Re: [Tinycc-devel] gcc/tcc nested macros difference

[Michael Matz]

Hi,

On Sun, 24 Apr 2016, Edmund Grimley Evans wrote:

> The C standard has a somewhat unclear description of how the
> preprocessor should work, but there is general agreement about what
> was intended, and the general agreement leaves no real option for
> alternative behaviour. Therefore, if two modern C compilers preprocess
> differently, then one of them is buggy... I think it's TCC that is
> buggy in this case. Has TCC always had this bug, or was it introduced
> recently?

Actually, while the text of the standard is somewhat confusing, I think it 
requires the behaviour of GCC.  6.10.3.6 (c99) Example 3, lists (somewhat 
shortened) this example:

#define f(a) f(x * (a))
#define x 2
#define g f
#define t(a) a
t(t(g)(0) + t)(1)

which is required to expand to:
f(2 * (0)) + t(1)

from which we can infer that object-like macros _are_ considered for 
expansion of function-like invocations ("g(0)" is replaced by "f(2 * 
(0))", which can only happen if "g" is replaced by "f" first), while 
function like macros are _not_ considered for non-function-like 
invocations (that wouldn't make much sense anyway).

TCC get's this example correct as well, but problems appear when there are 
nested invocations of this rule, as your next example shows:

> Simplified test case:
>
> #define Y(x) Z(x)
> #define X Y
> X(X(foo))

which tcc replaces by "Z(Y(1))" showing an inconsistency in TCC itself. 
Both X should be replaced in the same way (ultimately to Z).

So, your bug entry seems quite appropriate.


Ciao,
Michael.