[stmt.dcl] Add example of static/thread_local block variable initialization #6508
Conversation
…zation Update source/statements.tex Co-authored-by: A. Jiang <de34@live.cn>
Eisenwave
force-pushed
the
static-block-init-example
branch
from
d0e2f5d
to
30b281d
Compare
|
Updated version |
| // initialization of \tcode{s}, \tcode{s} remains uninitialized, and | ||
| // an attempt to initialize \tcode{s} (but not \tcode{x}) will be | ||
| // made the next time \tcode{f} is called. | ||
| thread_local const char* w = "world"; // OK, initializes \tcode{w} to \tcode{"world"} on the main thread. |
There was a problem hiding this comment.
w is still constant-initialized. Can we show non-constant initialization here? (e.g. thread_local const char* w = std::system_category().name();).
There was a problem hiding this comment.
Its not required to, and yes.
There was a problem hiding this comment.
Its not required to
I don't see why it is not required to. [expr.const] p2.2 and p13.3 seemly require that w is constant-initialized when the initializer is "world".
There was a problem hiding this comment.
I was thinking of the rules for static initialization, which only require that dynamic initialization may (but is not required to) be omitted when the initializer is a constant expression. I wasn't aware of [expr.const] p2.2.
There was a problem hiding this comment.
I don't see why it is not required to.
@frederick-vs-ja actually wait, I am right. It is true that e.g. int x = 0; in block scope is constant-initialized, but this is meaningless for block variables. Something being constant-initialized doesn't imply that it's constant initialized (https://eel.is/c++draft/basic.start.static#2). (pay attention to the hyphenation, constant-initialized and constant initialization are two completely different things)
There was a problem hiding this comment.
In this example we're initializing a variable with thread storage duration. So if it is constant-initialized, it has constant initialization.
| thread_local int k = 4; | ||
| std::cout << k + x; // OK, prints \tcode{"6"}. | ||
| std::cout << s << ' '; // OK, prints \tcode{"hello "}. | ||
| std::cout << w; // Undefined behavior if \tcode{w} hasn't been constant initialized\iref{basic.start.static}. |
There was a problem hiding this comment.
cplusplus/CWG#372 may be involved. But if we change w to have dynamic initialization, it seems safe to say this line raises UB unconditionally - a zero-initialized (null) const char* value is sufficient to raise UB.
This edit adds an example which clarifies at least two potential questions:
It also raises awareness for a particularly dangerous edge case, namely: