clang ("foo") accepted as char[] initializer

[efriedma-quic]

Bugzilla Link	8046
Version	unspecified
OS	Linux

Extended Description

Testcase:
char arr[] = ("foo");

clang gives no error. Strictly speaking, this isn't legal; Comeau gives "parentheses around a string initializer are nonstandard".

[arsenm]

Still no error given

[shafik]

Confirmed: https://godbolt.org/z/ccTY1KnW6

gcc give the following diagnostic w/ -pedantic:

<source>:1:6: warning: array 'arr' initialized by parenthesized string literal '("foo")'
    1 | char arr[] = ("foo");
      |      ^~~

[frederick-vs-ja]

I think we should submit a CWG issue to clarify that parenthesized string literal is legal here. It seems that many modern C++ compilers silently accept the example.

[shafik]

Looks like edg reject w/ -strict: https://godbolt.org/z/s77vb5jTn

It is an interesting extension and it seems to be used a lot in the wild:

https://sourcegraph.com/search?q=context:global+lang:c%2B%2B+/%5C%5B%5C%5D%5Cs%2B%5C%3D%5Cs%2B%5C%28%22.*%22%5C%29/+count:11000&patternType=keyword&sm=0

It would be nice to know why it came to be.

[philnik777]

Looks like edg reject w/ -strict: https://godbolt.org/z/s77vb5jTn

It is an interesting extension and it seems to be used a lot in the wild:

https://sourcegraph.com/search?q=context:global+lang:c%2B%2B+/%5C%5B%5C%5D%5Cs%2B%5C%3D%5Cs%2B%5C%28%22.*%22%5C%29/+count:11000&patternType=keyword&sm=0

It would be nice to know why it came to be.

Am I missing something? 124 results (ignoring clang's own constant-expression-cxx11.cpp) doesn't seem like a lot to me. That seems to me like the kind of scale where you could conceivably make this an error by default and then remove it in a few releases.

[shafik]

Looks like edg reject w/ -strict: https://godbolt.org/z/s77vb5jTn
It is an interesting extension and it seems to be used a lot in the wild:
https://sourcegraph.com/search?q=context:global+lang:c%2B%2B+/%5C%5B%5C%5D%5Cs%2B%5C%3D%5Cs%2B%5C%28%22.*%22%5C%29/+count:11000&patternType=keyword&sm=0
It would be nice to know why it came to be.

Am I missing something? 124 results (ignoring clang's own constant-expression-cxx11.cpp) doesn't seem like a lot to me. That seems to me like the kind of scale where you could conceivably make this an error by default and then remove it in a few releases.

Those are public hits, so perhaps there are more we don't have access to. The fact that implementations mostly seem to support it give me some pause but perhaps folks have phased it out over time and the need it filled does not apply today, don't know.