Provide single-threaded (no-locking) implementation of IMetadataEmit …

…family of APIs. (#52)

.gitignore

@@ -1,4 +1,5 @@
 artifacts/
 .vs/
 .idea/
-*~
+*~
+TestResults/

configure.cmake

@@ -55,4 +55,10 @@ endif()
 #
 # Agnostic compiler/platform settings
 #
-add_compile_definitions(__STDC_WANT_LIB_EXT1__=1) # https://en.cppreference.com/w/c/error#Bounds_checking
+add_compile_definitions(__STDC_WANT_LIB_EXT1__=1) # https://en.cppreference.com/w/c/error#Bounds_checking
+
+option(DNMD_ENABLE_PROFILING OFF)
+
+if (DNMD_ENABLE_PROFILING AND MSVC)
+    add_link_options(/PROFILE)
+endif()

src/dnmd/CMakeLists.txt

@@ -37,10 +37,12 @@ target_include_directories(dnmd PUBLIC $<INSTALL_INTERFACE:include>)
 target_include_directories(dnmd_pdb PUBLIC $<INSTALL_INTERFACE:include>)
 
 set_target_properties(dnmd PROPERTIES
-  PUBLIC_HEADER "../inc/dnmd.h;../inc/dnmd.hpp")
+  PUBLIC_HEADER "../inc/dnmd.h;../inc/dnmd.hpp"
+  POSITION_INDEPENDENT_CODE ON)
 
 set_target_properties(dnmd_pdb PROPERTIES
-  PUBLIC_HEADER "../inc/dnmd.h;../inc/dnmd.hpp;../inc/dnmd_pdb.h")
+  PUBLIC_HEADER "../inc/dnmd.h;../inc/dnmd.hpp;../inc/dnmd_pdb.h"
+  POSITION_INDEPENDENT_CODE ON)
 
 install(TARGETS dnmd dnmd_pdb EXPORT dnmd
   PUBLIC_HEADER DESTINATION include

src/dnmd/editor.c

@@ -568,7 +568,7 @@ static bool reserve_heap_space(mdeditor_t* editor, uint32_t space_size, mdtcol_t
     {
         // Set the default heap size based on likely reasonable sizes for the heaps.
         // In most images, there won't be more than three guids, so we can start with a small heap in that case.
-        const size_t initial_heap_size = heap_id == mdtc_hguid ? sizeof(mdguid_t) * 3 : 0x100;
+        size_t const initial_heap_size = heap_id == mdtc_hguid ? sizeof(mdguid_t) * 3 : 0x100;
         void* mem = alloc_mdmem(editor->cxt, initial_heap_size);
         if (mem == NULL)
             return false;
@@ -628,6 +628,14 @@ static bool reserve_heap_space(mdeditor_t* editor, uint32_t space_size, mdtcol_t
 
 uint32_t add_to_string_heap(mdcxt_t* cxt, char const* str)
 {
+    // II.24.2.3 - When the #String heap is present, the first entry is always the empty string (i.e., \0). 
+    // II.24.2.2 -  Streams need not be there if they are empty.
+    // We can avoid allocating the heap if the only entry is the empty string.
+    // Columns that point to the string heap can be 0 if there is no #String heap.
+    // In that case, they represent the empty string.
+    if (str[0] == '\0')
+        return 0;
+
     mdeditor_t* editor = get_editor(cxt);
     if (editor == NULL)
         return 0;
@@ -646,12 +654,21 @@ uint32_t add_to_string_heap(mdcxt_t* cxt, char const* str)
 
 uint32_t add_to_blob_heap(mdcxt_t* cxt, uint8_t const* data, uint32_t length)
 {
+    // II.24.2.4 - When the #Blob heap is present, the first entry is always the empty blob. 
+    // II.24.2.2 -  Streams need not be there if they are empty.
+    // We can avoid allocating the heap if the only entry is the empty blob.
+    // Columns that point to the blob heap can be 0 if there is no #Blob heap.
+    // In that case, they represent the empty blob.
+    if (length == 0)
+        return 0;
+
     mdeditor_t* editor = get_editor(cxt);
     if (editor == NULL)
         return 0;
+
     // TODO: Deduplicate heap
     uint8_t compressed_length[4];
-    size_t compressed_length_size = 0;
+    size_t compressed_length_size = ARRAY_SIZE(compressed_length);
     if (!compress_u32(length, compressed_length, &compressed_length_size))
         return 0;
 
