Re: flatten expr.h (version 2)

[Prathamesh Kulkarni <prathamesh dot kulkarni at linaro dot org>]

On 14 January 2015 at 16:40, Prathamesh Kulkarni
<prathamesh.kulkarni@linaro.org> wrote:
> On 14 January 2015 at 14:34, Richard Biener <rguenther@suse.de> wrote:
>> On Wed, 14 Jan 2015, Prathamesh Kulkarni wrote:
>>
>>> On 13 January 2015 at 22:02, Prathamesh Kulkarni
>>> <prathamesh.kulkarni@linaro.org> wrote:
>>> > On 13 January 2015 at 16:06, Prathamesh Kulkarni
>>> > <prathamesh.kulkarni@linaro.org> wrote:
>>> >> On 13 January 2015 at 15:34, Richard Biener <rguenther@suse.de> wrote:
>>> >>> On Sun, 11 Jan 2015, Prathamesh Kulkarni wrote:
>>> >>>
>>> >>>> Hi,
>>> >>>> This is a revamped expr.h flattening flattening patch rebased on
>>> >>>> tree.h and tree-core.h flattening patch (r219402).
>>> >>>> It depends upon the following patch to get committed.
>>> >>>> https://gcc.gnu.org/ml/gcc-patches/2015-01/msg00565.html
>>> >>>>
>>> >>>> Changes:
>>> >>>> * Removed all includes except tree-core.h. Put includes required by
>>> >>>> expr.h in a comment.
>>> >>>> * Moved stmt.c, expmed.c prototypes to stmt.h, expmed.h respectively.
>>> >>>> * Adjusted generator programs: genemit.c, gengtype.c, genopinit.c, genoutput.c.
>>> >>>> * Did not put includes in gcc-plugin.h since expr.h cannot be included
>>> >>>> by plugins
>>> >>>> (putting them broke building a file in c-family/ since expr.h is not
>>> >>>> allowed in front-ends)
>>> >>>> * Affects java front-end (expr.h is allowed in java front-end).
>>> >>>>
>>> >>>> Bootstrapped and tested on x86_64-unknown-linux-gnu with languages:
>>> >>>> all,go,ada,jit
>>> >>>> Built on all targets in config-list.mk with languages: all, go.
>>> >>>> OK to commit ?
>>> >>>
>>> >>> diff --git a/gcc/expr.c b/gcc/expr.c
>>> >>> index fc22862..824541e 100644
>>> >>> --- a/gcc/expr.c
>>> >>> +++ b/gcc/expr.c
>>> >>> @@ -41,11 +41,17 @@ along with GCC; see the file COPYING3.  If not see
>>> >>>  #include "regs.h"
>>> >>>  #include "hard-reg-set.h"
>>> >>>  #include "except.h"
>>> >>> -#include "input.h"
>>> >>>  #include "function.h"
>>> >>>  #include "insn-config.h"
>>> >>>  #include "insn-attr.h"
>>> >>>  /* Include expr.h after insn-config.h so we get HAVE_conditional_move.
>>> >>> */
>>> >>> +#include "hashtab.h"
>>> >>> +#include "emit-rtl.h"
>>> >>> +#include "expmed.h"
>>> >>> +#include "stmt.h"
>>> >>> +#include "statistics.h"
>>> >>> +#include "real.h"
>>> >>> +#include "fixed-value.h"
>>> >>>  #include "expr.h"
>>> >>>
>>> >>> Please move the comment to the proper place
>>> >> ah, my flattening tool doesn't look at comments. I will move the
>>> >> comment before expr.h include, thanks.
