第三部分 - 一个只保存在内存中、只能新增记录的单表数据库

为了使我们的数据库更加简单，在一开始的时候我们先加一些限制条件，比如：

只支持两个操作：插入一行和打印所有行
只存在内存中（不持久化到磁盘）
只支持一张表，并且这张表是硬编码（俗称“写死”）的

我们这张硬编码的表用来存储用户数据，它的结构是这样的：

字段名	类型
id	integer
username	varchar(32)
email	varchar(255)

表结构很简单，但它需要我们实现对多种数据类型和多种大小的文本数据类型的支持。

插入语句暂定这样：

insert 1 cstack foo@bar.com

这也就意味着我们需要更新我们的 prepare_statement 函数来解析多个参数

   if (strncmp(input_buffer->buffer, "insert", 6) == 0) {
     statement->type = STATEMENT_INSERT;
+    int args_assigned = sscanf(
+        input_buffer->buffer, "insert %d %s %s", &(statement->row_to_insert.id),
+        statement->row_to_insert.username, statement->row_to_insert.email);
+    if (args_assigned < 3) {
+      return PREPARE_SYNTAX_ERROR;
+    }
     return PREPARE_SUCCESS;
   }
   if (strcmp(input_buffer->buffer, "select") == 0) {

我们需要定义一个新的数据结构 Row 并将解析好的参数保存其中，同时在 Statement 中定义一个 Row 类型的 row_to_insert 字段。

+#define COLUMN_USERNAME_SIZE 32
+#define COLUMN_EMAIL_SIZE 255
+typedef struct {
+  uint32_t id;
+  char username[COLUMN_USERNAME_SIZE];
+  char email[COLUMN_EMAIL_SIZE];
+} Row;
+
 typedef struct {
   StatementType type;
+  Row row_to_insert;  // 只在插入语句使用
 } Statement;

现在我们需要将数据复制到一个可以呈现表的数据结构中。为了高效查找、插入和删除，SQLite 使用了 B-tree。但我们一开始会先用一些更简单的数据结构来代替。类似 B-tree，它会将数据行按页分组，但它不会将页编排成树，而是编排成数组。

下面是我的方案：

将行保存到被称之为“页”的内存块中
每一页保存尽可能多的行
行保存到页中时会被序列化成一个紧凑形式
页只在需要时才被创建
保存固定大小的页指针数组

我们首先定义一行数据的紧凑形式：

+#define size_of_attribute(Struct, Attribute) sizeof(((Struct*)0)->Attribute)
+
+const uint32_t ID_SIZE = size_of_attribute(Row, id);
+const uint32_t USERNAME_SIZE = size_of_attribute(Row, username);
+const uint32_t EMAIL_SIZE = size_of_attribute(Row, email);
+const uint32_t ID_OFFSET = 0;
+const uint32_t USERNAME_OFFSET = ID_OFFSET + ID_SIZE;
+const uint32_t EMAIL_OFFSET = USERNAME_OFFSET + USERNAME_SIZE;
+const uint32_t ROW_SIZE = ID_SIZE + USERNAME_SIZE + EMAIL_SIZE;

一个序列化后的行的结构是这样的：

列	大小（字节数）	偏移量
id	4	0
username	32	4
email	255	36
总计	291

我们还需要写一些代码来序列化和反序列化行。

+void serialize_row(Row* source, void* destination) {
+  memcpy(destination + ID_OFFSET, &(source->id), ID_SIZE);
+  memcpy(destination + USERNAME_OFFSET, &(source->username), USERNAME_SIZE);
+  memcpy(destination + EMAIL_OFFSET, &(source->email), EMAIL_SIZE);
+}
+
+void deserialize_row(void* source, Row* destination) {
+  memcpy(&(destination->id), source + ID_OFFSET, ID_SIZE);
+  memcpy(&(destination->username), source + USERNAME_OFFSET, USERNAME_SIZE);
+  memcpy(&(destination->email), source + EMAIL_OFFSET, EMAIL_SIZE);
+}

接下来，定义一个 表 结构体，里面包含了页指针数组和一个用于记录当前行数的变量：

+const uint32_t PAGE_SIZE = 4096;
+#define TABLE_MAX_PAGES 100
+const uint32_t ROWS_PER_PAGE = PAGE_SIZE / ROW_SIZE;
+const uint32_t TABLE_MAX_ROWS = ROWS_PER_PAGE * TABLE_MAX_PAGES;
+
+typedef struct {
+  uint32_t num_rows;
+  void* pages[TABLE_MAX_PAGES];
+} Table;

