=====================================\r
-The PDB TPI Stream\r
+The PDB TPI and IPI Streams\r
=====================================\r
+\r
+.. contents::\r
+ :local:\r
+\r
+.. _tpi_intro:\r
+\r
+Introduction\r
+============\r
+\r
+The PDB TPI Stream (Index 2) and IPI Stream (Index 3) contain information about\r
+all types used in the program. It is organized as a :ref:`header <tpi_header>`\r
+followed by a list of :doc:`CodeView Type Records <CodeViewTypes>`. Types are\r
+referenced from various streams and records throughout the PDB by their\r
+:ref:`type index <type_indices>`. In general, the sequence of type records\r
+following the :ref:`header <tpi_header>` forms a topologically sorted DAG\r
+(directed acyclic graph), which means that a type record B can only refer to\r
+the type A if ``A.TypeIndex < B.TypeIndex``. While there are rare cases where\r
+this property will not hold (particularly when dealing with object files\r
+compiled with MASM), an implementation should try very hard to make this\r
+property hold, as it means the entire type graph can be constructed in a single\r
+pass.\r
+\r
+.. important::\r
+ Type records form a topologically sorted DAG (directed acyclic graph).\r
+ \r
+.. _tpi_ipi:\r
+\r
+TPI vs IPI Stream\r
+=================\r
+\r
+Recent versions of the PDB format (aka all versions covered by this document)\r
+have 2 streams with identical layout, henceforth referred to as the TPI stream\r
+and IPI stream. Subsequent contents of this document describing the on-disk\r
+format apply equally whether it is for the TPI Stream or the IPI Stream. The\r
+only difference between the two is in *which* CodeView records are allowed to\r
+appear in each one, summarized by the following table:\r
+\r
++----------------------+---------------------+\r
+| TPI Stream | IPI Stream |\r
++======================+=====================+\r
+| LF_POINTER | LF_FUNC_ID |\r
++----------------------+---------------------+\r
+| LF_MODIFIER | LF_MFUNC_ID |\r
++----------------------+---------------------+\r
+| LF_PROCEDURE | LF_BUILDINFO |\r
++----------------------+---------------------+\r
+| LF_MFUNCTION | LF_SUBSTR_LIST |\r
++----------------------+---------------------+\r
+| LF_LABEL | LF_STRING_ID |\r
++----------------------+---------------------+\r
+| LF_ARGLIST | LF_UDT_SRC_LINE |\r
++----------------------+---------------------+\r
+| LF_FIELDLIST | LF_UDT_MOD_SRC_LINE |\r
++----------------------+---------------------+\r
+| LF_ARRAY | |\r
++----------------------+---------------------+\r
+| LF_CLASS | |\r
++----------------------+---------------------+\r
+| LF_STRUCTURE | |\r
++----------------------+---------------------+\r
+| LF_INTERFACE | |\r
++----------------------+---------------------+\r
+| LF_UNION | |\r
++----------------------+---------------------+\r
+| LF_ENUM | |\r
++----------------------+---------------------+\r
+| LF_TYPESERVER2 | |\r
++----------------------+---------------------+\r
+| LF_VFTABLE | |\r
++----------------------+---------------------+\r
+| LF_VTSHAPE | |\r
++----------------------+---------------------+\r
+| LF_BITFIELD | |\r
++----------------------+---------------------+\r
+| LF_METHODLIST | |\r
++----------------------+---------------------+\r
+| LF_PRECOMP | |\r
++----------------------+---------------------+\r
+| LF_ENDPRECOMP | |\r
++----------------------+---------------------+\r
+\r
+The usage of these records is described in more detail in\r
+:doc:`CodeView Type Records <CodeViewTypes>`.\r
+\r
+.. _type_indices:\r
+\r
+Type Indices\r
+============\r
+\r
+A type index is a 32-bit integer that uniquely identifies a type inside of an\r
+object file's ``.debug$T`` section or a PDB file's TPI or IPI stream. The\r
+value of the type index for the first type record from the TPI stream is given\r
+by the ``TypeIndexBegin`` member of the :ref:`TPI Stream Header <tpi_header>`\r
+although in practice this value is always equal to 0x1000 (4096).\r
+\r
+Any type index with a high bit set is considered to come from the IPI stream,\r
