Control: forwarded -1 http://www.cmake.org/Bug/view.php?id=15736
Hi Bob,
On 10.09.2015 14:50, Bob Tracy wrote:
> I think I may have found the issue. Not sure if other architectures
> could be affected.
>
> Looking at the cmake build artifacts under the "Build" directory, there
> are several instances of "feature_tests.c" and "feature_tests.cxx" with
> "const char features[]" initialized to '{ "" ... }' where I'm reasonably
> certain the intent is for the first byte of the "features" array to be a
> NULL. When "strings" is run on the resulting ".bin" file to extract the
> tokens of interest, there's no leading NULL byte in front of what is
> supposed to be the first non-NULL character in "features": that's where
> the mysterious leading "k" is coming from, and why only the first token
> in the array is "corrupted" in that manner. In other words, whichever
> feature appears first in the "const char features[]" array is going to
> be the one flagged as not supported.
>
> Replacing "" with "\0" correctly produces the intended NULL byte at the
> start of the array, and ultimately allows a working "cmake" to be built
> on Alpha.
>
> The question then is whether "" is supposed to compile to a NULL byte in
> the context of the "const char features[]" array, and whether this
> compiler behavior applies to other architectures. My fear is that we're
> simply getting lucky on x86, and any NULL in front of the array is there
> by accident.
>
> Not sure where/how to apply a proper fix.
Thanks for the analysis, I've forwarded this upstream.
Cheers,
Felix