之所以将页大小定义为 4KB，是因为在大多数计算机架构中虚拟内存页大小是 4KB。这意味着我们数据库中的一页对应于操作系统的一页。操作系统在换页时可以将我们的页整体换入或换出，而不是将其分开。

最大页数 100 是随意设置的，当我们后面改用树结构时，数据库的大小将只由文件大小决定。（当然我们会限制一次最多有多少页保存在内存中）

行不应该超出页的边界。由于页与页在内存中大概率不是紧挨在一起的，所以这个设定会使得行的读/写变得更加简单。

说到这里，下面是我们该如何确定要读/写的指定行在内存中的位置的方法：

+void* row_slot(Table* table, uint32_t row_num) {
+  uint32_t page_num = row_num / ROWS_PER_PAGE;
+  void* page = table->pages[page_num];
+  if (page == NULL) {
+    // Allocate memory only when we try to access page
+    page = table->pages[page_num] = malloc(PAGE_SIZE);
+  }
+  uint32_t row_offset = row_num % ROWS_PER_PAGE;
+  uint32_t byte_offset = row_offset * ROW_SIZE;
+  return page + byte_offset;
+}

现在我们可以通过 execute_statement 来从我们的表结构体中读/写数据了：

-void execute_statement(Statement* statement) {
+ExecuteResult execute_insert(Statement* statement, Table* table) {
+  if (table->num_rows >= TABLE_MAX_ROWS) {
+    return EXECUTE_TABLE_FULL;
+  }
+
+  Row* row_to_insert = &(statement->row_to_insert);
+
+  serialize_row(row_to_insert, row_slot(table, table->num_rows));
+  table->num_rows += 1;
+
+  return EXECUTE_SUCCESS;
+}
+
+ExecuteResult execute_select(Statement* statement, Table* table) {
+  Row row;
+  for (uint32_t i = 0; i < table->num_rows; i++) {
+    deserialize_row(row_slot(table, i), &row);
+    print_row(&row);
+  }
+  return EXECUTE_SUCCESS;
+}
+
+ExecuteResult execute_statement(Statement* statement, Table* table) {
   switch (statement->type) {
     case (STATEMENT_INSERT):
-      printf("This is where we would do an insert.\n");
-      break;
+      return execute_insert(statement, table);
     case (STATEMENT_SELECT):
-      printf("This is where we would do a select.\n");
-      break;
+      return execute_select(statement, table);
   }
 }

最后，我们需要初始化数据表，创建相应的内存释放函数，以及添加更多的错误处理逻辑：

+ Table* new_table() {
+  Table* table = malloc(sizeof(Table));
+  table->num_rows = 0;
+  for (uint32_t i = 0; i < TABLE_MAX_PAGES; i++) {
+     table->pages[i] = NULL;
+  }
+  return table;
+}
+
+void free_table(Table* table) {
+    for (int i = 0; table->pages[i]; i++) {
+        free(table->pages[i]);
+    }
+    free(table);
+}

 int main(int argc, char* argv[]) {
+  Table* table = new_table();
   InputBuffer* input_buffer = new_input_buffer();
   while (true) {
     print_prompt();
@@ -105,13 +203,22 @@ int main(int argc, char* argv[]) {
     switch (prepare_statement(input_buffer, &statement)) {
       case (PREPARE_SUCCESS):
         break;
+      case (PREPARE_SYNTAX_ERROR):
+        printf("Syntax error. Could not parse statement.\n");
+        continue;
       case (PREPARE_UNRECOGNIZED_STATEMENT):
         printf("Unrecognized keyword at start of '%s'.\n",
                input_buffer->buffer);
         continue;
     }

-    execute_statement(&statement);
-    printf("Executed.\n");
+    switch (execute_statement(&statement, table)) {
+      case (EXECUTE_SUCCESS):
+        printf("Executed.\n");
+        break;
+      case (EXECUTE_TABLE_FULL):
+        printf("Error: Table full.\n");
+        break;
+    }
   }
 }

有了上面这些改造，我们可以保存数据到我们的数据库中了！

~ ./db
db > insert 1 cstack foo@bar.com
Executed.
db > insert 2 bob bob@example.com
Executed.
db > select
(1, cstack, foo@bar.com)
(2, bob, bob@example.com)
Executed.
db > insert foo bar 1
Syntax error. Could not parse statement.
db > .exit
~

