Re: PR middle-end/18071 -Winline does not respect -fno-default-inline

["Manuel López-Ibáñez" <lopezibanez at gmail dot com>]

I hate to forget the patch :-(

Very sorry,

Manuel.


On 16/01/07, Manuel López-Ibáñez <lopezibanez@gmail.com> wrote:
> :ADDPATCH middle-end/18071:
>
> Winline complains about methods not inlined even when using
> -fno-default-inline. We have two bits in our tree representation:
>
> DECL_DECLARED_INLINE_P : Nonzero in a FUNCTION_DECL means that this
> function was declared inline, such as via the `inline' keyword in
> C/C++. This flag controls the linkage semantics of 'inline'; whether
> or not the function is inlined is controlled by DECL_INLINE.
>
> DECL_INLINE : Nonzero in a FUNCTION_DECL means this function can be
> substituted where it is called.
>
> Currently, with -fdefault-inline (the default) we set both to 1 for
> methods (even if the methods have no explicit "inline" keyword. With
> -fno-default-inline we set only DECL_DECLARED_INLINE_P.
>
> The definition of Winline is "Warn if a function can not be inlined
> and it was declared as inline." It seems that we should only warn if
> DECL_DECLARED_INLINE_P and don't care about the value of DECL_INLINE.
>
> So, this patch does two things:
>
> * It makes sure we emit the warning only if DECL_DECLARED_INLINE_P is
> true without taking into account the value of DECL_INLINE.
>
> * It does not set DECL_DECLARED_INLINE_P for methods unless they are
> declared inline or -fdefault-inline is true (which is the default).
>
> I am not sure whether the latter breaks the current description of
> -fno-default-inline:  "Note that these functions will have linkage
> like inline functions; they just won't be inlined by default".
>
> I also would like to test whether the methods were inlined by default
> and when using the keyword inline and -fno-default-inline.
> Unfortunately, I am not sure how to do this (looking in the testsuite
> didn't give any ideas). Also, I am not sure how to force the warnings
> to appear, so we can check that they are emitted when they should.
>
> So suggestions and comments are very welcome.
>
> Bootstrapped and regression tested with --enable-languages=all for
> revision 120770 on i686-pc-linux-gnu.
>
> OK for mainline?
>
> Manuel.
>
>
>
> 2007-01-16  Manuel Lopez-Ibanez  <manu@gcc.gnu.org>
>
>   PR middle-end/18071
>   * c-objc-common.c (c_cannot_inline_tree_fn): Internal notes like
> DECL_INLINE should not affect -Winline.
>   * tree-inline.c (inlinable_function_p): Likewise.
>
> cp/
>   * decl.c (start_method): Don't set DECL_DECLARED_INLINE_P if
> -fno-default-inline.
>
> testsuite/
>   * g++.dg/warn/Winline-5.C: New
>   * g++.dg/warn/Winline-6.C: New
>   * g++.dg/warn/Winline-7.C: New

Attachment: winline.diff
Description: Text document