@@ -670,16 +687,12 @@ uint32_t add_to_blob_heap(mdcxt_t* cxt, uint8_t const* data, uint32_t length)
 
 uint32_t add_to_user_string_heap(mdcxt_t* cxt, char16_t const* str)
 {
-    mdeditor_t* editor = get_editor(cxt);
-    if (editor == NULL)
-        return 0;
-    // TODO: Deduplicate heap
     uint32_t str_len;
     uint8_t has_special_char = 0;
     for (str_len = 0; str[str_len] != (char16_t)0; str_len++)
     {
         char16_t c = str[str_len];
-        // II.23.2.4
+        // II.24.2.4
         // There is an additional terminal byte which holds a 1 or 0.
         // The 1 signifies Unicode characters that require handling beyond
         // that normally provided for 8-bit encoding sets.
@@ -713,31 +726,61 @@ uint32_t add_to_user_string_heap(mdcxt_t* cxt, char16_t const* str)
         }
     }
 
-    uint8_t compressed_length[sizeof(str_len)];
-    size_t compressed_length_size = 0;
-    if (!compress_u32(str_len, compressed_length, &compressed_length_size))
+    // II.24.2.4 - When the #US heap is present, the first entry is always the empty blob. 
+    // II.24.2.2 -  Streams need not be there if they are empty.
+    // We can avoid allocating the heap if the only entry is the empty blob.
+    // Indices into the #US heap can be 0 if there is no #US heap.
+    // In that case, they represent the empty userstring blob.
+    if (str_len == 0)
         return 0;
 
-    uint32_t heap_slot_size = str_len + (uint32_t)compressed_length_size + 1;
+    mdeditor_t* editor = get_editor(cxt);
+    if (editor == NULL)
+        return 0;
+
+    // TODO: Deduplicate heap
+
+    // II.24.2.4
+    // Strings in the #US (user string) heap are encoded using 16-bit Unicode encodings.
+    // The count on each string is the number of bytes (not characters) in the string.
+    // Furthermore, there is an additional terminal byte (so all byte counts are odd, not even).
+    size_t us_blob_bytes = str_len * sizeof(char16_t) + 1;
+
+    // The string is too long to represent in the heap.
+    if (us_blob_bytes > INT32_MAX)
+        return 0;
+    uint8_t compressed_length[sizeof(uint32_t)];
+    size_t compressed_length_size = ARRAY_SIZE(compressed_length);
+    if (!compress_u32((uint32_t)us_blob_bytes, compressed_length, &compressed_length_size))
+        return 0;
+
+    uint32_t heap_slot_size = (uint32_t)us_blob_bytes + (uint32_t)compressed_length_size;
     uint32_t heap_offset;
     if (!reserve_heap_space(editor, heap_slot_size, mdtc_hus, false, &heap_offset))
     {
         return 0;
     }
 
+    // Copy the compressed blob length into the heap.
     memcpy(editor->user_string_heap.heap.ptr + heap_offset, compressed_length, compressed_length_size);
-    memcpy(editor->user_string_heap.heap.ptr + heap_offset + compressed_length_size, str, str_len);
+    // Copy the UTF-16-encoded user string into the heap.
+    memcpy(editor->user_string_heap.heap.ptr + heap_offset + compressed_length_size, str, us_blob_bytes - 1);
 
-    editor->user_string_heap.heap.ptr[heap_offset + compressed_length_size + str_len] = has_special_char;
+    // Set the trailing byte.
+    editor->user_string_heap.heap.ptr[heap_offset + compressed_length_size + us_blob_bytes - 1] = has_special_char;
     return heap_offset;
 }
 
+mdguid_t const empty_guid = { 0 };
+
 uint32_t add_to_guid_heap(mdcxt_t* cxt, mdguid_t guid)
 {
     mdeditor_t* editor = get_editor(cxt);
     if (editor == NULL)
         return 0;
     // TODO: Deduplicate heap
+    if (memcmp(&guid, &empty_guid, sizeof(mdguid_t)) == 0)
+        return 0;
 
     uint32_t heap_offset;
     if (!reserve_heap_space(editor, sizeof(mdguid_t), mdtc_hguid, false, &heap_offset))
@@ -746,7 +789,10 @@ uint32_t add_to_guid_heap(mdcxt_t* cxt, mdguid_t guid)
     }
 
