Re: [CIL users] _inline keyword

[Matt H. <ma...@cs...>]

Peter Hawkins wrote:
> Hi...
> 
> The following abomination turns up in linux-2.6.17.1/fs/jfs/jfs_incore.h:
>         struct {
>             unchar _unused[16]; /* 16: */
>             dxd_t _dxd;     /* 16: */
>             unchar _inline[128];    /* 128: inline symlink */
>             /* _inline_ea may overlay the last part of
>              * file._xtroot if maxentry = XTROOTINITSLOT
>              */
>             unchar _inline_ea[128]; /* 128: inline extended attr */
>         } link;
> 
> Unfortunately CIL parses the identifier "_inline" as the MSVC inline 
> keyword, whereas gcc accepts it as a valid identifier (as opposed to 
> inline, __inline, or __inline__). This is a case where accepting the 
> union of gcc and msvc syntax doesn't really cut it. To fix this the 
> lexer needs to have a notion of compiler personality, which I suspect 
> needs to reach further than this one issue.
> 
> Please fix?
> 
> Thanks,
> Peter
> 

Hi Peter,

Thanks for your recent patches.

There are several places where CIL's lexer and parser depend on the 
underlying compiler.  For example, "__thread" is a keyword on recent 
versions of gcc, but not older versions.  Here's a patch for _inline.

Cheers,
Matt

--- cil/src/frontc/clexer.mll	23 May 2006 21:31:21 -0000	1.106
+++ cil/src/frontc/clexer.mll	27 Jun 2006 19:55:46 -0000	1.107
@@ -146,7 +146,11 @@
        ("__inline__", fun loc -> INLINE loc);
        ("inline", fun loc -> INLINE loc);
        ("__inline", fun loc -> INLINE loc);
-      ("_inline", fun loc -> INLINE loc);
+      ("_inline", fun loc ->
+                      if !Cprint.msvcMode then
+                        INLINE loc
+                      else
+                        IDENT ("_inline", loc));
        ("__attribute__", fun loc -> ATTRIBUTE loc);
        ("__attribute", fun loc -> ATTRIBUTE loc);