llvm-project/clang/lib/Frontend/LayoutOverrideSource.cpp

209 lines
6.2 KiB
C++
Raw Normal View History

Extend the ExternalASTSource interface to allow the AST source to provide the layout of records, rather than letting Clang compute the layout itself. LLDB provides the motivation for this feature: because various layout-altering attributes (packed, aligned, etc.) don't get reliably get placed into DWARF, the record layouts computed by LLDB from the reconstructed records differ from the actual layouts, and badness occurs. This interface lets the DWARF data drive layout, so we don't need the attributes preserved to get the answer write. The testing methodology for this change is fun. I've introduced a variant of -fdump-record-layouts called -fdump-record-layouts-simple that always has the simple C format and provides size/alignment/field offsets. There is also a -cc1 option -foverride-record-layout=<file> to take the output of -fdump-record-layouts-simple and parse it to produce a set of overridden layouts, which is introduced into the AST via a testing-only ExternalASTSource (called LayoutOverrideSource). Each test contains a number of records to lay out, which use various layout-changing attributes, and then dumps the layouts. We then run the test again, using the preprocessor to eliminate the layout-changing attributes entirely (which would give us different layouts for the records), but supplying the previously-computed record layouts. Finally, we diff the layouts produced from the two runs to be sure that they are identical. Note that this code makes the assumption that we don't *have* to provide the offsets of bases or virtual bases to get the layout right, because the alignment attributes don't affect it. I believe this assumption holds, but if it does not, we can extend LayoutOverrideSource to also provide base offset information. Fixes the Clang side of <rdar://problem/10169539>. llvm-svn: 149055
2012-01-26 15:55:45 +08:00
//===--- LayoutOverrideSource.cpp --Override Record Layouts ---------------===//
//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
#include "clang/Frontend/LayoutOverrideSource.h"
#include "clang/AST/Decl.h"
#include "clang/Basic/CharInfo.h"
Extend the ExternalASTSource interface to allow the AST source to provide the layout of records, rather than letting Clang compute the layout itself. LLDB provides the motivation for this feature: because various layout-altering attributes (packed, aligned, etc.) don't get reliably get placed into DWARF, the record layouts computed by LLDB from the reconstructed records differ from the actual layouts, and badness occurs. This interface lets the DWARF data drive layout, so we don't need the attributes preserved to get the answer write. The testing methodology for this change is fun. I've introduced a variant of -fdump-record-layouts called -fdump-record-layouts-simple that always has the simple C format and provides size/alignment/field offsets. There is also a -cc1 option -foverride-record-layout=<file> to take the output of -fdump-record-layouts-simple and parse it to produce a set of overridden layouts, which is introduced into the AST via a testing-only ExternalASTSource (called LayoutOverrideSource). Each test contains a number of records to lay out, which use various layout-changing attributes, and then dumps the layouts. We then run the test again, using the preprocessor to eliminate the layout-changing attributes entirely (which would give us different layouts for the records), but supplying the previously-computed record layouts. Finally, we diff the layouts produced from the two runs to be sure that they are identical. Note that this code makes the assumption that we don't *have* to provide the offsets of bases or virtual bases to get the layout right, because the alignment attributes don't affect it. I believe this assumption holds, but if it does not, we can extend LayoutOverrideSource to also provide base offset information. Fixes the Clang side of <rdar://problem/10169539>. llvm-svn: 149055
2012-01-26 15:55:45 +08:00
#include "llvm/Support/raw_ostream.h"
#include <fstream>
#include <string>
using namespace clang;
/// \brief Parse a simple identifier.