     memcpy(editor->guid_heap.heap.ptr + heap_offset, &guid, sizeof(mdguid_t));
-    return heap_offset / sizeof(mdguid_t);
+    // II.22 -  The Guid heap is an array of GUIDs, each 16 bytes wide.  Its 
+    //    first element is numbered 1, its second 2, and so on. 
+    // So, we need to make the offset 1-based and at the scale of the GUID size.
+    return (heap_offset / sizeof(mdguid_t)) + 1;
 }
 
 bool append_heap(mdcxt_t* cxt, mdcxt_t* delta, mdtcol_t heap_id)

src/dnmd/entry.c

@@ -297,7 +297,7 @@ static bool initialize_minimal_table_rows(mdcxt_t* cxt)
     if (1 != md_set_column_value_as_utf8(global_type_cursor, mdtTypeDef_TypeNamespace, 1, &namespace))
         return false;
 
-    mdToken nil_typedef = CreateTokenType(mdtTypeDef);
+    mdToken nil_typedef = CreateTokenType(mdtid_TypeDef);
     if (1 != md_set_column_value_as_token(global_type_cursor, mdtTypeDef_Extends, 1, &nil_typedef))
         return false;
 
@@ -361,26 +361,21 @@ mdhandle_t md_create_new_pdb_handle()
 }
 #endif // DNMD_PORTABLE_PDB
 
-bool md_apply_delta(mdhandle_t handle, void const* data, size_t data_len)
+bool md_apply_delta(mdhandle_t handle, mdhandle_t delta_handle)
 {
     mdcxt_t* base = extract_mdcxt(handle);
     if (base == NULL)
         return false;
 
-    mdhandle_t h;
-    if (!md_create_handle(data, data_len, &h))
+    mdcxt_t* delta = extract_mdcxt(delta_handle);
+    if (delta == NULL)
         return false;
 
-    mdcxt_t* delta = extract_mdcxt(h);
-    assert(delta != NULL);
-
     // Verify the supplied delta is actually a delta file
     bool result = false;
     if (delta->context_flags & mdc_minimal_delta)
         result = merge_in_delta(base, delta);
 
-    // Free all data for the delta
-    md_destroy_handle(h);
     return result;
 }
 

src/dnmd/internal.h

@@ -430,4 +430,6 @@ bool sort_list_by_column(mdcursor_t parent, col_index_t list_col, col_index_t co
 // Add the heap with the specified id from the delta image to the cxt image.
 bool append_heap(mdcxt_t* cxt, mdcxt_t* delta, mdtcol_t heap_id);
 
+extern mdguid_t const empty_guid;
+
 #endif // _SRC_DNMD_INTERNAL_H_

src/dnmd/streams.c

@@ -5,6 +5,17 @@ bool try_get_string(mdcxt_t* cxt, size_t offset, char const** str)
     assert(cxt != NULL && str != NULL);
 
     mdstream_t* h = &cxt->strings_heap;
+
+    // II.24.2.3 - When the #String heap is present, the first entry is always the empty string (i.e., \0). 
+    // II.24.2.2 -  Streams need not be there if they are empty.
+    // If the offset into the heap is 0, we can treat that as a "null" index into the heap and return
+    // the empty string.
+    if (h->size == 0 && offset == 0)
+    {
+        *str = "\0"; // II.24.2.3 'The first character must be the '\0' character.
+        return true;
+    }
+
     if (h->size <= offset)
         return false;
 
@@ -81,6 +92,15 @@ bool try_get_blob(mdcxt_t* cxt, size_t offset, uint8_t const** blob, uint32_t* b
     assert(cxt != NULL && blob != NULL && blob_len != NULL);
 
     mdstream_t* h = &cxt->blob_heap;
+
+    if (h->size == 0 && offset == 0)
+    {
+        // The first element must be the 0 - II.24.2.4
+        *blob = h->ptr;
+        *blob_len = 0;
+        return true;
+    }
+
     if (h->size <= offset)
         return false;
 

src/dnmd/tables.c

@@ -772,11 +772,18 @@ bool initialize_new_table_details(
     // Set the new table's row count temporarily to 1 to ensure that we initialize the table.
     table_row_counts[id] = 1;
 
+    // We'll treat any new table that has keys as sorted.
+    // We only want to do this for tables with keys as tables without keys
+    // never use the is_sorted bit.
+    md_key_info_t const* keys;
+    uint8_t key_count = get_table_keys(id, &keys);
+    bool has_keys = key_count != 0;
+
     if (!initialize_table_details(
         table_row_counts,
         cxt->context_flags,
         id,
-        false,
+        has_keys,
         table))
         return false;