Hi, LLVM. I have a question I'd like to get put into the official style guidelines: do we prefer NULL or 0 for C++ objects? I've seen both throughout the code.
Personally I prefer NULL, since it establishes that something is a pointer and not an integer (or integer-constructed object, but thankfully we avoid implicit conversions in LLVM/Clang). But I think I read somewhere that 0 is more C++esque. (And C++11 nullptr's not available yet, of course.)
Similarly, when testing for the null pointer, is it better to use:
1. (x) and (!x)
2. (x) and (x == NULL) // or 0
3. (x != NULL) and (x == NULL)
4. (NULL != x) and (NULL == x)
5. ...some other combination or something I haven't thought of
I don't care about this one as much. Still, in this case I think the first is fairly standard, though the second does make the explicit test for NULL stand out a little more (fewer copy/paste errors, perhaps?). IMO (3) and (4) are overkill.
Thoughts?
Jordy
But I think I read somewhere that 0 is more C++esque.
I believe Stroustrup espoused this at one point (perhaps even on his
website) on the basis that using NULL gives you a false sense of
security - which isn't entirely true now that compilers (GCC & clang
presumably) will warn you about using NULL in non-pointer contexts.
(And C++11 nullptr's not available yet, of course.)
My 2c is that this is a good reason to use NULL over 0 - it'll be way
easier to s/NULL/nullptr/ when upgrading to C++11 in the future than
it'll be to hunt down all those 0s used as pointers. Though I suppose
we can always enhance the compiler to warn about 0 used in pointer
contexts in C++11...
Hmm, we could do that today - hmm, is that something we should do? I
think it'd be rather neat/helpful - it could have a fixup & all that.
1. (x) and (!x)
This is my usual approach/preference.
Then you can also do things like:
if (T *x = foo())
...
using NULL gives you a false sense of security - which isn't entirely
true now that compilers (GCC & clang
presumably) will warn you about using NULL in non-pointer contexts.
Hmm, nope, can't seem to find that warning in clang - gcc gives it though:
foo.cpp:4:11: warning: converting to non-pointer type "int" from NULL
I use #4 most of the time.
Rationale:
* Make it explicit
* Constant on the left hand side.
-- Marshall
Marshall Clow Idio Software <mailto:mclow.lists@gmail.com>
A.D. 1517: Martin Luther nails his 95 Theses to the church door and is promptly moderated down to (-1, Flamebait).
-- Yu Suzuki
Please don't do this in LLVM code. Unless you're the terminator of a varargs call, I'd prefer to use 0 instead of NULL, and definitely constants on right hand sides of comparisons.
-Chris
For the curious, the way that gcc does this is by doing:
#define NULL __null
and then treating __null as 0 in all ways, except for the purposes of warning tracking. clang already encapsulates __null as GNUNullExpr, so in principle (although beyond me) it shouldn't be hard to add warnings, wherever it gets converted to an integer.
Chris