一鼓作气，我们写一些测试用例：

我们后面会大改存储数据的数据结构，有了测试用例可以方便回归测试
有一系列边界条件我们没有手动测试过（例如写满整张表）

这些问题我们将在下一章解决。下面是本章的完整代码变更：

@@ -2,6 +2,7 @@
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
+#include <stdint.h>

 typedef struct {
   char* buffer;
@@ -10,6 +11,105 @@ typedef struct {
 } InputBuffer;

+typedef enum { EXECUTE_SUCCESS, EXECUTE_TABLE_FULL } ExecuteResult;
+
+typedef enum {
+  META_COMMAND_SUCCESS,
+  META_COMMAND_UNRECOGNIZED_COMMAND
+} MetaCommandResult;
+
+typedef enum {
+  PREPARE_SUCCESS,
+  PREPARE_SYNTAX_ERROR,
+  PREPARE_UNRECOGNIZED_STATEMENT
+ } PrepareResult;
+
+typedef enum { STATEMENT_INSERT, STATEMENT_SELECT } StatementType;
+
+#define COLUMN_USERNAME_SIZE 32
+#define COLUMN_EMAIL_SIZE 255
+typedef struct {
+  uint32_t id;
+  char username[COLUMN_USERNAME_SIZE];
+  char email[COLUMN_EMAIL_SIZE];
+} Row;
+
+typedef struct {
+  StatementType type;
+  Row row_to_insert; // 只在插入语句使用
+} Statement;
+
+#define size_of_attribute(Struct, Attribute) sizeof(((Struct*)0)->Attribute)
+
+const uint32_t ID_SIZE = size_of_attribute(Row, id);
+const uint32_t USERNAME_SIZE = size_of_attribute(Row, username);
+const uint32_t EMAIL_SIZE = size_of_attribute(Row, email);
+const uint32_t ID_OFFSET = 0;
+const uint32_t USERNAME_OFFSET = ID_OFFSET + ID_SIZE;
+const uint32_t EMAIL_OFFSET = USERNAME_OFFSET + USERNAME_SIZE;
+const uint32_t ROW_SIZE = ID_SIZE + USERNAME_SIZE + EMAIL_SIZE;
+
+const uint32_t PAGE_SIZE = 4096;
+#define TABLE_MAX_PAGES 100
+const uint32_t ROWS_PER_PAGE = PAGE_SIZE / ROW_SIZE;
+const uint32_t TABLE_MAX_ROWS = ROWS_PER_PAGE * TABLE_MAX_PAGES;
+
+typedef struct {
+  uint32_t num_rows;
+  void* pages[TABLE_MAX_PAGES];
+} Table;
+
+void print_row(Row* row) {
+  printf("(%d, %s, %s)\n", row->id, row->username, row->email);
+}
+
+void serialize_row(Row* source, void* destination) {
+  memcpy(destination + ID_OFFSET, &(source->id), ID_SIZE);
+  memcpy(destination + USERNAME_OFFSET, &(source->username), USERNAME_SIZE);
+  memcpy(destination + EMAIL_OFFSET, &(source->email), EMAIL_SIZE);
+}
+
+void deserialize_row(void *source, Row* destination) {
+  memcpy(&(destination->id), source + ID_OFFSET, ID_SIZE);
+  memcpy(&(destination->username), source + USERNAME_OFFSET, USERNAME_SIZE);
+  memcpy(&(destination->email), source + EMAIL_OFFSET, EMAIL_SIZE);
+}
+
+void* row_slot(Table* table, uint32_t row_num) {
+  uint32_t page_num = row_num / ROWS_PER_PAGE;
+  void *page = table->pages[page_num];
+  if (page == NULL) {
+     // Allocate memory only when we try to access page
+     page = table->pages[page_num] = malloc(PAGE_SIZE);
+  }
+  uint32_t row_offset = row_num % ROWS_PER_PAGE;
+  uint32_t byte_offset = row_offset * ROW_SIZE;
+  return page + byte_offset;
+}
+
+Table* new_table() {
+  Table* table = malloc(sizeof(Table));
+  table->num_rows = 0;
+  for (uint32_t i = 0; i < TABLE_MAX_PAGES; i++) {
+     table->pages[i] = NULL;
+  }
+  return table;
+}
+
+void free_table(Table* table) {
+  for (int i = 0; table->pages[i]; i++) {
+     free(table->pages[i]);
+  }
+  free(table);
+}
+
 InputBuffer* new_input_buffer() {
   InputBuffer* input_buffer = malloc(sizeof(InputBuffer));
   input_buffer->buffer = NULL;
@@ -40,17 +140,105 @@ void close_input_buffer(InputBuffer* input_buffer) {
     free(input_buffer);
 }

