Re: [PATCH] nat/fork-inferior: include linux-ptrace.h

[Simon Marchi <simon.marchi@polymtl.ca>]

On 2018-06-25 04:05, Thomas Petazzoni wrote:
> To decide whether fork() or vfork() should be used, fork-inferior.c
> uses the following test:
> 
>   #if !(defined(__UCLIBC__) && defined(HAS_NOMMU))
> 
> However, HAS_NOMMU is never defined, because it gets defined in
> linux-ptrace.h, which is not included by fork-inferior.c. Due to this,
> gdbserver fails to build on noMMU architectures. This commit fixes
> that by simply including linux-ptrace.h.
> 
> This bug was introduced by commit
> 2090129c36c7e582943b7d300968d19b46160d84 ("Share fork_inferior et al
> with gdbserver"). Indeed, the same fork()/vfork() selection was done,
> but in another file where linux-ptrace.h was included.
> 
> Fixes the following build issue:
> 
> ../nat/fork-inferior.c: In function 'pid_t fork_inferior(const char*,
> const string&, char**, void (*)(), void (*)(int), void (*)(), const
> char*, void (*)(const char*, char* const*, char* const*))':
> ../nat/fork-inferior.c:376:11: error: 'fork' was not declared in this 
> scope
>      pid = fork ();
>            ^~~~
> ../nat/fork-inferior.c:376:11: note: suggested alternative: 'vfork'
>      pid = fork ();
>            ^~~~
>            vfork
> 
> Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
> ---
>  gdb/nat/fork-inferior.c | 1 +
>  1 file changed, 1 insertion(+)
> 
> diff --git a/gdb/nat/fork-inferior.c b/gdb/nat/fork-inferior.c
> index 8b59387fa5..05167628a6 100644
> --- a/gdb/nat/fork-inferior.c
> +++ b/gdb/nat/fork-inferior.c
> @@ -26,6 +26,7 @@
>  #include "common-gdbthread.h"
>  #include "signals-state-save-restore.h"
>  #include "gdb_tilde_expand.h"
> +#include "linux-ptrace.h"
>  #include <vector>
> 
>  extern char **environ;

Hi Thomas,

fork-inferior.c is also included in native builds for BSDs, AIX, Solaris 
and Darwin (see gdb/configure.nat).  I am a bit concerned that 
linux-ptrace.h could use some Linux-specific things, and thus would 
break the other builds.  However, I built-tested on FreeBSD and it seems 
fine.  Worst case, we can probably wrap this include in "#ifdef 
__linux__" if that becomes a problem.

Do you have push access, or do you prefer if I push the patch for you?

I suppose that error was caught by a Buildroot autobuilder?  Would it be 
possible to have the config, so I can add a similar configuration to my 
collection of cross-compiled GDB builds I use for build-testing?

Thanks,

Simon