711   builtin_context.initializeBuiltins(PP.getIdentifierTable(),
712                                      m_compiler->getLangOpts());
713 
714   m_compiler->createASTContext();
715   clang::ASTContext &ast_context = m_compiler->getASTContext();
716 
717   m_ast_context = std::make_unique<TypeSystemClang>(
718       "Expression ASTContext for '" + m_filename + "'", ast_context);
719 
720   std::string module_name("$__lldb_module");
721 
722   m_llvm_context = std::make_unique<LLVMContext>();
723   m_code_generator.reset(CreateLLVMCodeGen(
724       m_compiler->getDiagnostics(), module_name,
725       &m_compiler->getVirtualFileSystem(), m_compiler->getHeaderSearchOpts(),
726       m_compiler->getPreprocessorOpts(), m_compiler->getCodeGenOpts(),
727       *m_llvm_context));
728 }
729 
730 ClangExpressionParser::~ClangExpressionParser() = default;
731 
732 namespace {
733 
734 /// \class CodeComplete
735 ///
736 /// A code completion consumer for the clang Sema that is responsible for
737 /// creating the completion suggestions when a user requests completion
738 /// of an incomplete `expr` invocation.
739 class CodeComplete : public CodeCompleteConsumer {
740   CodeCompletionTUInfo m_info;
741 
742   std::string m_expr;
743   unsigned m_position = 0;
744   /// The printing policy we use when printing declarations for our completion
745   /// descriptions.
746   clang::PrintingPolicy m_desc_policy;
747 
748   struct CompletionWithPriority {
749     CompletionResult::Completion completion;
750     /// See CodeCompletionResult::Priority;
751     unsigned Priority;
752 
753     /// Establishes a deterministic order in a list of CompletionWithPriority.
754     /// The order returned here is the order in which the completions are
755     /// displayed to the user.
756     bool operator<(const CompletionWithPriority &o) const {
757       // High priority results should come first.
758       if (Priority != o.Priority)
759         return Priority > o.Priority;
760 
761       // Identical priority, so just make sure it's a deterministic order.
762       return completion.GetUniqueKey() < o.completion.GetUniqueKey();
763     }
764   };
765 
766   /// The stored completions.
767   /// Warning: These are in a non-deterministic order until they are sorted
768   /// and returned back to the caller.
769   std::vector<CompletionWithPriority> m_completions;
770 
771   /// Returns true if the given character can be used in an identifier.
772   /// This also returns true for numbers because for completion we usually
773   /// just iterate backwards over iterators.
774   ///
775   /// Note: lldb uses '$' in its internal identifiers, so we also allow this.
776   static bool IsIdChar(char c) {
777     return c == '_' || std::isalnum(c) || c == '$';
778   }
779 
780   /// Returns true if the given character is used to separate arguments
781   /// in the command line of lldb.
782   static bool IsTokenSeparator(char c) { return c == ' ' || c == '\t'; }
783 
784   /// Drops all tokens in front of the expression that are unrelated for
785   /// the completion of the cmd line. 'unrelated' means here that the token
786   /// is not interested for the lldb completion API result.
787   StringRef dropUnrelatedFrontTokens(StringRef cmd) const {
788     if (cmd.empty())
789       return cmd;
790 
791     // If we are at the start of a word, then all tokens are unrelated to
792     // the current completion logic.
793     if (IsTokenSeparator(cmd.back()))
794       return StringRef();
795 
796     // Remove all previous tokens from the string as they are unrelated
797     // to completing the current token.
798     StringRef to_remove = cmd;
799     while (!to_remove.empty() && !IsTokenSeparator(to_remove.back())) {
800       to_remove = to_remove.drop_back();
801     }
802     cmd = cmd.drop_front(to_remove.size());
803 
804     return cmd;
805   }
806 
807   /// Removes the last identifier token from the given cmd line.
808   StringRef removeLastToken(StringRef cmd) const {
809     while (!cmd.empty() && IsIdChar(cmd.back())) {
810       cmd = cmd.drop_back();
811     }
812     return cmd;
813   }
814 
815   /// Attempts to merge the given completion from the given position into the
816   /// existing command. Returns the completion string that can be returned to
817   /// the lldb completion API.
818   std::string mergeCompletion(StringRef existing, unsigned pos,
819                               StringRef completion) const {
820     StringRef existing_command = existing.substr(0, pos);
821     // We rewrite the last token with the completion, so let's drop that
822     // token from the command.