>>> >>>
>>> >>> diff --git a/gcc/expr.h b/gcc/expr.h
>>> >>> index a7638b8..f1be8dc 100644
>>> >>> --- a/gcc/expr.h
>>> >>> +++ b/gcc/expr.h
>>> >>> @@ -20,7 +20,8 @@ along with GCC; see the file COPYING3.  If not see
>>> >>>  #ifndef GCC_EXPR_H
>>> >>>  #define GCC_EXPR_H
>>> >>>
>>> >>> -/* For inhibit_defer_pop */
>>> >>> +/* expr.h required includes */
>>> >>> +#if 0
>>> >>>  #include "hashtab.h"
>>> >>>  #include "hash-set.h"
>>> >>>  #include "vec.h"
>>> >>> @@ -29,15 +30,17 @@ along with GCC; see the file COPYING3.  If not see
>>> >>>  #include "hard-reg-set.h"
>>> >>>  #include "input.h"
>>> >>>  #include "function.h"
>>> >>> -/* For XEXP, GEN_INT, rtx_code */
>>> >>>  #include "rtl.h"
>>> >>> -/* For optimize_size */
>>> >>>  #include "flags.h"
>>> >>> -/* For tree_fits_[su]hwi_p, tree_to_[su]hwi, fold_convert, size_binop,
>>> >>> -   ssize_int, TREE_CODE, TYPE_SIZE, int_size_in_bytes,    */
>>> >>>  #include "tree-core.h"
>>> >>> -/* For GET_MODE_BITSIZE, word_mode */
>>> >>>  #include "insn-config.h"
>>> >>> +#include "alias.h"
>>> >>> +#include "emit-rtl.h"
>>> >>> +#include "expmed.h"
>>> >>> +#include "stmt.h"
>>> >>> +#endif
>>> >>>
>>> >>> Err, please remove the #if 0 section
>>> >> I kept it because if something breaks later (hopefully not!), it will
>>> >> be easier to fix.
>>> >> I will remove it.
>>> >>>
>>> >>> +
>>> >>> +#include "tree-core.h"
>>> >>>
>>> >>> Why?  The original comment says
>>> >>>
>>> >>> -/* For tree_fits_[su]hwi_p, tree_to_[su]hwi, fold_convert, size_binop,
>>> >>> -   ssize_int, TREE_CODE, TYPE_SIZE, int_size_in_bytes,    */
>>> >>>
>>> >>> but all those are declared in tree.h.  Which means the files including
>>> >>> expr.h must already include tree.h.
>>> >>>
>>> >>> If that's not the reason we need to include tree-core.h from expr.c
>>> >>> please add a comment explaining why.
>>> >> bt-load.c fails to compile because it includes expr.h but does not
>>> >> include tree.h
>>> >> I will place tree.h include in all files that include expr.h and rebuild.
>>> > This is not going to work, since tree.h is now flattened. Shall also
>>> > require including all headers required by
>>> > tree.h in all files that include expr.h. Could we retain tree-core.h
>>> > in expr.h for now ?
>>> > Or should I insert tree.h (along with tree.h required includes) in all
>>> > files that include expr.h ?
>>> I am including tree.h along with required includes in all files that
>>> include expr.h.
>>> This removes all includes from expr.h.
>>
>> Good.
> The attached patch, refactors expr.h so that it only exports
> prototypes from expr.c.
> I kept int_expr_size() in explow.c instead of moving it to expr.c because it
> called tree_expr_size() which is static in explow.c and has other callers too.
Not really. It only has two callers expr_size() and int_expr_size().
I will move definitions of int_expr_size(), tree_expr_size() and
expr_size() to expr.c
in followup patch.
>
> Changes:
> * New header files dojump.h, explow.h.
> * Modified generator programs: gengtype.c, genemit.c, genopinit.c,
> genoutput.c, config/tilepro/gen-mul-tables.cc.
> * Affects java front-end (java/builtins.c includes expr.h).
>
> Bootstrapped and tested on x86_64-unknown-linux-gnu with languages:
> all, go, ada, jit, obj-c++.
> Build in progress on all targets in config-list.mk with languages: all, go.
> Assuming the build goes fine, OK to commit ?
>
> Thanks,
> Prathamesh
>>
>> Richard.