現在MVC架構開發中。m部分是提供底層數據。無論是cs還是bs越來越看重數據對象的業務處理，而不是以前原生的sql得到的數據。
因此，1個通用的關係數據庫字段對應對模型對象的框架就比較重要了。有了他可以節省大量的開發時間。
本篇側重簡要分析django中的orm。
對於orm，既然是通用，那麼就存在5個重要問題。
1：如何多數據庫的支持
2：對象字段類型的提供
3：sql->object如何實現轉換
4：object->sql如何存儲
5：數據庫字段關聯關係在對象中的對應。
11分析如上幾個問題。
1：對於多數據庫支持，這個問題比較好解決，因爲sql通用的操作代碼比較多。差別在細微和一些數據庫特性。數據庫聯接上。基於分層思想。
高層封裝sql通用方法類。特性方法。和數據庫操作方法類。在具體對應的數據庫實現文件中針對性改寫。
根據猜想，分析django.db.backends模塊下文件。得到
backends
….
|----sqlite3
|----__init__.py
|----base.py 基於sqlite3構造繼承數據庫基類。填充sqlte3特性，完善sqlte3的運算操作與遊標
|----client.py 本地執行sqlite3 構造db文件
|----creation.py 提供構造數據庫的model與數據庫字段對應關係，填充創建方法
|----introspection.py 提供自省對應的model與數據庫字段關係，填充自省方法
__init__.py 數據庫基礎，特性，運算操作，自省，本地執行，驗證 等基類封裝
creation.py 構造數據庫基類封裝
signals.py 數據庫構造信號
util.py 遊標，調試遊標（記錄時間）封裝問題2：對象字段類型的提供。
sql語句與iphone的視圖與android的控件一樣。都是1個提供比較同樣方法或者屬性的集合。細微差別在子類實現時候差異化。
對於sql提供的字段，有整形，布爾型，字符，浮點，日期等。。。
這段比較不熟悉，因此分析models的模塊得到如下目錄
models
|----fields
|----__init__.py 對象的基類與對象類
|----fies.py 文件對象
|----proxy.py 代理類，暫時還不知道用途
|----related.py 關聯關係（1對多，多對多，反向，主見，1對1）
|----subclassing.py 暫時不知道
|----sql
|----__init__.py
|----aggregates.py 集合類，用於組合sql字段模版，重點是as_sql
|----complier.py sql語句組合類，最終查詢是通過此類實現
|----constants.py 常量
|----datastructures.py 暫時不清楚
|----expressions.py 評估程序？暫時不清楚
|----query.py 查詢
|----subqueries.py 增刪改查表達式（從Qeury文件中繼承）
|----where.py
__init__.py
aggregates.py 集合類，平均數，統計，最大，最小，stdDev，求和，方差
base.py 模型基類
constants.py
deletion.py 刪除的集合類collecor
expression.py 邏輯運算表達式
loading.py 加載 加載應用程序於模型，註冊模型，等方法的封裝
manager.py 對象管理類。對象管理描述符類
options.py 模型對象支持的選項
query.py 查詢集合類（基類，值類，值列表，時間，空）
query_utils.py 查詢包裝
related.py
signals.py 操作相關的信號定義大致根據如上的代碼詳細查看源代碼會方便理解。
有了sql的連接管理，有了model提供字段，如何管理對象。。
從django的使用上可以看出。
模型.管理.操作()[切片處理]得到QuerySet。其中QuerySet封裝在models.query.py中。
下面就剩下3個問題。
3：sql->object，從QuerySet入手。分析iterator()
def iterator(self):
"""
An iterator over the results from applying this QuerySet to the
database.
"""
...