static std::string parseName(StringRef S) {
if (S.empty() || !isIdentifierHead(S[0]))
return "";
unsigned Offset = 1;
while (Offset < S.size() && isIdentifierBody(S[Offset]))
Extend the ExternalASTSource interface to allow the AST source to provide the layout of records, rather than letting Clang compute the layout itself. LLDB provides the motivation for this feature: because various layout-altering attributes (packed, aligned, etc.) don't get reliably get placed into DWARF, the record layouts computed by LLDB from the reconstructed records differ from the actual layouts, and badness occurs. This interface lets the DWARF data drive layout, so we don't need the attributes preserved to get the answer write. The testing methodology for this change is fun. I've introduced a variant of -fdump-record-layouts called -fdump-record-layouts-simple that always has the simple C format and provides size/alignment/field offsets. There is also a -cc1 option -foverride-record-layout=<file> to take the output of -fdump-record-layouts-simple and parse it to produce a set of overridden layouts, which is introduced into the AST via a testing-only ExternalASTSource (called LayoutOverrideSource). Each test contains a number of records to lay out, which use various layout-changing attributes, and then dumps the layouts. We then run the test again, using the preprocessor to eliminate the layout-changing attributes entirely (which would give us different layouts for the records), but supplying the previously-computed record layouts. Finally, we diff the layouts produced from the two runs to be sure that they are identical. Note that this code makes the assumption that we don't *have* to provide the offsets of bases or virtual bases to get the layout right, because the alignment attributes don't affect it. I believe this assumption holds, but if it does not, we can extend LayoutOverrideSource to also provide base offset information. Fixes the Clang side of <rdar://problem/10169539>. llvm-svn: 149055
2012-01-26 15:55:45 +08:00
++Offset;
return S.substr(0, Offset).str();
}
LayoutOverrideSource::LayoutOverrideSource(StringRef Filename) {
Extend the ExternalASTSource interface to allow the AST source to provide the layout of records, rather than letting Clang compute the layout itself. LLDB provides the motivation for this feature: because various layout-altering attributes (packed, aligned, etc.) don't get reliably get placed into DWARF, the record layouts computed by LLDB from the reconstructed records differ from the actual layouts, and badness occurs. This interface lets the DWARF data drive layout, so we don't need the attributes preserved to get the answer write. The testing methodology for this change is fun. I've introduced a variant of -fdump-record-layouts called -fdump-record-layouts-simple that always has the simple C format and provides size/alignment/field offsets. There is also a -cc1 option -foverride-record-layout=<file> to take the output of -fdump-record-layouts-simple and parse it to produce a set of overridden layouts, which is introduced into the AST via a testing-only ExternalASTSource (called LayoutOverrideSource). Each test contains a number of records to lay out, which use various layout-changing attributes, and then dumps the layouts. We then run the test again, using the preprocessor to eliminate the layout-changing attributes entirely (which would give us different layouts for the records), but supplying the previously-computed record layouts. Finally, we diff the layouts produced from the two runs to be sure that they are identical. Note that this code makes the assumption that we don't *have* to provide the offsets of bases or virtual bases to get the layout right, because the alignment attributes don't affect it. I believe this assumption holds, but if it does not, we can extend LayoutOverrideSource to also provide base offset information. Fixes the Clang side of <rdar://problem/10169539>. llvm-svn: 149055
2012-01-26 15:55:45 +08:00
std::ifstream Input(Filename.str().c_str());
if (!Input.is_open())
return;
// Parse the output of -fdump-record-layouts.
std::string CurrentType;
Layout CurrentLayout;
bool ExpectingType = false;
while (Input.good()) {
std::string Line;
getline(Input, Line);
StringRef LineStr(Line);
// Determine whether the following line will start a
if (LineStr.find("*** Dumping AST Record Layout") != StringRef::npos) {
// Flush the last type/layout, if there is one.
if (!CurrentType.empty())
Layouts[CurrentType] = CurrentLayout;
CurrentLayout = Layout();
ExpectingType = true;
continue;
}
// If we're expecting a type, grab it.
if (ExpectingType) {
ExpectingType = false;
StringRef::size_type Pos;
if ((Pos = LineStr.find("struct ")) != StringRef::npos)
LineStr = LineStr.substr(Pos + strlen("struct "));
else if ((Pos = LineStr.find("class ")) != StringRef::npos)
LineStr = LineStr.substr(Pos + strlen("class "));
else if ((Pos = LineStr.find("union ")) != StringRef::npos)
LineStr = LineStr.substr(Pos + strlen("union "));
else
continue;
// Find the name of the type.