823     existing_command = removeLastToken(existing_command);
824     // We also should remove all previous tokens from the command as they
825     // would otherwise be added to the completion that already has the
826     // completion.
827     existing_command = dropUnrelatedFrontTokens(existing_command);
828     return existing_command.str() + completion.str();
829   }
830 
831 public:
832   /// Constructs a CodeComplete consumer that can be attached to a Sema.
833   ///
834   /// \param[out] expr
835   ///    The whole expression string that we are currently parsing. This
836   ///    string needs to be equal to the input the user typed, and NOT the
837   ///    final code that Clang is parsing.
838   /// \param[out] position
839   ///    The character position of the user cursor in the `expr` parameter.
840   ///
841   CodeComplete(clang::LangOptions ops, std::string expr, unsigned position)
842       : CodeCompleteConsumer(CodeCompleteOptions()),
843         m_info(std::make_shared<GlobalCodeCompletionAllocator>()), m_expr(expr),
844         m_position(position), m_desc_policy(ops) {
845 
846     // Ensure that the printing policy is producing a description that is as
847     // short as possible.
848     m_desc_policy.SuppressScope = true;
849     m_desc_policy.SuppressTagKeyword = true;
850     m_desc_policy.FullyQualifiedName = false;
851     m_desc_policy.TerseOutput = true;
852     m_desc_policy.IncludeNewlines = false;
853     m_desc_policy.UseVoidForZeroParams = false;
854     m_desc_policy.Bool = true;
855   }
856 
857   /// \name Code-completion filtering
858   /// Check if the result should be filtered out.
859   bool isResultFilteredOut(StringRef Filter,
860                            CodeCompletionResult Result) override {
861     // This code is mostly copied from CodeCompleteConsumer.
862     switch (Result.Kind) {
863     case CodeCompletionResult::RK_Declaration:
864       return !(
865           Result.Declaration->getIdentifier() &&
866           Result.Declaration->getIdentifier()->getName().startswith(Filter));
867     case CodeCompletionResult::RK_Keyword:
868       return !StringRef(Result.Keyword).startswith(Filter);
869     case CodeCompletionResult::RK_Macro:
870       return !Result.Macro->getName().startswith(Filter);
871     case CodeCompletionResult::RK_Pattern:
872       return !StringRef(Result.Pattern->getAsString()).startswith(Filter);
873     }
874     // If we trigger this assert or the above switch yields a warning, then
875     // CodeCompletionResult has been enhanced with more kinds of completion
876     // results. Expand the switch above in this case.
877     assert(false && "Unknown completion result type?");
878     // If we reach this, then we should just ignore whatever kind of unknown
879     // result we got back. We probably can't turn it into any kind of useful
880     // completion suggestion with the existing code.
881     return true;
882   }
883 
884 private:
885   /// Generate the completion strings for the given CodeCompletionResult.
886   /// Note that this function has to process results that could come in
887   /// non-deterministic order, so this function should have no side effects.
888   /// To make this easier to enforce, this function and all its parameters
889   /// should always be const-qualified.
890   /// \return Returns llvm::None if no completion should be provided for the
891   ///         given CodeCompletionResult.
892   llvm::Optional<CompletionWithPriority>
893   getCompletionForResult(const CodeCompletionResult &R) const {
894     std::string ToInsert;
895     std::string Description;
896     // Handle the different completion kinds that come from the Sema.
897     switch (R.Kind) {
898     case CodeCompletionResult::RK_Declaration: {
899       const NamedDecl *D = R.Declaration;
900       ToInsert = R.Declaration->getNameAsString();
901       // If we have a function decl that has no arguments we want to
902       // complete the empty parantheses for the user. If the function has
903       // arguments, we at least complete the opening bracket.
904       if (const FunctionDecl *F = dyn_cast<FunctionDecl>(D)) {
905         if (F->getNumParams() == 0)
906           ToInsert += "()";
907         else
908           ToInsert += "(";
909         raw_string_ostream OS(Description);
910         F->print(OS, m_desc_policy, false);
911         OS.flush();
912       } else if (const VarDecl *V = dyn_cast<VarDecl>(D)) {
913         Description = V->getType().getAsString(m_desc_policy);
914       } else if (const FieldDecl *F = dyn_cast<FieldDecl>(D)) {
915         Description = F->getType().getAsString(m_desc_policy);
916       } else if (const NamespaceDecl *N = dyn_cast<NamespaceDecl>(D)) {
917         // If we try to complete a namespace, then we can directly append
918         // the '::'.
