Upc2c issue 52

[PHHargrove]

Please consider this replacement implementation of StmtPrinter::VisitImplicitValueInitExpr() which resolves issue #52 in the upc2c tracker.

-Paul

[PHHargrove]

Please HOLD OFF on this pull request. I am getting crashes in upc2c on the bug858a.c testcase that are not present w/o my changes.

[PHHargrove]

The following is sufficient to trigger the crash:

int main(void) {
    typedef char type_f2 [6];
    type_f2 f2[3] = { [1] = {[2] = 3,4} };
}

[PHHargrove]

The crash has been fixed by my latest commit. The issue was typedef'ed array types. Please resume consideration of this pull request.

[swatanabe]

[steven@localhost ~]$ cat test.upc
struct X { } arr[2] = { [1] = {} };
[steven@localhost ~]$ intrepid/build/bin/upc2c test.upc 
Segmentation fault (core dumped)

Note that the berkeley translator rejects this code, saying

Error during remote HTTP translation:
upcc: error during UPC-to-C translation (sgiupc stage): 
/tmp/web-trans-S3YD0q.upc:1: Incomplete or unknown type: X

[PHHargrove]

Steve,

Was that crash w/ or w/o my changes applied?

Reminder that '{}' as an initializer-list is NOT valid ISO C99 (even if the empty struct is, which I am uncertain of):

$ gcc -std=c99 -pedantic -c test.c
test.c:1:8: warning: struct has no members [-pedantic]
test.c:1:31: warning: ISO C forbids empty initializer braces [-pedantic]

Of course, regardless of the legality of the code, no input should crash the compiler. I will determine if my version of the code produces this SEGV and fix if so.

[swatanabe]

This is with your changes. The problem is in accessing the first member which doesn't exist for an empty struct.

[PHHargrove]

This is with your changes. The problem is in accessing the first member which doesn't exist for an empty struct.

Right, thanks. Not sure if I can fix today but I will take responsibility for this.

FWIW: I am not sure there exists a portable initiailizer-list for an empty struct. So, there may be no choice but to emit the non-portable "{}" for this case (and only this one).

-Paul

[swatanabe]

On 02/21/2014 03:30 PM, Paul H. Hargrove wrote:

FWIW: I am not sure there exists a portable initiailizer-list for an empty struct. So, there may be no choice but to emit the non-portable "{}" for this case (and only this one).

Agree, with the addition of zero-size arrays, which clang also accepts.

[PHHargrove]

C99 doesn't allow an empty structure.

Quoting from ISO C99 6.7.2.1 (Structure and union specifiers), Semantics paragraph 7:

... The struct-declaration-list is a sequence of declarations for the members of the structure or union. If the struct-declaration-list contains no named members, the behavior is undefined. ...

Gcc documents empty structs as an extension: http://gcc.gnu.org/onlinedocs/gcc/Empty-Structures.html

So, I'll output the non-portable "{}" for this case, instead of crashing.

If there is a consensus to make empty-structs an error that should probably be done prior to upc2c.

[PHHargrove]

I have the empty struct under control (emit '{}' instead of crashing). Will push to by branch after removing my debug output.

The empty array case seems to hit a different bug:

$ cat t2.c
typedef int Z_t[0];
Z_t arr[2] = { [1] = {} };
$ ./bin/clang -c t2.c && echo OK - No error || echo Non-zero exit
OK - No error
$ ./bin/clangupc -x upc -c t2.c && echo OK - No error || echo Non-zero exit
OK - No error
$ ./bin/upc2c t2.c && echo OK - No error || echo Non-zero exit
error: initializer for aggregate with no elements requires explicit braces
1 error generated.
Non-zero exit

[swatanabe]

On 02/21/2014 04:20 PM, Paul H. Hargrove wrote:

The empty array case seems to hit a different bug:

error: initializer for aggregate with no elements requires explicit braces
1 error generated.

I didn't get that. What I used was

struct X { int arr[0]; } test[2] = { [1] = {} };

which becomes:

struct X { int arr[0]; } test[2] = { /implicit/{{0}}, { } };

which gives me:

[steven@localhost ~]$ gcc test.c -c test.c:3:1: warning: excess elements in array initializer [enabled by default] } test[2] = { /implicit/{{0}}, { } }; ^

In Christ, Steven Watanabe

[PHHargrove]

I wrote:

The empty array case seems to hit a different bug

And that issue is present on the master branch too - not my fault. Will investigate later and open a new issue if it seems appropriate to do so. In the meantime I don't even know what the AST will look like for that code

Steve provided a different example: a struct containing an empty array (vs. my array of empty arrays). I will attack that case now...