+MetaCommandResult do_meta_command(InputBuffer* input_buffer, Table *table) {
+  if (strcmp(input_buffer->buffer, ".exit") == 0) {
+    close_input_buffer(input_buffer);
+    free_table(table);
+    exit(EXIT_SUCCESS);
+  } else {
+    return META_COMMAND_UNRECOGNIZED_COMMAND;
+  }
+}
+
+PrepareResult prepare_statement(InputBuffer* input_buffer,
+                                Statement* statement) {
+  if (strncmp(input_buffer->buffer, "insert", 6) == 0) {
+    statement->type = STATEMENT_INSERT;
+    int args_assigned = sscanf(
+	input_buffer->buffer, "insert %d %s %s", &(statement->row_to_insert.id),
+	statement->row_to_insert.username, statement->row_to_insert.email
+	);
+    if (args_assigned < 3) {
+	return PREPARE_SYNTAX_ERROR;
+    }
+    return PREPARE_SUCCESS;
+  }
+  if (strcmp(input_buffer->buffer, "select") == 0) {
+    statement->type = STATEMENT_SELECT;
+    return PREPARE_SUCCESS;
+  }
+
+  return PREPARE_UNRECOGNIZED_STATEMENT;
+}
+
+ExecuteResult execute_insert(Statement* statement, Table* table) {
+  if (table->num_rows >= TABLE_MAX_ROWS) {
+     return EXECUTE_TABLE_FULL;
+  }
+
+  Row* row_to_insert = &(statement->row_to_insert);
+
+  serialize_row(row_to_insert, row_slot(table, table->num_rows));
+  table->num_rows += 1;
+
+  return EXECUTE_SUCCESS;
+}
+
+ExecuteResult execute_select(Statement* statement, Table* table) {
+  Row row;
+  for (uint32_t i = 0; i < table->num_rows; i++) {
+     deserialize_row(row_slot(table, i), &row);
+     print_row(&row);
+  }
+  return EXECUTE_SUCCESS;
+}
+
+ExecuteResult execute_statement(Statement* statement, Table *table) {
+  switch (statement->type) {
+    case (STATEMENT_INSERT):
+       	return execute_insert(statement, table);
+    case (STATEMENT_SELECT):
+	return execute_select(statement, table);
+  }
+}
+
 int main(int argc, char* argv[]) {
+  Table* table = new_table();
   InputBuffer* input_buffer = new_input_buffer();
   while (true) {
     print_prompt();
     read_input(input_buffer);

-    if (strcmp(input_buffer->buffer, ".exit") == 0) {
-      close_input_buffer(input_buffer);
-      exit(EXIT_SUCCESS);
-    } else {
-      printf("Unrecognized command '%s'.\n", input_buffer->buffer);
+    if (input_buffer->buffer[0] == '.') {
+      switch (do_meta_command(input_buffer, table)) {
+        case (META_COMMAND_SUCCESS):
+          continue;
+        case (META_COMMAND_UNRECOGNIZED_COMMAND):
+          printf("Unrecognized command '%s'\n", input_buffer->buffer);
+          continue;
+      }
+    }
+
+    Statement statement;
+    switch (prepare_statement(input_buffer, &statement)) {
+      case (PREPARE_SUCCESS):
+        break;
+      case (PREPARE_SYNTAX_ERROR):
+	printf("Syntax error. Could not parse statement.\n");
+	continue;
+      case (PREPARE_UNRECOGNIZED_STATEMENT):
+        printf("Unrecognized keyword at start of '%s'.\n",
+               input_buffer->buffer);
+        continue;
+    }
+
+    switch (execute_statement(&statement, table)) {
+	case (EXECUTE_SUCCESS):
+	    printf("Executed.\n");
+	    break;
+	case (EXECUTE_TABLE_FULL):
+	    printf("Error: Table full.\n");
+	    break;
     }
   }
 }

第二部分 - 世界上最简单的 SQL 编译器和虚拟机

第四部分 - 第一组测试（...以及 bug 们）