919         if (!N->isAnonymousNamespace())
920           ToInsert += "::";
921       }
922       break;
923     }
924     case CodeCompletionResult::RK_Keyword:
925       ToInsert = R.Keyword;
926       break;
927     case CodeCompletionResult::RK_Macro:
928       ToInsert = R.Macro->getName().str();
929       break;
930     case CodeCompletionResult::RK_Pattern:
931       ToInsert = R.Pattern->getTypedText();
932       break;
933     }
934     // We also filter some internal lldb identifiers here. The user
935     // shouldn't see these.
936     if (llvm::StringRef(ToInsert).startswith("$__lldb_"))
937       return llvm::None;
938     if (ToInsert.empty())
939       return llvm::None;
940     // Merge the suggested Token into the existing command line to comply
941     // with the kind of result the lldb API expects.
942     std::string CompletionSuggestion =
943         mergeCompletion(m_expr, m_position, ToInsert);
944 
945     CompletionResult::Completion completion(CompletionSuggestion, Description,
946                                             CompletionMode::Normal);
947     return {{completion, R.Priority}};
948   }
949 
950 public:
951   /// Adds the completions to the given CompletionRequest.
952   void GetCompletions(CompletionRequest &request) {
953     // Bring m_completions into a deterministic order and pass it on to the
954     // CompletionRequest.
955     llvm::sort(m_completions);
956 
957     for (const CompletionWithPriority &C : m_completions)
958       request.AddCompletion(C.completion.GetCompletion(),
959                             C.completion.GetDescription(),
960                             C.completion.GetMode());
961   }
962 
963   /// \name Code-completion callbacks
964   /// Process the finalized code-completion results.
965   void ProcessCodeCompleteResults(Sema &SemaRef, CodeCompletionContext Context,
966                                   CodeCompletionResult *Results,
967                                   unsigned NumResults) override {
968 
969     // The Sema put the incomplete token we try to complete in here during
970     // lexing, so we need to retrieve it here to know what we are completing.
971     StringRef Filter = SemaRef.getPreprocessor().getCodeCompletionFilter();
972 
973     // Iterate over all the results. Filter out results we don't want and
974     // process the rest.
975     for (unsigned I = 0; I != NumResults; ++I) {
976       // Filter the results with the information from the Sema.
977       if (!Filter.empty() && isResultFilteredOut(Filter, Results[I]))
978         continue;
979 
980       CodeCompletionResult &R = Results[I];
981       llvm::Optional<CompletionWithPriority> CompletionAndPriority =
982           getCompletionForResult(R);
983       if (!CompletionAndPriority)
984         continue;
985       m_completions.push_back(*CompletionAndPriority);
986     }
987   }
988 
989   /// \param S the semantic-analyzer object for which code-completion is being
990   /// done.
991   ///
992   /// \param CurrentArg the index of the current argument.
993   ///
994   /// \param Candidates an array of overload candidates.
995   ///
996   /// \param NumCandidates the number of overload candidates
997   void ProcessOverloadCandidates(Sema &S, unsigned CurrentArg,
998                                  OverloadCandidate *Candidates,
999                                  unsigned NumCandidates,
1000                                  SourceLocation OpenParLoc,
1001                                  bool Braced) override {
1002     // At the moment we don't filter out any overloaded candidates.
1003   }
1004 
1005   CodeCompletionAllocator &getAllocator() override {
1006     return m_info.getAllocator();
1007   }
1008 
1009   CodeCompletionTUInfo &getCodeCompletionTUInfo() override { return m_info; }
1010 };
1011 } // namespace
1012 
1013 bool ClangExpressionParser::Complete(CompletionRequest &request, unsigned line,
1014                                      unsigned pos, unsigned typed_pos) {
1015   DiagnosticManager mgr;
1016   // We need the raw user expression here because that's what the CodeComplete
1017   // class uses to provide completion suggestions.
1018   // However, the `Text` method only gives us the transformed expression here.
1019   // To actually get the raw user input here, we have to cast our expression to
1020   // the LLVMUserExpression which exposes the right API. This should never fail
1021   // as we always have a ClangUserExpression whenever we call this.
1022   ClangUserExpression *llvm_expr = cast<ClangUserExpression>(&m_expr);
1023   CodeComplete CC(m_compiler->getLangOpts(), llvm_expr->GetUserText(),
1024                   typed_pos);
1025   // We don't need a code generator for parsing.
1026   m_code_generator.reset();
1027   // Start parsing the expression with our custom code completion consumer.