# Cache db and model outside the loop
db = self.db
model = self.model
compiler = self.query.get_compiler(using=db)
if fill_cache:
klass_info = get_klass_info(model, max_depth=max_depth,
requested=requested, only_load=only_load) 得到類信息
for row in compiler.results_iter(): 從數據庫遍歷字段
if fill_cache:
obj, _ = get_cached_row(row, index_start, db, klass_info,
offset=len(aggregate_select)) 獲取字段
else:
# Omit aggregates in object creation.
row_data = row[index_start:aggregate_start]
if skip:
obj = model_cls(**dict(zip(init_list, row_data)))
else:
obj = model(*row_data) 根據字段生成模型對象
# Store the source database of the object
obj._state.db = db
# This object came from the database; it's not being added.
obj._state.adding = False
if extra_select:
for i, k in enumerate(extra_select):
setattr(obj, k, row[i])
# Add the aggregates to the model
if aggregate_select:
for i, aggregate in enumerate(aggregate_select):
setattr(obj, aggregate, row[i + aggregate_start])
# Add the known related objects to the model, if there are any
if self._known_related_objects:
for field, rel_objs in self._known_related_objects.items():
pk = getattr(obj, field.get_attname())
try:
rel_obj = rel_objs[pk]
except KeyError:
pass # may happen in qs1 | qs2 scenarios
else:
setattr(obj, field.name, rel_obj)
yield objdef get_cached_row(row, index_start, using, klass_info, offset=0):
"""
Helper function that recursively returns an object with the specified
related attributes already populated.
This method may be called recursively to populate deep select_related()
clauses.
Arguments:
* row - the row of data returned by the database cursor
* index_start - the index of the row at which data for this
object is known to start
* offset - the number of additional fields that are known to
exist in row for `klass`. This usually means the number of
annotated results on `klass`.
* using - the database alias on which the query is being executed.
* klass_info - result of the get_klass_info function
"""
if klass_info is None:
return None
klass, field_names, field_count, related_fields, reverse_related_fields, pk_idx = klass_info
fields = row[index_start : index_start + field_count]
# If the pk column is None (or the Oracle equivalent ''), then the related
# object must be non-existent - set the relation to None.
if fields[pk_idx] == None or fields[pk_idx] == '':
obj = None
elif field_names:
obj = klass(**dict(zip(field_names, fields)))
else:
obj = klass(*fields)生成對象
# If an object was retrieved, set the database state.
if obj:
obj._state.db = using
obj._state.adding = False
# Instantiate related fields
index_end = index_start + field_count + offset
# Iterate over each related object, populating any
# select_related() fields
for f, klass_info in related_fields:
# Recursively retrieve the data for the related object
cached_row = get_cached_row(row, index_end, using, klass_info)
# If the recursive descent found an object, populate the
# descriptor caches relevant to the object
if cached_row:
rel_obj, index_end = cached_row
if obj is not None:
# If the base object exists, populate the
# descriptor cache
setattr(obj, f.get_cache_name(), rel_obj)
if f.unique and rel_obj is not None:
# If the field is unique, populate the
# reverse descriptor cache on the related object
setattr(rel_obj, f.related.get_cache_name(), obj)
# Now do the same, but for reverse related objects.
# Only handle the restricted case - i.e., don't do a depth
# descent into reverse relations unless explicitly requested
for f, klass_info in reverse_related_fields:
# Recursively retrieve the data for the related object
cached_row = get_cached_row(row, index_end, using, klass_info)
# If the recursive descent found an object, populate the
# descriptor caches relevant to the object
if cached_row:
rel_obj, index_end = cached_row
if obj is not None:
# If the field is unique, populate the
# reverse descriptor cache
setattr(obj, f.related.get_cache_name(), rel_obj)
if rel_obj is not None:
# If the related object exists, populate
# the descriptor cache.
setattr(rel_obj, f.get_cache_name(), obj)
# Now populate all the non-local field values on the related
# object. If this object has deferred fields, we need to use
# the opts from the original model to get non-local fields
# correctly.
opts = rel_obj._meta
if getattr(rel_obj, '_deferred'):
opts = opts.proxy_for_model._meta
for rel_field, rel_model in opts.get_fields_with_model():
if rel_model is not None:
setattr(rel_obj, rel_field.attname, getattr(obj, rel_field.attname))
# populate the field cache for any related object
# that has already been retrieved
if rel_field.rel:
try:
cached_obj = getattr(obj, rel_field.get_cache_name())
setattr(rel_obj, rel_field.get_cache_name(), cached_obj)
except AttributeError:
# Related object hasn't been cached yet
pass
return obj, index_end以上是簡要分析。
下面還有如何把object->sql
object.save()方法，執行以後會將數據存儲到db。因此分析model基類的save方法。分析到save_base方法
def save_base(self, raw=False, cls=None, origin=None, force_insert=False,
force_update=False, using=None, update_fields=None):
...
