| 1 |
.. title:: clang-tidy - bugprone-unchecked-optional-access |
| 2 |
|
| 3 |
bugprone-unchecked-optional-access |
| 4 |
================================== |
| 5 |
|
| 6 |
*Note*: This check uses a flow-sensitive static analysis to produce its |
| 7 |
results. Therefore, it may be more resource intensive (RAM, CPU) than the |
| 8 |
average clang-tidy check. |
| 9 |
|
| 10 |
This check identifies unsafe accesses to values contained in |
| 11 |
``std::optional<T>``, ``absl::optional<T>``, ``base::Optional<T>``, or |
| 12 |
``folly::Optional<T>`` objects. Below we will refer to all these types |
| 13 |
collectively as ``optional<T>``. |
| 14 |
|
| 15 |
An access to the value of an ``optional<T>`` occurs when one of its ``value``, |
| 16 |
``operator*``, or ``operator->`` member functions is invoked. To align with |
| 17 |
common misconceptions, the check considers these member functions as equivalent, |
| 18 |
even though there are subtle differences related to exceptions versus undefined |
| 19 |
behavior. See *Additional notes*, below, for more information on this topic. |
| 20 |
|
| 21 |
An access to the value of an ``optional<T>`` is considered safe if and only if |
| 22 |
code in the local scope (for example, a function body) ensures that the |
| 23 |
``optional<T>`` has a value in all possible execution paths that can reach the |
| 24 |
access. That should happen either through an explicit check, using the |
| 25 |
``optional<T>::has_value`` member function, or by constructing the |
| 26 |
``optional<T>`` in a way that shows that it unambiguously holds a value (e.g |
| 27 |
using ``std::make_optional`` which always returns a populated |
| 28 |
``std::optional<T>``). |
| 29 |
|
| 30 |
Below we list some examples, starting with unsafe optional access patterns, |
| 31 |
followed by safe access patterns. |
| 32 |
|
| 33 |
Unsafe access patterns |
| 34 |
~~~~~~~~~~~~~~~~~~~~~~ |
| 35 |
|
| 36 |
Access the value without checking if it exists |
| 37 |
---------------------------------------------- |
| 38 |
|
| 39 |
The check flags accesses to the value that are not locally guarded by |
| 40 |
existence check: |
| 41 |
|
| 42 |
.. code-block:: c++ |
| 43 |
|
| 44 |
void f(std::optional<int> opt) { |
| 45 |
use(*opt); // unsafe: it is unclear whether `opt` has a value. |
| 46 |
} |
| 47 |
|
| 48 |
Access the value in the wrong branch |
| 49 |
------------------------------------ |
| 50 |
|
| 51 |
The check is aware of the state of an optional object in different |
| 52 |
branches of the code. For example: |
| 53 |
|
| 54 |
.. code-block:: c++ |
| 55 |
|
| 56 |
void f(std::optional<int> opt) { |
| 57 |
if (opt.has_value()) { |
| 58 |
} else { |
| 59 |
use(opt.value()); // unsafe: it is clear that `opt` does *not* have a value. |
| 60 |
} |
| 61 |
} |
| 62 |
|
| 63 |
Assume a function result to be stable |
| 64 |
------------------------------------- |
| 65 |
|
| 66 |
The check is aware that function results might not be stable. That is, |
| 67 |
consecutive calls to the same function might return different values. |
| 68 |
For example: |
| 69 |
|
| 70 |
.. code-block:: c++ |
| 71 |
|
| 72 |
void f(Foo foo) { |
| 73 |
if (foo.opt().has_value()) { |
| 74 |
use(*foo.opt()); // unsafe: it is unclear whether `foo.opt()` has a value. |
| 75 |
} |
| 76 |
} |
| 77 |
|
| 78 |
Rely on invariants of uncommon APIs |
| 79 |
----------------------------------- |
| 80 |
|
| 81 |
The check is unaware of invariants of uncommon APIs. For example: |
| 82 |
|
| 83 |
.. code-block:: c++ |
| 84 |
|
| 85 |
void f(Foo foo) { |
| 86 |
if (foo.HasProperty("bar")) { |
| 87 |
use(*foo.GetProperty("bar")); // unsafe: it is unclear whether `foo.GetProperty("bar")` has a value. |
| 88 |
} |
| 89 |
} |
| 90 |
|