1028   ParseInternal(mgr, &CC, line, pos);
1029   CC.GetCompletions(request);
1030   return true;
1031 }
1032 
1033 unsigned ClangExpressionParser::Parse(DiagnosticManager &diagnostic_manager) {
1034   return ParseInternal(diagnostic_manager);
1035 }
1036 
1037 unsigned
1038 ClangExpressionParser::ParseInternal(DiagnosticManager &diagnostic_manager,
1039                                      CodeCompleteConsumer *completion_consumer,
1040                                      unsigned completion_line,
1041                                      unsigned completion_column) {
1042   ClangDiagnosticManagerAdapter *adapter =
1043       static_cast<ClangDiagnosticManagerAdapter *>(
1044           m_compiler->getDiagnostics().getClient());
1045 
1046   adapter->ResetManager(&diagnostic_manager);
1047 
1048   const char *expr_text = m_expr.Text();
1049 
1050   clang::SourceManager &source_mgr = m_compiler->getSourceManager();
1051   bool created_main_file = false;
1052 
1053   // Clang wants to do completion on a real file known by Clang's file manager,
1054   // so we have to create one to make this work.
1055   // TODO: We probably could also simulate to Clang's file manager that there
1056   // is a real file that contains our code.
1057   bool should_create_file = completion_consumer != nullptr;
1058 
1059   // We also want a real file on disk if we generate full debug info.
1060   should_create_file |= m_compiler->getCodeGenOpts().getDebugInfo() ==
1061                         codegenoptions::FullDebugInfo;
1062 
1063   if (should_create_file) {
1064     int temp_fd = -1;
1065     llvm::SmallString<128> result_path;
1066     if (FileSpec tmpdir_file_spec = HostInfo::GetProcessTempDir()) {
1067       tmpdir_file_spec.AppendPathComponent("lldb-%%%%%%.expr");
1068       std::string temp_source_path = tmpdir_file_spec.GetPath();
1069       llvm::sys::fs::createUniqueFile(temp_source_path, temp_fd, result_path);
1070     } else {
1071       llvm::sys::fs::createTemporaryFile("lldb", "expr", temp_fd, result_path);
1072     }
1073 
1074     if (temp_fd != -1) {
1075       lldb_private::NativeFile file(temp_fd, File::eOpenOptionWriteOnly, true);
1076       const size_t expr_text_len = strlen(expr_text);
1077       size_t bytes_written = expr_text_len;
1078       if (file.Write(expr_text, bytes_written).Success()) {
1079         if (bytes_written == expr_text_len) {
1080           file.Close();
1081           if (auto fileEntry = m_compiler->getFileManager().getOptionalFileRef(
1082                   result_path)) {
1083             source_mgr.setMainFileID(source_mgr.createFileID(
1084                 *fileEntry,
1085                 SourceLocation(), SrcMgr::C_User));
1086             created_main_file = true;
1087           }
1088         }
1089       }
1090     }
1091   }
1092 
1093   if (!created_main_file) {
1094     std::unique_ptr<MemoryBuffer> memory_buffer =
1095         MemoryBuffer::getMemBufferCopy(expr_text, m_filename);
1096     source_mgr.setMainFileID(source_mgr.createFileID(std::move(memory_buffer)));
1097   }
1098 
1099   adapter->BeginSourceFile(m_compiler->getLangOpts(),
1100                            &m_compiler->getPreprocessor());
1101 
1102   ClangExpressionHelper *type_system_helper =
1103       dyn_cast<ClangExpressionHelper>(m_expr.GetTypeSystemHelper());
1104 
1105   // If we want to parse for code completion, we need to attach our code
1106   // completion consumer to the Sema and specify a completion position.
1107   // While parsing the Sema will call this consumer with the provided
1108   // completion suggestions.
1109   if (completion_consumer) {
1110     auto main_file = source_mgr.getFileEntryForID(source_mgr.getMainFileID());
1111     auto &PP = m_compiler->getPreprocessor();
1112     // Lines and columns start at 1 in Clang, but code completion positions are
1113     // indexed from 0, so we need to add 1 to the line and column here.
