On 08/26/18 05:34, Jeff Law wrote:
> On 08/24/2018 01:52 PM, Bernd Edlinger wrote:
>> Hi,
>>
>> this updated patch fixes one regression with current trunk due
>> to a new test case. Sorry for the confusion.
>>
>> The change to the previous version is:
>> 1) the check to avoid folding on empty char arrays is restored.
>> 2) A null-termination character is added except when the string is full
>> length.
>>
>>
>> Bootstrapped and reg-tested on x86_64-pc-linux-gnu.
>> Is it OK for trunk?
>>
>>
>> Thanks
>> Bernd.
>>
>>
>> patch-bracedstr-v2.diff
>>
>>
>> c-family:
>> 2018-08-24 Bernd Edlinger <bernd.edlin...@hotmail.de>
>>
>> * c-common.c (braced_list_to_string): Remove eval parameter.
>> Add some more checks. Always create zero-terminated STRING_CST.
>> * c-common.h (braced_list_to_string): Adjust prototype.
>>
>> c:
>> 2018-08-24 Bernd Edlinger <bernd.edlin...@hotmail.de>
>>
>> * c-decl.c (finish_decl): Call braced_list_to_string here ...
>> * c-parser.c (c_parser_declaration_or_fndef): ... instead of here.
>>
>>
>> cp:
>> 2018-08-24 Bernd Edlinger <bernd.edlin...@hotmail.de>
>>
>> * decl.c (eval_check_narrowing): Remove.
>> (check_initializer): Move call to braced_list_to_string from here ...
>> * typeck2.c (store_init_value): ... to here.
>> (digest_init_r): Remove handing of signed/unsigned char strings.
>>
>> testsuite:
>> 2018-08-24 Bernd Edlinger <bernd.edlin...@hotmail.de>
>>
>> * c-c++-common/array-init.c: New test.
>> * g++.dg/init/string2.C: Remove xfail.
> My concern here is that you removed code that was explicitly added
> during the initial review work of the patch that turned braced
> initializers into strings.
>
Yes, you mean BRACE_ENCLOSED_INITIALIZER_P
which is (TREE_CODE (NODE) == CONSTRUCTOR && TREE_TYPE (NODE) ==
init_list_type_node)
I tried that first but the TREE_TYPE of the CONSTRUCTOR was no longer
init_list_type_node
at that point. I think the middle-end needs the structure type here, and
digest_init must
have fixed that.
This did not break in the debugger:
+ if (TREE_CODE (type) == ARRAY_TYPE
+ && TREE_CODE (value) == CONSTRUCTOR)
+ {
+ tree typ1 = TYPE_MAIN_VARIANT (TREE_TYPE (type));
+ if (typ1 == char_type_node
+ || typ1 == signed_char_type_node
+ || typ1 == unsigned_char_type_node)
+ {
+ if (BRACE_ENCLOSED_INITIALIZER_P(value))
+ asm("int3");
+ value = braced_list_to_string (type, value);
+ }
+ }
+
while this
+ if (!BRACE_ENCLOSED_INITIALIZER_P(value))
+ printf("%p - %p\n", TREE_TYPE(value), init_list_type_node);
does this:
$ cat test.cc
char x[] = {1,2,3};
$ ../gcc-build/gcc/xgcc -B ../gcc-build/gcc/ -S test.cc
0x7fdedb821b28 - 0x7fdedb81f738
while the string is folded as expected.
>
>> - a[] = { 1, 2, 333, 0 }; // { dg-warning
>> "\\\[\(-Wnarrowing|-Woverflow\)" "" { target { ! c++98_only } } }
>> + a[] = { 1, 2, 333, 0 }; // { dg-warning
>> "\\\[\(-Wnarrowing|-Woverflow\)" }
> This is not an XFAIL, this is a selector. Essentially it says that the
> diagnostic is appropriate when not in c++98 mode.
>
> You can see that to be the case if you compile the test in c++98 mode
> without your change. It will compile with no errors or warnings.
>
> However, after your change it issues a warning for the narrowing
> conversion in c++98 mode, which AFAICT it should not do.
>
This just restores the state _before_ Martin's braced initializer patch.
So I have the impression that is actually a regression due to the
original braced initializer patch.
It is unfortunate that Martin did not check that.
Bernd.