CurrentType = parseName(LineStr);
CurrentLayout = Layout();
continue;
}
// Check for the size of the type.
StringRef::size_type Pos = LineStr.find(" Size:");
Extend the ExternalASTSource interface to allow the AST source to provide the layout of records, rather than letting Clang compute the layout itself. LLDB provides the motivation for this feature: because various layout-altering attributes (packed, aligned, etc.) don't get reliably get placed into DWARF, the record layouts computed by LLDB from the reconstructed records differ from the actual layouts, and badness occurs. This interface lets the DWARF data drive layout, so we don't need the attributes preserved to get the answer write. The testing methodology for this change is fun. I've introduced a variant of -fdump-record-layouts called -fdump-record-layouts-simple that always has the simple C format and provides size/alignment/field offsets. There is also a -cc1 option -foverride-record-layout=<file> to take the output of -fdump-record-layouts-simple and parse it to produce a set of overridden layouts, which is introduced into the AST via a testing-only ExternalASTSource (called LayoutOverrideSource). Each test contains a number of records to lay out, which use various layout-changing attributes, and then dumps the layouts. We then run the test again, using the preprocessor to eliminate the layout-changing attributes entirely (which would give us different layouts for the records), but supplying the previously-computed record layouts. Finally, we diff the layouts produced from the two runs to be sure that they are identical. Note that this code makes the assumption that we don't *have* to provide the offsets of bases or virtual bases to get the layout right, because the alignment attributes don't affect it. I believe this assumption holds, but if it does not, we can extend LayoutOverrideSource to also provide base offset information. Fixes the Clang side of <rdar://problem/10169539>. llvm-svn: 149055
2012-01-26 15:55:45 +08:00
if (Pos != StringRef::npos) {
// Skip past the " Size:" prefix.
LineStr = LineStr.substr(Pos + strlen(" Size:"));
Extend the ExternalASTSource interface to allow the AST source to provide the layout of records, rather than letting Clang compute the layout itself. LLDB provides the motivation for this feature: because various layout-altering attributes (packed, aligned, etc.) don't get reliably get placed into DWARF, the record layouts computed by LLDB from the reconstructed records differ from the actual layouts, and badness occurs. This interface lets the DWARF data drive layout, so we don't need the attributes preserved to get the answer write. The testing methodology for this change is fun. I've introduced a variant of -fdump-record-layouts called -fdump-record-layouts-simple that always has the simple C format and provides size/alignment/field offsets. There is also a -cc1 option -foverride-record-layout=<file> to take the output of -fdump-record-layouts-simple and parse it to produce a set of overridden layouts, which is introduced into the AST via a testing-only ExternalASTSource (called LayoutOverrideSource). Each test contains a number of records to lay out, which use various layout-changing attributes, and then dumps the layouts. We then run the test again, using the preprocessor to eliminate the layout-changing attributes entirely (which would give us different layouts for the records), but supplying the previously-computed record layouts. Finally, we diff the layouts produced from the two runs to be sure that they are identical. Note that this code makes the assumption that we don't *have* to provide the offsets of bases or virtual bases to get the layout right, because the alignment attributes don't affect it. I believe this assumption holds, but if it does not, we can extend LayoutOverrideSource to also provide base offset information. Fixes the Clang side of <rdar://problem/10169539>. llvm-svn: 149055
2012-01-26 15:55:45 +08:00
unsigned long long Size = 0;
(void)LineStr.getAsInteger(10, Size);
CurrentLayout.Size = Size;
continue;
}
// Check for the alignment of the type.
