Re: [PATCH 1/3] [gdbsupport] Add gdb::{waitpid,read,write,close}

[Simon Marchi <simark@simark.ca>]

On 2024-11-29 11:23, Tom de Vries wrote:
> On 11/29/24 17:15, Simon Marchi wrote:
>>
>>
>> On 2024-11-22 11:46, Tom de Vries wrote:
>>> On 11/5/24 13:07, Tom de Vries wrote:
>>>> On 11/1/24 18:35, Tom Tromey wrote:
>>>>>>>>>> "Tom" == Tom de Vries <tdevries@suse.de> writes:
>>>>>
>>>>> Tom> +namespace gdb {
>>>>> Tom> +
>>>>> Tom> +extern "C" {
>>>>>
>>>>> I didn't even know this was valid ... but is it really needed?
>>>>> It seems wrong.
>>>>>
>>>>
>>>> It seemed to work:
>>>> ...
>>>> $ cat test.c
>>>> namespace gdb {
>>>>
>>>>     extern "C" {
>>>>       int foo (void) { return 1; };
>>>>     }
>>>> }
>>>>
>>>> $ cat test2.c
>>>> namespace gdb {
>>>>     extern "C" {
>>>>       int foo (void);
>>>>     }
>>>> }
>>>>
>>>> int
>>>> main (void)
>>>> {
>>>>     return gdb::foo ();
>>>> }
>>>> $ g++ test.c test2.c
>>>> ...
>>>> but the linkage symbol is plain foo:
>>>> ...
>>>> $ nm ./a.out  | grep foo
>>>> 00000000004004c7 T foo
>>>> ...
>>>>
>>>> So, you're right, this is wrong.
>>>>
>>>> I seem to have ended up with gdb_waitpid instead of gdb::waitpid in v2, similar to gdb_select, but thinking about it now I slightly prefer gdb::waitpid.  I suppose it doesn't matter that much.
>>>>
>>>
>>> When looking at this again today, it bugged me enough rewrite v2 in the gdb::waitpid style.  Committed after retesting.
>>
>> When building for mingw-w64, I see this, which seems related:
>>
>>        CXX    cli/cli-cmds.o
>>      In file included from /home/simark/src/binutils-gdb/gdb/cli/cli-cmds.c:58:
>>      /home/simark/src/binutils-gdb/gdb/../gdbsupport/eintr.h: In function ‘pid_t gdb::waitpid(pid_t, int*, int)’:
>>      /home/simark/src/binutils-gdb/gdb/../gdbsupport/eintr.h:77:35: error: ‘::waitpid’ has not been declared; did you mean ‘gdb::waitpid’?
>>         77 |   return gdb::handle_eintr (-1, ::waitpid, pid, wstatus, options);
>>            |                                   ^~~~~~~
>>            |                                   gdb::waitpid
>>      /home/simark/src/binutils-gdb/gdb/../gdbsupport/eintr.h:75:1: note: ‘gdb::waitpid’ declared here
>>         75 | waitpid (pid_t pid, int *wstatus, int options)
>>            | ^~~~~~~
>>
>> Note that to get to it, you have to step past this other build error in
>> gdbserver (by doing `make all-gdb` for instance):
>>
>>    https://inbox.sourceware.org/gdb-patches/75926446-4c8a-40ee-8ed8-ab55f38e1520@simark.ca/T/#m194744b18419e37dfb4ec82c8e0ce73fb017ba17
> 
> Hi Simon,
> 
> thanks for reporting this.
> 
> I don't have a windows setup unfortunately, so I have no way of reproducing this, and at this point I have no idea why waitpid is not declared.

For just build testing, I cross-compile with a mingw-w64 cross compiler,
which is available on most distros, looks like it is on Suse:

https://en.opensuse.org/MinGW

The problem is that you might have to cross build the packages gdb
requires.  If you're lucky, your distro also packages them for
mingw-w64.  For instance, it looks like Suse packages libgmp:

https://software.opensuse.org/package/mingw64-gmp?locale=ro

These are the configure flags I currently use for this build:

   '--host=x86_64-w64-mingw32' \
   'CFLAGS=-g3 -O0' \
   'CC=ccache x86_64-w64-mingw32-gcc' \
   'CXXFLAGS=-g3 -O0 -fmax-errors=1' \
   'CXX=ccache x86_64-w64-mingw32-g++' \
   '--disable-source-highlight' \
   '--disable-gold' \
   '--disable-ld' \
   '--disable-binutils' \
   '--disable-gprof' \
   '--disable-gas' \
   '--without-python' \
   '--disable-nls'

I'll try to get it to build inside a Suse container.

Simon