On Mon, 2014-09-22 at 12:26 -0400, Andrew MacLeod wrote: > After being reminded of the tm.h issues brought up last november (here: > https://gcc.gnu.org/ml/gcc-patches/2013-11/msg01731.html ), I started > looking back into it. > > The general summary is the any header file which has a conditional on a > target macro could be affected by include file reordering... ie, if a > header file has > #ifdef BLAH > <whatever> > > and we move the header files around a bit, it wouldn't be immediately > obvious if the order where changes and BLAH was define later in the > include structure. The results for this header files would be different > because <whatever> would no longer be in the preprocess source, and it > may take a long time to discover the difference. > > Josephs solution was to identify these and instead put a default > definition in default.h ... then change all the uses to #if instead.. ie, > #if BLAH > > This way we can ensure that the definition has been seen and it will be > a compile error if not. > > > I looked at all the target macros listed by Joseph, and decide to start > with the ones which are used *only* in an "if defined" situation in a .h > files of some sort. > ie > #ifdef > #ifndef > or #if define() > > If this happens in a .c file, then changing the order of .h's shouldn't > matter. (Unless the .c file is doing something really screwy. So for > the moment, ignoring that.)
There appears to be a particular implicit order in which headers must be included. I notice that e.g. tm.h has: #ifndef GCC_TM_H #define GCC_TM_H so if we're going with this "no header file includes any other header file" model, would it make sense to add something like: #ifndef GCC_TM_H #error tm.h must have been included by this point /* We need tm.h here so that we can see: BAR, BAZ, QUUX, etc. */ #endif to header files needing it, thus expressing the expected dependencies explicitly? > So looking at only .h files, I found a number of default definitions in > rtl.h, which this patch moves to defaults.h. I was going to #error if > defaults.h wasn't included, but found that to be unnecessary since it > uses some of those macros and would cause a compile failure anyway if it > wasn't included. All the other uses of these particular macros use the > #if model, so no further adjustment is required for them. > > This patch bootstraps on x86_64-unknown-linux-gnu, and regressions are > running. Assuming no issues show up, OK for mainline? > > The remaining macros I will have a closer look at and most will likely > need the full treatment moving from #ifdef to the #if model. I plan to > do these next. The macros which fit this category for potential header > trouble (along with their usage count) are: > > 2 USE_COLLECT2 > 3 TARGET_TERMINATE_DW2_EH_FRAME_INFO > 3 TARGET_HAVE_CTORS_DTORS > 3 REGMODE_NATURAL_SIZE > 4 CC_STATUS_MDEP_INIT > 4 CC_STATUS_MDEP_INIT > 4 MODE_BASE_REG_REG_CLASS > 4 REGNO_MODE_OK_FOR_REG_BASE_P > 5 MODE_BASE_REG_CLASS > 5 XCOFF_DEBUGGING_INFO > 6 VMS_DEBUGGING_INFO > 6 TARGET_ASM_OUTPUT_ANCHOR > 6 EH_FRAME_SECTION_NAME > 8 MODE_CODE_BASE_REG_CLASS > 9 HARD_REGNO_CALLER_SAVE_MODE > 9 HARD_REGNO_CALL_PART_CLOBBERED > 9 SDB_DEBUGGING_INFO > 9 CC_STATUS_MDEP > 9 REGNO_MODE_CODE_OK_FOR_BASE_P > 10 SECONDARY_RELOAD_CLASS > 10 TARGET_VXWORKS_RTP > 14 NO_DOT_IN_LABEL > 14 REGNO_MODE_OK_FOR_BASE_P > 17 STACK_REGS > 19 TARGET_ASM_DESTRUCTOR > 20 OBJECT_FORMAT_ELF > 20 TARGET_ASM_CONSTRUCTOR > 24 NO_DOLLAR_IN_LABEL > 26 DBX_DEBUGGING_INFO > 30 CPLUSPLUS_CPP_SPEC > 42 ASM_OUTPUT_DEF > > > >