Pos = LineStr.find("Alignment:");
if (Pos != StringRef::npos) {
// Skip past the "Alignment:" prefix.
LineStr = LineStr.substr(Pos + strlen("Alignment:"));
unsigned long long Alignment = 0;
(void)LineStr.getAsInteger(10, Alignment);
CurrentLayout.Align = Alignment;
continue;
}
// Check for the size/alignment of the type.
Pos = LineStr.find("sizeof=");
if (Pos != StringRef::npos) {
/* Skip past the sizeof= prefix. */
LineStr = LineStr.substr(Pos + strlen("sizeof="));
// Parse size.
unsigned long long Size = 0;
(void)LineStr.getAsInteger(10, Size);
CurrentLayout.Size = Size;
Pos = LineStr.find("align=");
if (Pos != StringRef::npos) {
/* Skip past the align= prefix. */
LineStr = LineStr.substr(Pos + strlen("align="));
// Parse alignment.
unsigned long long Alignment = 0;
(void)LineStr.getAsInteger(10, Alignment);
CurrentLayout.Align = Alignment;
}
continue;
}
// Check for the field offsets of the type.
Pos = LineStr.find("FieldOffsets: [");
if (Pos == StringRef::npos)
continue;
LineStr = LineStr.substr(Pos + strlen("FieldOffsets: ["));
while (!LineStr.empty() && isDigit(LineStr[0])) {
Extend the ExternalASTSource interface to allow the AST source to provide the layout of records, rather than letting Clang compute the layout itself. LLDB provides the motivation for this feature: because various layout-altering attributes (packed, aligned, etc.) don't get reliably get placed into DWARF, the record layouts computed by LLDB from the reconstructed records differ from the actual layouts, and badness occurs. This interface lets the DWARF data drive layout, so we don't need the attributes preserved to get the answer write. The testing methodology for this change is fun. I've introduced a variant of -fdump-record-layouts called -fdump-record-layouts-simple that always has the simple C format and provides size/alignment/field offsets. There is also a -cc1 option -foverride-record-layout=<file> to take the output of -fdump-record-layouts-simple and parse it to produce a set of overridden layouts, which is introduced into the AST via a testing-only ExternalASTSource (called LayoutOverrideSource). Each test contains a number of records to lay out, which use various layout-changing attributes, and then dumps the layouts. We then run the test again, using the preprocessor to eliminate the layout-changing attributes entirely (which would give us different layouts for the records), but supplying the previously-computed record layouts. Finally, we diff the layouts produced from the two runs to be sure that they are identical. Note that this code makes the assumption that we don't *have* to provide the offsets of bases or virtual bases to get the layout right, because the alignment attributes don't affect it. I believe this assumption holds, but if it does not, we can extend LayoutOverrideSource to also provide base offset information. Fixes the Clang side of <rdar://problem/10169539>. llvm-svn: 149055
2012-01-26 15:55:45 +08:00
// Parse this offset.
