[Bug binutils/28684] New: [RISCV] 32-bit --enable-targets=all build breakage issue

[nelsonc1225 at sourceware dot org]

https://sourceware.org/bugzilla/show_bug.cgi?id=28684

            Bug ID: 28684
           Summary: [RISCV] 32-bit --enable-targets=all build breakage
                    issue
           Product: binutils
           Version: unspecified
            Status: NEW
          Severity: normal
          Priority: P2
         Component: binutils
          Assignee: unassigned at sourceware dot org
          Reporter: nelsonc1225 at sourceware dot org
  Target Milestone: ---

The 32-bit build failed again when --enable-targets=all, the original
discussion was here,
https://sourceware.org/pipermail/binutils/2021-November/118485.html

For now, the dis-assembler (objdump) need to call the riscv architecture parser
from bfd/elfxx-riscv.c
(https://sourceware.org/pipermail/binutils/2021-November/118444.html), but this
will break the 32-bit host --enable-targets=all build when building the
libopcodes.a,

[Copied from Alan's comments]
usr/local/bin/ld: ../opcodes/.libs/libopcodes.a(riscv-dis.o): in function
`riscv_disassemble_insn':
/home/alan/src/binutils-gdb/opcodes/riscv-dis.c:541: undefined reference to
`riscv_multi_subset_supports'
/usr/local/bin/ld: ../opcodes/.libs/libopcodes.a(riscv-dis.o): in function
`riscv_get_disassembler':
/home/alan/src/binutils-gdb/opcodes/riscv-dis.c:892: undefined reference to
`riscv_release_subset_list'
/usr/local/bin/ld: /home/alan/src/binutils-gdb/opcodes/riscv-dis.c:893:
undefined reference to `riscv_parse_subset'
collect2: error: ld returned 1 exit status
make[4]: *** [Makefile:940: objdump] Error 1

As I known, there are two solutions could fix the problem,

The first one was suggested by Alan Modra, and the idea is great - don't
compile some opcodes files when bfd is 32-bit only,
https://sourceware.org/pipermail/binutils/2021-November/118500.html
https://sourceware.org/pipermail/binutils/2021-December/118751.html

But seems like this caused the simulator and gdb build failed, when building
bfd in 32-bit mode.  Therefore, Andrew Bugress have a proposed solution to make
gdb can build as usual,
https://sourceware.org/pipermail/gdb-patches/2021-December/184365.html

At the meantime, Jim Wilson also remind us that my previous RFC patch may
works, and then we probably won't need the extra gdb fixes, and the idea of
solution was also came from him,
https://sourceware.org/pipermail/binutils/2021-November/118498.html

I'm OK that we can commit Andrew's gdb patch first, and then probably spend
some times to make sure if my RFC patch do works.  If so, then we can revert
the gdb patch at that time.  I collect all the information here, in case I
forgot the details when coming back to see this issue again, or someone is
interested and try to resolve it.

Thanks
Nelson

-- 
You are receiving this mail because:
You are on the CC list for the bug.