| 91 |
Check if a value exists, then pass the optional to another function |
| 92 |
------------------------------------------------------------------- |
| 93 |
|
| 94 |
The check relies on local reasoning. The check and value access must |
| 95 |
both happen in the same function. An access is considered unsafe even if |
| 96 |
the caller of the function performing the access ensures that the |
| 97 |
optional has a value. For example: |
| 98 |
|
| 99 |
.. code-block:: c++ |
| 100 |
|
| 101 |
void g(std::optional<int> opt) { |
| 102 |
use(*opt); // unsafe: it is unclear whether `opt` has a value. |
| 103 |
} |
| 104 |
|
| 105 |
void f(std::optional<int> opt) { |
| 106 |
if (opt.has_value()) { |
| 107 |
g(opt); |
| 108 |
} |
| 109 |
} |
| 110 |
|
| 111 |
Safe access patterns |
| 112 |
~~~~~~~~~~~~~~~~~~~~ |
| 113 |
|
| 114 |
Check if a value exists, then access the value |
| 115 |
---------------------------------------------- |
| 116 |
|
| 117 |
The check recognizes all straightforward ways for checking if a value |
| 118 |
exists and accessing the value contained in an optional object. For |
| 119 |
example: |
| 120 |
|
| 121 |
.. code-block:: c++ |
| 122 |
|
| 123 |
void f(std::optional<int> opt) { |
| 124 |
if (opt.has_value()) { |
| 125 |
use(*opt); |
| 126 |
} |
| 127 |
} |
| 128 |
|
| 129 |
|
| 130 |
Check if a value exists, then access the value from a copy |
| 131 |
---------------------------------------------------------- |
| 132 |
|
| 133 |
The criteria that the check uses is semantic, not syntactic. It |
| 134 |
recognizes when a copy of the optional object being accessed is known to |
| 135 |
have a value. For example: |
| 136 |
|
| 137 |
.. code-block:: c++ |
| 138 |
|
| 139 |
void f(std::optional<int> opt1) { |
| 140 |
if (opt1.has_value()) { |
| 141 |
std::optional<int> opt2 = opt1; |
| 142 |
use(*opt2); |
| 143 |
} |
| 144 |
} |
| 145 |
|
| 146 |
|
| 147 |
Ensure that a value exists using common macros |
| 148 |
---------------------------------------------- |
| 149 |
|
| 150 |
The check is aware of common macros like ``CHECK`` and ``DCHECK``. Those can be |
| 151 |
used to ensure that an optional object has a value. For example: |
| 152 |
|
| 153 |
.. code-block:: c++ |
| 154 |
|
| 155 |
void f(std::optional<int> opt) { |
| 156 |
DCHECK(opt.has_value()); |
| 157 |
use(*opt); |
| 158 |
} |
| 159 |
|
| 160 |
Ensure that a value exists, then access the value in a correlated branch |
| 161 |
------------------------------------------------------------------------ |
| 162 |
|
| 163 |
The check is aware of correlated branches in the code and can figure out |
| 164 |
when an optional object is ensured to have a value on all execution |
| 165 |
paths that lead to an access. For example: |
| 166 |
|
| 167 |
.. code-block:: c++ |
| 168 |
|
| 169 |
void f(std::optional<int> opt) { |
| 170 |
bool safe = false; |
| 171 |
if (opt.has_value() && SomeOtherCondition()) { |
| 172 |
safe = true; |
| 173 |
} |
| 174 |
// ... more code... |
| 175 |
if (safe) { |
| 176 |
use(*opt); |
| 177 |
} |
| 178 |
} |
| 179 |
|
| 180 |
Stabilize function results |
| 181 |
~~~~~~~~~~~~~~~~~~~~~~~~~~ |
| 182 |
|
| 183 |
Since function results are not assumed to be stable across calls, it is best to |
| 184 |
store the result of the function call in a local variable and use that variable |
| 185 |
to access the value. For example: |
| 186 |
|
| 187 |
.. code-block:: c++ |
| 188 |
|
| 189 |
void f(Foo foo) { |
| 190 |
if (const auto& foo_opt = foo.opt(); foo_opt.has_value()) { |
| 191 |
use(*foo_opt); |
| 192 |
} |
| 193 |
} |
| 194 |
|
| 195 |
Do not rely on uncommon-API invariants |
| 196 |
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ |
| 197 |
|
| 198 |
