warn about initializers containing more elements than initialized value

[llvmbot]

Bugzilla Link	9669
Resolution	FIXED
Resolved on	Apr 11, 2011 14:49
Version	trunk
OS	Linux
Reporter	LLVM Bugzilla Contributor
CC	@DougGregor,@efriedma-quic,@tkremenek

Extended Description

Hi there, I was wondering if it's sensible to add a compile-time check for out of bounds definitions for arrays like g++ already does for strings.

impulze@localhost ~ $ printf 'int main(){char f[1]="f";}' | g++ -x c++ -
: In function ‘int main()’:
:1:22: error: initializer-string for array of chars is too long

impulze@localhost ~ $ printf 'int main(){char f[1]="f";}' | clang++ -x c++ -

impulze@localhost ~ $ clang ++ --version
clang version 3.0 (/var/cache/paludis/distfiles/scm/clang f65076ebf5e8bc5ae7b77fe3c3a65dc5db8d763f)
Target: x86_64-pc-linux-gnu
Thread model: posix

Not sure if this is maybe related tohttp://llvm.org/bugs/show_bug.cgi?id=5028

[lattner]

Try out -fcatch-undefined-behavior

[lattner]

Also, -Warray-bounds. But sorry, this is a specific case with initializers. We should handle this.

[efriedma-quic]

The given construct is legal in C, but not C++. Taking a look.

[efriedma-quic]

r129260.

[tkremenek]

The given construct is legal in C, but not C++. Taking a look.

While legal in C, shouldn't we still warn? The code clearly doesn't do what the developer intended.

[tkremenek]

Re-opened for further discussion.

[efriedma-quic]

I'm pretty sure we'd get a lot of useless warnings if we warn by default in C...

[tkremenek]

I'm pretty sure we'd get a lot of useless warnings if we warn by default in
C...

Why is that? How would:

char f[1] = "f";

ever do what someone intended?

[efriedma-quic]

One example from ffmpeg:
const char ff_id3v2_tags[][4] = {
"TALB", "TBPM", "TCOM", "TCON", "TCOP", "TDLY", "TENC", "TEXT",
"TFLT", "TIT1", "TIT2", "TIT3", "TKEY", "TLAN", "TLEN", "TMED",
"TOAL", "TOFN", "TOLY", "TOPE", "TOWN", "TPE1", "TPE2", "TPE3",
"TPE4", "TPOS", "TPUB", "TRCK", "TRSN", "TRSO", "TSRC", "TSSE",
{ 0 },
};

One example from gcc:
static const char template[IDENT_LENGTH] = "gpch.012";

It's a useful construct if you know what you're doing; not all char arrays need to be null-terminated.

[tkremenek]

One example from ffmpeg:
const char ff_id3v2_tags[][4] = {
"TALB", "TBPM", "TCOM", "TCON", "TCOP", "TDLY", "TENC", "TEXT",
"TFLT", "TIT1", "TIT2", "TIT3", "TKEY", "TLAN", "TLEN", "TMED",
"TOAL", "TOFN", "TOLY", "TOPE", "TOWN", "TPE1", "TPE2", "TPE3",
"TPE4", "TPOS", "TPUB", "TRCK", "TRSN", "TRSO", "TSRC", "TSSE",
{ 0 },
};

One example from gcc:
static const char template[IDENT_LENGTH] = "gpch.012";

It's a useful construct if you know what you're doing; not all char arrays need
to be null-terminated.

Fair enough. It's unfortunate that there is no difference between a char array and a C string in the C type system.