1114     ++completion_line;
1115     ++completion_column;
1116     PP.SetCodeCompletionPoint(main_file, completion_line, completion_column);
1117   }
1118 
1119   ASTConsumer *ast_transformer =
1120       type_system_helper->ASTTransformer(m_code_generator.get());
1121 
1122   std::unique_ptr<clang::ASTConsumer> Consumer;
1123   if (ast_transformer) {
1124     Consumer = std::make_unique<ASTConsumerForwarder>(ast_transformer);
1125   } else if (m_code_generator) {
1126     Consumer = std::make_unique<ASTConsumerForwarder>(m_code_generator.get());
1127   } else {
1128     Consumer = std::make_unique<ASTConsumer>();
1129   }
1130 
1131   clang::ASTContext &ast_context = m_compiler->getASTContext();
1132 
1133   m_compiler->setSema(new Sema(m_compiler->getPreprocessor(), ast_context,
1134                                *Consumer, TU_Complete, completion_consumer));
1135   m_compiler->setASTConsumer(std::move(Consumer));
1136 
1137   if (ast_context.getLangOpts().Modules) {
1138     m_compiler->createASTReader();
1139     m_ast_context->setSema(&m_compiler->getSema());
1140   }
1141 
1142   ClangExpressionDeclMap *decl_map = type_system_helper->DeclMap();
1143   if (decl_map) {
1144     decl_map->InstallCodeGenerator(&m_compiler->getASTConsumer());
1145     decl_map->InstallDiagnosticManager(diagnostic_manager);
1146 
1147     clang::ExternalASTSource *ast_source = decl_map->CreateProxy();
1148 
1149     if (ast_context.getExternalSource()) {
1150       auto module_wrapper =
1151           new ExternalASTSourceWrapper(ast_context.getExternalSource());
1152 
1153       auto ast_source_wrapper = new ExternalASTSourceWrapper(ast_source);
1154 
1155       auto multiplexer =
1156           new SemaSourceWithPriorities(*module_wrapper, *ast_source_wrapper);
1157       IntrusiveRefCntPtr<ExternalASTSource> Source(multiplexer);
1158       ast_context.setExternalSource(Source);
1159     } else {
1160       ast_context.setExternalSource(ast_source);
1161     }
1162     decl_map->InstallASTContext(*m_ast_context);
1163   }
1164 
1165   // Check that the ASTReader is properly attached to ASTContext and Sema.
1166   if (ast_context.getLangOpts().Modules) {
1167     assert(m_compiler->getASTContext().getExternalSource() &&
1168            "ASTContext doesn't know about the ASTReader?");
1169     assert(m_compiler->getSema().getExternalSource() &&
1170            "Sema doesn't know about the ASTReader?");
1171   }
1172 
1173   {
1174     llvm::CrashRecoveryContextCleanupRegistrar<Sema> CleanupSema(
1175         &m_compiler->getSema());
1176     ParseAST(m_compiler->getSema(), false, false);
1177   }
1178 
1179   // Make sure we have no pointer to the Sema we are about to destroy.
1180   if (ast_context.getLangOpts().Modules)
1181     m_ast_context->setSema(nullptr);
1182   // Destroy the Sema. This is necessary because we want to emulate the
1183   // original behavior of ParseAST (which also destroys the Sema after parsing).
1184   m_compiler->setSema(nullptr);
1185 
1186   adapter->EndSourceFile();
1187 
1188   unsigned num_errors = adapter->getNumErrors();
1189 
1190   if (m_pp_callbacks && m_pp_callbacks->hasErrors()) {
1191     num_errors++;
1192     diagnostic_manager.PutString(eDiagnosticSeverityError,
1193                                  "while importing modules:");
1194     diagnostic_manager.AppendMessageToDiagnostic(
1195         m_pp_callbacks->getErrorString());
1196   }
1197 
1198   if (!num_errors) {
1199     type_system_helper->CommitPersistentDecls();
1200   }
1201 
1202   adapter->ResetManager();
1203 
1204   return num_errors;
1205 }
1206 
1207 std::string
1208 ClangExpressionParser::GetClangTargetABI(const ArchSpec &target_arch) {
1209   std::string abi;
1210 
1211   if (target_arch.IsMIPS()) {
1212     switch (target_arch.GetFlags() & ArchSpec::eMIPSABI_mask) {
1213     case ArchSpec::eMIPSABI_N64:
1214       abi = "n64";
1215       break;
1216     case ArchSpec::eMIPSABI_N32:
1217       abi = "n32";
1218       break;
1219     case ArchSpec::eMIPSABI_O32:
1220       abi = "o32";
1221       break;
1222     default:
1223       break;
1224     }
1225   }
1226   return abi;
1227 }
1228 
1229 /// Applies the given Fix-It hint to the given commit.
1230 static void ApplyFixIt(const FixItHint &fixit, clang::edit::Commit &commit) {
1231   // This is cobbed from clang::Rewrite::FixItRewriter.