When uncommon APIs guarantee that an optional has contents, do not rely on it -- |
| 199 |
instead, check explicitly that the optional object has a value. For example: |
| 200 |
|
| 201 |
.. code-block:: c++ |
| 202 |
|
| 203 |
void f(Foo foo) { |
| 204 |
if (const auto& property = foo.GetProperty("bar")) { |
| 205 |
use(*property); |
| 206 |
} |
| 207 |
} |
| 208 |
|
| 209 |
instead of the `HasProperty`, `GetProperty` pairing we saw above. |
| 210 |
|
| 211 |
Do not rely on caller-performed checks |
| 212 |
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ |
| 213 |
|
| 214 |
If you know that all of a function's callers have checked that an optional |
| 215 |
argument has a value, either change the function to take the value directly or |
| 216 |
check the optional again in the local scope of the callee. For example: |
| 217 |
|
| 218 |
.. code-block:: c++ |
| 219 |
|
| 220 |
void g(int val) { |
| 221 |
use(val); |
| 222 |
} |
| 223 |
|
| 224 |
void f(std::optional<int> opt) { |
| 225 |
if (opt.has_value()) { |
| 226 |
g(*opt); |
| 227 |
} |
| 228 |
} |
| 229 |
|
| 230 |
and |
| 231 |
|
| 232 |
.. code-block:: c++ |
| 233 |
|
| 234 |
struct S { |
| 235 |
std::optional<int> opt; |
| 236 |
int x; |
| 237 |
}; |
| 238 |
|
| 239 |
void g(const S &s) { |
| 240 |
if (s.opt.has_value() && s.x > 10) { |
| 241 |
use(*s.opt); |
| 242 |
} |
| 243 |
|
| 244 |
void f(S s) { |
| 245 |
if (s.opt.has_value()) { |
| 246 |
g(s); |
| 247 |
} |
| 248 |
} |
| 249 |
|
| 250 |
Additional notes |
| 251 |
~~~~~~~~~~~~~~~~ |
| 252 |
|
| 253 |
Aliases created via ``using`` declarations |
| 254 |
------------------------------------------ |
| 255 |
|
| 256 |
The check is aware of aliases of optional types that are created via |
| 257 |
``using`` declarations. For example: |
| 258 |
|
| 259 |
.. code-block:: c++ |
| 260 |
|
| 261 |
using OptionalInt = std::optional<int>; |
| 262 |
|
| 263 |
void f(OptionalInt opt) { |
| 264 |
use(opt.value()); // unsafe: it is unclear whether `opt` has a value. |
| 265 |
} |
| 266 |
|
| 267 |
Lambdas |
| 268 |
------- |
| 269 |
|
| 270 |
The check does not currently report unsafe optional accesses in lambdas. |
| 271 |
A future version will expand the scope to lambdas, following the rules |
| 272 |
outlined above. It is best to follow the same principles when using |
| 273 |
optionals in lambdas. |
| 274 |
|
| 275 |
Access with ``operator*()`` vs. ``value()`` |
| 276 |
------------------------------------------- |
| 277 |
|
| 278 |
Given that ``value()`` has well-defined behavior (either throwing an exception |
| 279 |
or terminating the program), why treat it the same as ``operator*()`` which |
| 280 |
causes undefined behavior (UB)? That is, why is it considered unsafe to access |
| 281 |
an optional with ``value()``, if it's not provably populated with a value? For |
| 282 |
that matter, why is ``CHECK()`` followed by ``operator*()`` any better than |
| 283 |
``value()``, given that they are semantically equivalent (on configurations that |
| 284 |
disable exceptions)? |
| 285 |
|
| 286 |
The answer is that we assume most users do not realize the difference between |
| 287 |
``value()`` and ``operator*()``. Shifting to ``operator*()`` and some form of |
| 288 |
explicit value-presence check or explicit program termination has two |
| 289 |
advantages: |
| 290 |
|
| 291 |
* Readability. The check, and any potential side effects like program |
| 292 |
shutdown, are very clear in the code. Separating access from checks can |
| 293 |
actually make the checks more obvious. |
| 294 |
|
| 295 |
* Performance. A single check can cover many or even all accesses within |
| 296 |
scope. This gives the user the best of both worlds -- the safety of a |
| 297 |
dynamic check, but without incurring redundant costs. |