+although this appears to be more of a hack, and LLVM does not generate type\r
+indices of this nature. They can, however, be observed in Microsoft PDBs\r
+occasionally, so one should be prepared to handle them. Note that having the\r
+high bit set is not a necessary condition to determine whether a type index\r
+comes from the IPI stream, it is only sufficient.\r
+\r
+Once the high bit is cleared, any type index >= ``TypeIndexBegin`` is presumed\r
+to come from the appropriate stream, and any type index less than this is a\r
+bitmask which can be decomposed as follows:\r
+\r
+.. code-block:: none\r
+\r
+ .---------------------------.------.----------.\r
+ | Unused | Mode | Kind |\r
+ '---------------------------'------'----------'\r
+ |+32 |+12 |+8 |+0\r
+\r
+\r
+- **Kind** - A value from the following enum:\r
+\r
+.. code-block:: c++\r
+\r
+ enum class SimpleTypeKind : uint32_t {\r
+ None = 0x0000, // uncharacterized type (no type)\r
+ Void = 0x0003, // void\r
+ NotTranslated = 0x0007, // type not translated by cvpack\r
+ HResult = 0x0008, // OLE/COM HRESULT\r
+\r
+ SignedCharacter = 0x0010, // 8 bit signed\r
+ UnsignedCharacter = 0x0020, // 8 bit unsigned\r
+ NarrowCharacter = 0x0070, // really a char\r
+ WideCharacter = 0x0071, // wide char\r
+ Character16 = 0x007a, // char16_t\r
+ Character32 = 0x007b, // char32_t\r
+\r
+ SByte = 0x0068, // 8 bit signed int\r
+ Byte = 0x0069, // 8 bit unsigned int\r
+ Int16Short = 0x0011, // 16 bit signed\r
+ UInt16Short = 0x0021, // 16 bit unsigned\r
+ Int16 = 0x0072, // 16 bit signed int\r
+ UInt16 = 0x0073, // 16 bit unsigned int\r
+ Int32Long = 0x0012, // 32 bit signed\r
+ UInt32Long = 0x0022, // 32 bit unsigned\r
+ Int32 = 0x0074, // 32 bit signed int\r
+ UInt32 = 0x0075, // 32 bit unsigned int\r
+ Int64Quad = 0x0013, // 64 bit signed\r
+ UInt64Quad = 0x0023, // 64 bit unsigned\r
+ Int64 = 0x0076, // 64 bit signed int\r
+ UInt64 = 0x0077, // 64 bit unsigned int\r
+ Int128Oct = 0x0014, // 128 bit signed int\r
+ UInt128Oct = 0x0024, // 128 bit unsigned int\r
+ Int128 = 0x0078, // 128 bit signed int\r
+ UInt128 = 0x0079, // 128 bit unsigned int\r
+\r
+ Float16 = 0x0046, // 16 bit real\r
+ Float32 = 0x0040, // 32 bit real\r
+ Float32PartialPrecision = 0x0045, // 32 bit PP real\r
+ Float48 = 0x0044, // 48 bit real\r
+ Float64 = 0x0041, // 64 bit real\r
+ Float80 = 0x0042, // 80 bit real\r
+ Float128 = 0x0043, // 128 bit real\r
+\r
+ Complex16 = 0x0056, // 16 bit complex\r
+ Complex32 = 0x0050, // 32 bit complex\r
+ Complex32PartialPrecision = 0x0055, // 32 bit PP complex\r
+ Complex48 = 0x0054, // 48 bit complex\r
+ Complex64 = 0x0051, // 64 bit complex\r
+ Complex80 = 0x0052, // 80 bit complex\r
+ Complex128 = 0x0053, // 128 bit complex\r
+\r
+ Boolean8 = 0x0030, // 8 bit boolean\r
+ Boolean16 = 0x0031, // 16 bit boolean\r
+ Boolean32 = 0x0032, // 32 bit boolean\r
+ Boolean64 = 0x0033, // 64 bit boolean\r
+ Boolean128 = 0x0034, // 128 bit boolean\r
+ };\r
+\r
+- **Mode** - A value from the following enum:\r
+\r
+.. code-block:: c++\r
+\r
+ enum class SimpleTypeMode : uint32_t {\r
+ Direct = 0, // Not a pointer\r
+ NearPointer = 1, // Near pointer\r
+ FarPointer = 2, // Far pointer\r
+ HugePointer = 3, // Huge pointer\r
+ NearPointer32 = 4, // 32 bit near pointer\r
+ FarPointer32 = 5, // 32 bit far pointer\r
+ NearPointer64 = 6, // 64 bit near pointer\r
+ NearPointer128 = 7 // 128 bit near pointer\r
+ };\r
+ \r
+Note that for pointers, the bitness is represented in the mode. So a ``void*``\r
+would have a type index with ``Mode=NearPointer32, Kind=Void`` if built for 32-bits\r
+but a type index with ``Mode=NearPointer64, Kind=Void`` if built for 64-bits.\r
+\r
+By convention, the type index for ``std::nullptr_t`` is constructed the same way\r
+as the type index for ``void*``, but using the bitless enumeration value\r
+``NearPointer``.\r
+\r
+\r
+\r
+.. _tpi_header:\r
+\r
+Stream Header\r
+=============\r
+At offset 0 of the TPI Stream is a header with the following layout:\r
+\r
+\r
+.. code-block:: c++\r
+\r