1232   if (fixit.CodeToInsert.empty()) {
1233     if (fixit.InsertFromRange.isValid()) {
1234       commit.insertFromRange(fixit.RemoveRange.getBegin(),
1235                              fixit.InsertFromRange, /*afterToken=*/false,
1236                              fixit.BeforePreviousInsertions);
1237       return;
1238     }
1239     commit.remove(fixit.RemoveRange);
1240     return;
1241   }
1242   if (fixit.RemoveRange.isTokenRange() ||
1243       fixit.RemoveRange.getBegin() != fixit.RemoveRange.getEnd()) {
1244     commit.replace(fixit.RemoveRange, fixit.CodeToInsert);
1245     return;
1246   }
1247   commit.insert(fixit.RemoveRange.getBegin(), fixit.CodeToInsert,
1248                 /*afterToken=*/false, fixit.BeforePreviousInsertions);
1249 }
1250 
1251 bool ClangExpressionParser::RewriteExpression(
1252     DiagnosticManager &diagnostic_manager) {
1253   clang::SourceManager &source_manager = m_compiler->getSourceManager();
1254   clang::edit::EditedSource editor(source_manager, m_compiler->getLangOpts(),
1255                                    nullptr);
1256   clang::edit::Commit commit(editor);
1257   clang::Rewriter rewriter(source_manager, m_compiler->getLangOpts());
1258 
1259   class RewritesReceiver : public edit::EditsReceiver {
1260     Rewriter &rewrite;
1261 
1262   public:
1263     RewritesReceiver(Rewriter &in_rewrite) : rewrite(in_rewrite) {}
1264 
1265     void insert(SourceLocation loc, StringRef text) override {
1266       rewrite.InsertText(loc, text);
1267     }
1268     void replace(CharSourceRange range, StringRef text) override {
1269       rewrite.ReplaceText(range.getBegin(), rewrite.getRangeSize(range), text);
1270     }
1271   };
1272 
1273   RewritesReceiver rewrites_receiver(rewriter);
1274 
1275   const DiagnosticList &diagnostics = diagnostic_manager.Diagnostics();
1276   size_t num_diags = diagnostics.size();
1277   if (num_diags == 0)
1278     return false;
1279 
1280   for (const auto &diag : diagnostic_manager.Diagnostics()) {
1281     const auto *diagnostic = llvm::dyn_cast<ClangDiagnostic>(diag.get());
1282     if (!diagnostic)
1283       continue;
1284     if (!diagnostic->HasFixIts())
1285       continue;
1286     for (const FixItHint &fixit : diagnostic->FixIts())
1287       ApplyFixIt(fixit, commit);
1288   }
1289 
1290   // FIXME - do we want to try to propagate specific errors here?
1291   if (!commit.isCommitable())
1292     return false;
1293   else if (!editor.commit(commit))
1294     return false;
1295 
1296   // Now play all the edits, and stash the result in the diagnostic manager.
1297   editor.applyRewrites(rewrites_receiver);
1298   RewriteBuffer &main_file_buffer =
1299       rewriter.getEditBuffer(source_manager.getMainFileID());
1300 
1301   std::string fixed_expression;
1302   llvm::raw_string_ostream out_stream(fixed_expression);
1303 
1304   main_file_buffer.write(out_stream);
1305   out_stream.flush();
1306   diagnostic_manager.SetFixedExpression(fixed_expression);
1307 
1308   return true;
1309 }
1310 
1311 static bool FindFunctionInModule(ConstString &mangled_name,
1312                                  llvm::Module *module, const char *orig_name) {
1313   for (const auto &func : module->getFunctionList()) {
1314     const StringRef &name = func.getName();
1315     if (name.contains(orig_name)) {
1316       mangled_name.SetString(name);
1317       return true;
1318     }
1319   }
1320 
1321   return false;
1322 }
1323 
1324 lldb_private::Status ClangExpressionParser::PrepareForExecution(
1325     lldb::addr_t &func_addr, lldb::addr_t &func_end,
1326     lldb::IRExecutionUnitSP &execution_unit_sp, ExecutionContext &exe_ctx,
1327     bool &can_interpret, ExecutionPolicy execution_policy) {
1328   func_addr = LLDB_INVALID_ADDRESS;
1329   func_end = LLDB_INVALID_ADDRESS;
1330   Log *log = GetLog(LLDBLog::Expressions);
1331 
1332   lldb_private::Status err;
1333 
1334   std::unique_ptr<llvm::Module> llvm_module_up(
1335       m_code_generator->ReleaseModule());
1336 
1337   if (!llvm_module_up) {
1338     err.SetErrorToGenericError();
1339     err.SetErrorString("IR doesn't contain a module");