manager = cls._base_manager
...
result = manager._insert([self], fields=fields, return_id=update_pk, using=using, raw=raw)這裏可以看出是插入部分manager內部
def _insert(self, objs, fields, **kwargs):
return insert_query(self.model, objs, fields, **kwargs)insert_query中
def insert_query(model, objs, fields, return_id=False, raw=False, using=None):
query = sql.InsertQuery(model)生成插入查詢對象
query.insert_values(fields, objs, raw=raw)插入值
return query.get_compiler(using=using).execute_sql(return_id)獲得生成sql對象，並執行sql語句插入值部分
def insert_values(self, fields, objs, raw=False):
self.fields = fields
# Check that no Promise object reaches the DB. Refs #10498.
for field in fields:
for obj in objs:
value = getattr(obj, field.attname)
if isinstance(value, Promise):
setattr(obj, field.attname, force_text(value))
self.objs = objs
self.raw = raw獲取sql編譯對象上面已經說過。
查看執行語句
def execute_sql(self, return_id=False):
assert not (return_id and len(self.query.objs) != 1)
self.return_id = return_id
cursor = self.connection.cursor()
for sql, params in self.as_sql():
cursor.execute(sql, params)
if not (return_id and cursor):
return
if self.connection.features.can_return_id_from_insert:
return self.connection.ops.fetch_returned_insert_id(cursor)
return self.connection.ops.last_insert_id(cursor,
self.query.model._meta.db_table, self.query.model._meta.pk.column)看到這裏，看as_sql()方法
def as_sql(self):
qn = self.connection.ops.quote_name
opts = self.query.model._meta
result = ['INSERT INTO %s' % qn(opts.db_table)]
has_fields = bool(self.query.fields)
fields = self.query.fields if has_fields else [opts.pk]
result.append('(%s)' % ', '.join([qn(f.column) for f in fields]))
if has_fields:
params = values = [
[
f.get_db_prep_save(getattr(obj, f.attname) if self.query.raw else f.pre_save(obj, True), connection=self.connection)
for f in fields
]
for obj in self.query.objs
]
else:
values = [[self.connection.ops.pk_default_value()] for obj in self.query.objs]
params = [[]]
fields = [None]
can_bulk = (not any(hasattr(field, "get_placeholder") for field in fields) and
not self.return_id and self.connection.features.has_bulk_insert)
if can_bulk:
placeholders = [["%s"] * len(fields)]
else:
placeholders = [
[self.placeholder(field, v) for field, v in zip(fields, val)]
for val in values
]
# Oracle Spatial needs to remove some values due to #10888
params = self.connection.ops.modify_insert_params(placeholders, params)
if self.return_id and self.connection.features.can_return_id_from_insert:
params = params[0]
col = "%s.%s" % (qn(opts.db_table), qn(opts.pk.column))
result.append("VALUES (%s)" % ", ".join(placeholders[0]))
r_fmt, r_params = self.connection.ops.return_insert_id()
# Skip empty r_fmt to allow subclasses to customize behaviour for
# 3rd party backends. Refs #19096.
if r_fmt:
result.append(r_fmt % col)
params += r_params
return [(" ".join(result), tuple(params))]
if can_bulk:
result.append(self.connection.ops.bulk_insert_sql(fields, len(values)))
return [(" ".join(result), tuple([v for val in values for v in val]))]
else:
return [
(" ".join(result + ["VALUES (%s)" % ", ".join(p)]), vals)
for p, vals in zip(placeholders, params)
]因此object->sql部分邏輯是
object->得到表名，字段名，和值
格式化到 insert into 表 (字段) values (值) 並執行
最後1個問題。。有點複雜的說。要深入學習下。