Re: GDB 13.1 and clang

[John Baldwin <jhb@FreeBSD.org>]

On 3/8/23 1:22 PM, Chris Johns wrote:
> On 9/3/2023 5:22 am, Simon Marchi wrote:
>> On 3/8/23 15:07, Chris Johns wrote:
>>> Hi,
>>>
>>> I have just tried to build the latest RTEMS tools and we now use gdb 13.1. I am
>>> seeing a build failure on FreeBSD 13.1 of:
>>>
>>> In file included from ../../gdb-13.1/gdb/xml-tdesc.c:23:
>>> In file included from ../../gdb-13.1/gdb/target.h:42:
>>> In file included from ../../gdb-13.1/gdb/infrun.h:21:
>>> In file included from ../../gdb-13.1/gdb/gdbthread.h:26:
>>> In file included from ../../gdb-13.1/gdb/breakpoint.h:38:
>>> ../../gdb-13.1/gdb/target/waitstatus.h:113:1: error: use of undeclared
>>> identifier 'DIAGNOSTIC_ERROR_SWITCH'
>>> DIAGNOSTIC_ERROR_SWITCH
>>> ^
>>>
>>> I only have clang installed on the FreeBSD machine.
>>>
>>> A quick review of include/diagnostics.h seems to show support for
>>> DIAGNOSTIC_ERROR_SWITCH only in the gcc area?
>>
>> Hmm, I see it in the clang section:
>>
>> # define DIAGNOSTIC_ERROR_SWITCH \
>>    DIAGNOSTIC_ERROR ("-Wswitch")
>>
> 
> It seems I need to do a better review before posting :)
> 
>> https://sourceware.org/cgit/binutils-gdb/tree/include/diagnostics.h?h=gdb-13.1-release&id=4f3e26ac6ee31f7bc4b04abd8bdb944e7f1fc5d2#n76
>>
>> Unless there's a typo I don't see.
>>
>> diagnostics.h is included at the top of waitstatus.h.  Is it possible
>> that another unrelated diagnostics.h gets included on FreeBSD?  You
>> could inspect the preprocessed file to see what the preprocessor
>> included for diagnostics.h.
> 
> There is another version installed here:
> 
> $ pkg which /usr/local/include/diagnostics.h
> /usr/local/include/diagnostics.h was installed by package binutils-2.37_2,1
> 
> We have /usr/local/include early in the include list on FreeBSD to pick up some
> of the required libraries installed as packages. I will check for a package
> update or I revisit the build flags to see how it can be handled.
> 
> Thank you for the prompt and helpful responses.

Yes, I manually delete all the headers installed by binutils into
/usr/local/include when building GDB from source locally to work
around this. :(  I'm not sure of the best fix for this.  I tracked
it down to the change to add zstd as ZSTD_CFLAGS is before BFD_CFLAGS
in INTERNAL_CFLAGS_BASE in Makefile.in in this commit:

commit 2cac01e3ffff74898c54fa5e6418817f5578adb6
Author: Fangrui Song <maskray@google.com>
Date:   Mon Sep 26 19:50:13 2022 -0700

     binutils, gdb: support zstd compressed debug sections
     
     PR29397 PR29563: Add new configure option --with-zstd which defaults to
     auto.  If pkgconfig/libzstd.pc is found, define HAVE_ZSTD and support
     zstd compressed debug sections for most tools.
     
     * bfd: for addr2line, objdump --dwarf, gdb, etc
     * gas: support --compress-debug-sections=zstd
     * ld: support ELFCOMPRESS_ZSTD input and --compress-debug-sections=zstd
     * objcopy: support ELFCOMPRESS_ZSTD input for
       --decompress-debug-sections and --compress-debug-sections=zstd
     * gdb: support ELFCOMPRESS_ZSTD input.  The bfd change references zstd
       symbols, so gdb has to link against -lzstd in this patch.
     
     If zstd is not supported, ELFCOMPRESS_ZSTD input triggers an error.  We
     can avoid HAVE_ZSTD if binutils-gdb imports zstd/ like zlib/, but this
     is too heavyweight, so don't do it for now.
     
     ```
     % ld/ld-new a.o
     ld/ld-new: a.o: section .debug_abbrev is compressed with zstd, but BFD is not built with zstd support
     ...
     
     % ld/ld-new a.o --compress-debug-sections=zstd
     ld/ld-new: --compress-debug-sections=zstd: ld is not built with zstd support
     
     % binutils/objcopy --compress-debug-sections=zstd a.o b.o
     binutils/objcopy: --compress-debug-sections=zstd: binutils is not built with zstd support
     
     % binutils/objcopy b.o --decompress-debug-sections
     binutils/objcopy: zstd.o: section .debug_abbrev is compressed with zstd, but BFD is not built with zstd support
     ...
     ```

I have not tried moving ZSTD_CFLAGS after BFD_CFLAGS, but if it worked I
think the right solution is to ensure all the "local" includes for things
in git come first before any includes that pull from the installed system.
I think this would mean moving READLINE_CFLAGS (for systems that use a
system readline), ZLIBINC, and ZSTD_CFLAGS after INCLUDE_CFLAGS.

I just tried this and it worked for me locally:

diff --git a/gdb/Makefile.in b/gdb/Makefile.in
index 6e39383eb93..40497541880 100644
--- a/gdb/Makefile.in
+++ b/gdb/Makefile.in
@@ -629,8 +629,8 @@ INTERNAL_CPPFLAGS = $(CPPFLAGS) @GUILE_CPPFLAGS@ @PYTHON_CPPFLAGS@ \
  # INTERNAL_CFLAGS is the aggregate of all other *CFLAGS macros.
  INTERNAL_CFLAGS_BASE = \
  	$(GLOBAL_CFLAGS) $(PROFILE_CFLAGS) \
-	$(GDB_CFLAGS) $(OPCODES_CFLAGS) $(READLINE_CFLAGS) $(ZLIBINC) \
-	$(ZSTD_CFLAGS) $(BFD_CFLAGS) $(INCLUDE_CFLAGS) $(LIBDECNUMBER_CFLAGS) \
+	$(GDB_CFLAGS) $(OPCODES_CFLAGS) $(BFD_CFLAGS) $(INCLUDE_CFLAGS) \
+	$(READLINE_CFLAGS) $(ZLIBINC) $(ZSTD_CFLAGS) $(LIBDECNUMBER_CFLAGS) \
  	$(INTL_CFLAGS) $(INCGNU) $(INCSUPPORT) $(LIBBACKTRACE_INC) \
  	$(ENABLE_CFLAGS) $(INTERNAL_CPPFLAGS) $(SRCHIGH_CFLAGS) \
  	$(TOP_CFLAGS) $(PTHREAD_CFLAGS) $(DEBUGINFOD_CFLAGS) $(GMPINC) \

-- 
John Baldwin