1340     return err;
1341   }
1342 
1343   ConstString function_name;
1344 
1345   if (execution_policy != eExecutionPolicyTopLevel) {
1346     // Find the actual name of the function (it's often mangled somehow)
1347 
1348     if (!FindFunctionInModule(function_name, llvm_module_up.get(),
1349                               m_expr.FunctionName())) {
1350       err.SetErrorToGenericError();
1351       err.SetErrorStringWithFormat("Couldn't find %s() in the module",
1352                                    m_expr.FunctionName());
1353       return err;
1354     } else {
1355       LLDB_LOGF(log, "Found function %s for %s", function_name.AsCString(),
1356                 m_expr.FunctionName());
1357     }
1358   }
1359 
1360   SymbolContext sc;
1361 
1362   if (lldb::StackFrameSP frame_sp = exe_ctx.GetFrameSP()) {
1363     sc = frame_sp->GetSymbolContext(lldb::eSymbolContextEverything);
1364   } else if (lldb::TargetSP target_sp = exe_ctx.GetTargetSP()) {
1365     sc.target_sp = target_sp;
1366   }
1367 
1368   LLVMUserExpression::IRPasses custom_passes;
1369   {
1370     auto lang = m_expr.Language();
1371     LLDB_LOGF(log, "%s - Current expression language is %s\n", __FUNCTION__,
1372               Language::GetNameForLanguageType(lang));
1373     lldb::ProcessSP process_sp = exe_ctx.GetProcessSP();
1374     if (process_sp && lang != lldb::eLanguageTypeUnknown) {
1375       auto runtime = process_sp->GetLanguageRuntime(lang);
1376       if (runtime)
1377         runtime->GetIRPasses(custom_passes);
1378     }
1379   }
1380 
1381   if (custom_passes.EarlyPasses) {
1382     LLDB_LOGF(log,
1383               "%s - Running Early IR Passes from LanguageRuntime on "
1384               "expression module '%s'",
1385               __FUNCTION__, m_expr.FunctionName());
1386 
1387     custom_passes.EarlyPasses->run(*llvm_module_up);
1388   }
1389 
1390   execution_unit_sp = std::make_shared<IRExecutionUnit>(
1391       m_llvm_context, // handed off here
1392       llvm_module_up, // handed off here
1393       function_name, exe_ctx.GetTargetSP(), sc,
1394       m_compiler->getTargetOpts().Features);
1395 
1396   ClangExpressionHelper *type_system_helper =
1397       dyn_cast<ClangExpressionHelper>(m_expr.GetTypeSystemHelper());
1398   ClangExpressionDeclMap *decl_map =
1399       type_system_helper->DeclMap(); // result can be NULL
1400 
1401   if (decl_map) {
1402     StreamString error_stream;
1403     IRForTarget ir_for_target(decl_map, m_expr.NeedsVariableResolution(),
1404                               *execution_unit_sp, error_stream,
1405                               function_name.AsCString());
1406 
1407     if (!ir_for_target.runOnModule(*execution_unit_sp->GetModule())) {
1408       err.SetErrorString(error_stream.GetString());
1409       return err;
1410     }
1411 
1412     Process *process = exe_ctx.GetProcessPtr();
1413 
1414     if (execution_policy != eExecutionPolicyAlways &&
1415         execution_policy != eExecutionPolicyTopLevel) {
1416       lldb_private::Status interpret_error;
1417 
1418       bool interpret_function_calls =
1419           !process ? false : process->CanInterpretFunctionCalls();
1420       can_interpret = IRInterpreter::CanInterpret(
1421           *execution_unit_sp->GetModule(), *execution_unit_sp->GetFunction(),
1422           interpret_error, interpret_function_calls);
1423 
1424       if (!can_interpret && execution_policy == eExecutionPolicyNever) {
1425         err.SetErrorStringWithFormat(
1426             "Can't evaluate the expression without a running target due to: %s",
1427             interpret_error.AsCString());
1428         return err;
1429       }
1430     }
1431 
1432     if (!process && execution_policy == eExecutionPolicyAlways) {
1433       err.SetErrorString("Expression needed to run in the target, but the "
1434                          "target can't be run");
1435       return err;
1436     }
1437 
1438     if (!process && execution_policy == eExecutionPolicyTopLevel) {
1439       err.SetErrorString("Top-level code needs to be inserted into a runnable "
1440                          "target, but the target can't be run");