unsigned Idx = 1;
while (Idx < LineStr.size() && isDigit(LineStr[Idx]))
Extend the ExternalASTSource interface to allow the AST source to provide the layout of records, rather than letting Clang compute the layout itself. LLDB provides the motivation for this feature: because various layout-altering attributes (packed, aligned, etc.) don't get reliably get placed into DWARF, the record layouts computed by LLDB from the reconstructed records differ from the actual layouts, and badness occurs. This interface lets the DWARF data drive layout, so we don't need the attributes preserved to get the answer write. The testing methodology for this change is fun. I've introduced a variant of -fdump-record-layouts called -fdump-record-layouts-simple that always has the simple C format and provides size/alignment/field offsets. There is also a -cc1 option -foverride-record-layout=<file> to take the output of -fdump-record-layouts-simple and parse it to produce a set of overridden layouts, which is introduced into the AST via a testing-only ExternalASTSource (called LayoutOverrideSource). Each test contains a number of records to lay out, which use various layout-changing attributes, and then dumps the layouts. We then run the test again, using the preprocessor to eliminate the layout-changing attributes entirely (which would give us different layouts for the records), but supplying the previously-computed record layouts. Finally, we diff the layouts produced from the two runs to be sure that they are identical. Note that this code makes the assumption that we don't *have* to provide the offsets of bases or virtual bases to get the layout right, because the alignment attributes don't affect it. I believe this assumption holds, but if it does not, we can extend LayoutOverrideSource to also provide base offset information. Fixes the Clang side of <rdar://problem/10169539>. llvm-svn: 149055
2012-01-26 15:55:45 +08:00
++Idx;
unsigned long long Offset = 0;
(void)LineStr.substr(0, Idx).getAsInteger(10, Offset);
CurrentLayout.FieldOffsets.push_back(Offset);
// Skip over this offset, the following comma, and any spaces.
LineStr = LineStr.substr(Idx + 1);
while (!LineStr.empty() && isWhitespace(LineStr[0]))
Extend the ExternalASTSource interface to allow the AST source to provide the layout of records, rather than letting Clang compute the layout itself. LLDB provides the motivation for this feature: because various layout-altering attributes (packed, aligned, etc.) don't get reliably get placed into DWARF, the record layouts computed by LLDB from the reconstructed records differ from the actual layouts, and badness occurs. This interface lets the DWARF data drive layout, so we don't need the attributes preserved to get the answer write. The testing methodology for this change is fun. I've introduced a variant of -fdump-record-layouts called -fdump-record-layouts-simple that always has the simple C format and provides size/alignment/field offsets. There is also a -cc1 option -foverride-record-layout=<file> to take the output of -fdump-record-layouts-simple and parse it to produce a set of overridden layouts, which is introduced into the AST via a testing-only ExternalASTSource (called LayoutOverrideSource). Each test contains a number of records to lay out, which use various layout-changing attributes, and then dumps the layouts. We then run the test again, using the preprocessor to eliminate the layout-changing attributes entirely (which would give us different layouts for the records), but supplying the previously-computed record layouts. Finally, we diff the layouts produced from the two runs to be sure that they are identical. Note that this code makes the assumption that we don't *have* to provide the offsets of bases or virtual bases to get the layout right, because the alignment attributes don't affect it. I believe this assumption holds, but if it does not, we can extend LayoutOverrideSource to also provide base offset information. Fixes the Clang side of <rdar://problem/10169539>. llvm-svn: 149055
2012-01-26 15:55:45 +08:00
LineStr = LineStr.substr(1);
}
}
// Flush the last type/layout, if there is one.
if (!CurrentType.empty())
Layouts[CurrentType] = CurrentLayout;
}
bool
LayoutOverrideSource::layoutRecordType(const RecordDecl *Record,
uint64_t &Size, uint64_t &Alignment,
llvm::DenseMap<const FieldDecl *, uint64_t> &FieldOffsets,
llvm::DenseMap<const CXXRecordDecl *, CharUnits> &BaseOffsets,
llvm::DenseMap<const CXXRecordDecl *, CharUnits> &VirtualBaseOffsets)
{
// We can't override unnamed declarations.
if (!Record->getIdentifier())
return false;
// Check whether we have a layout for this record.
llvm::StringMap<Layout>::iterator Known = Layouts.find(Record->getName());
if (Known == Layouts.end())
return false;
// Provide field layouts.