+ struct TpiStreamHeader {\r
+ uint32_t Version;\r
+ uint32_t HeaderSize;\r
+ uint32_t TypeIndexBegin;\r
+ uint32_t TypeIndexEnd;\r
+ uint32_t TypeRecordBytes;\r
+\r
+ uint16_t HashStreamIndex;\r
+ uint16_t HashAuxStreamIndex;\r
+ uint32_t HashKeySize;\r
+ uint32_t NumHashBuckets;\r
+\r
+ int32_t HashValueBufferOffset;\r
+ uint32_t HashValueBufferLength;\r
+ \r
+ int32_t IndexOffsetBufferOffset;\r
+ uint32_t IndexOffsetBufferLength;\r
+\r
+ int32_t HashAdjBufferOffset;\r
+ uint32_t HashAdjBufferLength;\r
+ };\r
+ \r
+- **Version** - A value from the following enum.\r
+\r
+.. code-block:: c++\r
+\r
+ enum class TpiStreamVersion : uint32_t {\r
+ V40 = 19950410,\r
+ V41 = 19951122,\r
+ V50 = 19961031,\r
+ V70 = 19990903,\r
+ V80 = 20040203,\r
+ };\r
+\r
+Similar to the :doc:`PDB Stream <PdbStream>`, this value always appears to be\r
+``V80``, and no other values have been observed. It is assumed that should\r
+another value be observed, the layout described by this document may not be\r
+accurate.\r
+\r
+- **HeaderSize** - ``sizeof(TpiStreamHeader)``\r
+ \r
+- **TypeIndexBegin** - The numeric value of the type index representing the\r
+ first type record in the TPI stream. This is usually the value 0x1000 as type\r
+ indices lower than this are reserved (see :ref:`Type Indices <type_indices>` for\r
+ a discussion of reserved type indices).\r
+ \r
+- **TypeIndexEnd** - One greater than the numeric value of the type index\r
+ representing the last type record in the TPI stream. The total number of type\r
+ records in the TPI stream can be computed as ``TypeIndexEnd - TypeIndexBegin``.\r
+ \r
+- **TypeRecordBytes** - The number of bytes of type record data following the header.\r
+ \r
+- **HashStreamIndex** - The index of a stream which contains a list of hashes for\r
+ every type record. This value may be -1, indicating that hash information is not\r
+ present. In practice a valid stream index is always observed, so any producer\r
+ implementation should be prepared to emit this stream to ensure compatibility with\r
+ tools which may expect it to be present.\r
+ \r
+- **HashAuxStreamIndex** - Presumably the index of a stream which contains a separate\r
+ hash table, although this has not been observed in practice and it's unclear what it\r
+ might be used for.\r
+ \r
+- **HashKeySize** - The size of a hash value (usually 4 bytes).\r
+\r
+- **NumHashBuckets** - The number of buckets used to generate the hash values in the\r
+ aforementioned hash streams.\r
+\r
+- **HashValueBufferOffset / HashValueBufferLength** - The offset and size within\r
+ the TPI Hash Stream of the list of hash values. It should be assumed that there\r
+ are either 0 hash values, or a number equal to the number of type records in the\r
+ TPI stream (``TypeIndexEnd - TypeEndBegin``). Thus, if ``HashBufferLength`` is\r
+ not equal to ``(TypeIndexEnd - TypeEndBegin) * HashKeySize`` we can consider the\r
+ PDB malformed.\r
+\r
+- **IndexOffsetBufferOffset / IndexOffsetBufferLength** - The offset and size\r
+ within the TPI Hash Stream of the Type Index Offsets Buffer. This is a list of\r
+ pairs of uint32_t's where the first value is a :ref:`Type Index <type_indices>`\r
+ and the second value is the offset in the type record data of the type with this\r
+ index. This can be used to do a binary search followed bin a linear search to\r
+ get amortized O(log n) lookup by type index.\r
+\r
+- **HashAdjBufferOffset / HashAdjBufferLength** - \r
+\r
+.. _tpi_records:\r
+\r
+CodeView Type Record List\r
+=========================\r
+Following the header, there are ``TypeRecordBytes`` bytes of data that represent a\r
+variable length array of :doc:`CodeView type records <CodeViewTypes>`. The number\r
+of such records (e.g. the length of the array) can be determined by computing the\r
+value ``Header.TypeIndexEnd - Header.TypeIndexBegin``.\r
+\r
+log(n) random access is provided by way of the Type Index Offsets array (if present)\r
+described previously.
\ No newline at end of file
projects / llvm / commitdiff