1441       return err;
1442     }
1443 
1444     if (execution_policy == eExecutionPolicyAlways ||
1445         (execution_policy != eExecutionPolicyTopLevel && !can_interpret)) {
1446       if (m_expr.NeedsValidation() && process) {
1447         if (!process->GetDynamicCheckers()) {
1448           ClangDynamicCheckerFunctions *dynamic_checkers =
1449               new ClangDynamicCheckerFunctions();
1450 
1451           DiagnosticManager install_diagnostics;
1452 
1453           if (!dynamic_checkers->Install(install_diagnostics, exe_ctx)) {
1454             if (install_diagnostics.Diagnostics().size())
1455               err.SetErrorString(install_diagnostics.GetString().c_str());
1456             else
1457               err.SetErrorString("couldn't install checkers, unknown error");
1458 
1459             return err;
1460           }
1461 
1462           process->SetDynamicCheckers(dynamic_checkers);
1463 
1464           LLDB_LOGF(log, "== [ClangExpressionParser::PrepareForExecution] "
1465                          "Finished installing dynamic checkers ==");
1466         }
1467 
1468         if (auto *checker_funcs = llvm::dyn_cast<ClangDynamicCheckerFunctions>(
1469                 process->GetDynamicCheckers())) {
1470           IRDynamicChecks ir_dynamic_checks(*checker_funcs,
1471                                             function_name.AsCString());
1472 
1473           llvm::Module *module = execution_unit_sp->GetModule();
1474           if (!module || !ir_dynamic_checks.runOnModule(*module)) {
1475             err.SetErrorToGenericError();
1476             err.SetErrorString("Couldn't add dynamic checks to the expression");
1477             return err;
1478           }
1479 
1480           if (custom_passes.LatePasses) {
1481             LLDB_LOGF(log,
1482                       "%s - Running Late IR Passes from LanguageRuntime on "
1483                       "expression module '%s'",
1484                       __FUNCTION__, m_expr.FunctionName());
1485 
1486             custom_passes.LatePasses->run(*module);
1487           }
1488         }
1489       }
1490     }
1491 
1492     if (execution_policy == eExecutionPolicyAlways ||
1493         execution_policy == eExecutionPolicyTopLevel || !can_interpret) {
1494       execution_unit_sp->GetRunnableInfo(err, func_addr, func_end);
1495     }
1496   } else {
1497     execution_unit_sp->GetRunnableInfo(err, func_addr, func_end);
1498   }
1499 
1500   return err;
1501 }
1502 
1503 lldb_private::Status ClangExpressionParser::RunStaticInitializers(
1504     lldb::IRExecutionUnitSP &execution_unit_sp, ExecutionContext &exe_ctx) {
1505   lldb_private::Status err;
1506 
1507   lldbassert(execution_unit_sp.get());
1508   lldbassert(exe_ctx.HasThreadScope());
1509 
1510   if (!execution_unit_sp.get()) {
1511     err.SetErrorString(
1512         "can't run static initializers for a NULL execution unit");
1513     return err;
1514   }
1515 
1516   if (!exe_ctx.HasThreadScope()) {
1517     err.SetErrorString("can't run static initializers without a thread");
1518     return err;
1519   }
1520 
1521   std::vector<lldb::addr_t> static_initializers;
1522 
1523   execution_unit_sp->GetStaticInitializers(static_initializers);
1524 
1525   for (lldb::addr_t static_initializer : static_initializers) {
1526     EvaluateExpressionOptions options;
1527 
1528     lldb::ThreadPlanSP call_static_initializer(new ThreadPlanCallFunction(
1529         exe_ctx.GetThreadRef(), Address(static_initializer), CompilerType(),
1530         llvm::ArrayRef<lldb::addr_t>(), options));
1531 
1532     DiagnosticManager execution_errors;
1533     lldb::ExpressionResults results =
1534         exe_ctx.GetThreadRef().GetProcess()->RunThreadPlan(
1535             exe_ctx, call_static_initializer, options, execution_errors);
1536 
1537     if (results != lldb::eExpressionCompleted) {
1538       err.SetErrorStringWithFormat("couldn't run static initializer: %s",
1539                                    execution_errors.GetString().c_str());
1540       return err;
1541     }
1542   }
1543 
1544   return err;
1545 }
1546