unsigned NumFields = 0;
for (RecordDecl::field_iterator F = Record->field_begin(),
FEnd = Record->field_end();
F != FEnd; ++F, ++NumFields) {
if (NumFields >= Known->second.FieldOffsets.size())
continue;
FieldOffsets[*F] = Known->second.FieldOffsets[NumFields];
Extend the ExternalASTSource interface to allow the AST source to provide the layout of records, rather than letting Clang compute the layout itself. LLDB provides the motivation for this feature: because various layout-altering attributes (packed, aligned, etc.) don't get reliably get placed into DWARF, the record layouts computed by LLDB from the reconstructed records differ from the actual layouts, and badness occurs. This interface lets the DWARF data drive layout, so we don't need the attributes preserved to get the answer write. The testing methodology for this change is fun. I've introduced a variant of -fdump-record-layouts called -fdump-record-layouts-simple that always has the simple C format and provides size/alignment/field offsets. There is also a -cc1 option -foverride-record-layout=<file> to take the output of -fdump-record-layouts-simple and parse it to produce a set of overridden layouts, which is introduced into the AST via a testing-only ExternalASTSource (called LayoutOverrideSource). Each test contains a number of records to lay out, which use various layout-changing attributes, and then dumps the layouts. We then run the test again, using the preprocessor to eliminate the layout-changing attributes entirely (which would give us different layouts for the records), but supplying the previously-computed record layouts. Finally, we diff the layouts produced from the two runs to be sure that they are identical. Note that this code makes the assumption that we don't *have* to provide the offsets of bases or virtual bases to get the layout right, because the alignment attributes don't affect it. I believe this assumption holds, but if it does not, we can extend LayoutOverrideSource to also provide base offset information. Fixes the Clang side of <rdar://problem/10169539>. llvm-svn: 149055
2012-01-26 15:55:45 +08:00
}
// Wrong number of fields.
if (NumFields != Known->second.FieldOffsets.size())
return false;
Size = Known->second.Size;
Alignment = Known->second.Align;
return true;
}
LLVM_DUMP_METHOD void LayoutOverrideSource::dump() {
raw_ostream &OS = llvm::errs();
Extend the ExternalASTSource interface to allow the AST source to provide the layout of records, rather than letting Clang compute the layout itself. LLDB provides the motivation for this feature: because various layout-altering attributes (packed, aligned, etc.) don't get reliably get placed into DWARF, the record layouts computed by LLDB from the reconstructed records differ from the actual layouts, and badness occurs. This interface lets the DWARF data drive layout, so we don't need the attributes preserved to get the answer write. The testing methodology for this change is fun. I've introduced a variant of -fdump-record-layouts called -fdump-record-layouts-simple that always has the simple C format and provides size/alignment/field offsets. There is also a -cc1 option -foverride-record-layout=<file> to take the output of -fdump-record-layouts-simple and parse it to produce a set of overridden layouts, which is introduced into the AST via a testing-only ExternalASTSource (called LayoutOverrideSource). Each test contains a number of records to lay out, which use various layout-changing attributes, and then dumps the layouts. We then run the test again, using the preprocessor to eliminate the layout-changing attributes entirely (which would give us different layouts for the records), but supplying the previously-computed record layouts. Finally, we diff the layouts produced from the two runs to be sure that they are identical. Note that this code makes the assumption that we don't *have* to provide the offsets of bases or virtual bases to get the layout right, because the alignment attributes don't affect it. I believe this assumption holds, but if it does not, we can extend LayoutOverrideSource to also provide base offset information. Fixes the Clang side of <rdar://problem/10169539>. llvm-svn: 149055
2012-01-26 15:55:45 +08:00
for (llvm::StringMap<Layout>::iterator L = Layouts.begin(),
LEnd = Layouts.end();
L != LEnd; ++L) {
OS << "Type: blah " << L->first() << '\n';
OS << " Size:" << L->second.Size << '\n';
OS << " Alignment:" << L->second.Align << '\n';
OS << " FieldOffsets: [";
for (unsigned I = 0, N = L->second.FieldOffsets.size(); I != N; ++I) {
if (I)
OS << ", ";
OS << L->second.FieldOffsets[I];
}
OS << "]\n";
}
}