Re: [PATCH 4/5] Fix exit/signal code on Cygwin

[Pedro Alves <pedro@palves.net>]

On 2026-05-22 08:15, Eli Zaretskii wrote:
>> From: Pedro Alves <pedro@palves.net>

...

>> This commit fixes it.  To avoid duplicating code, it adds a new
>> native_exit_code_to_target_status function in nat/windows-nat.c used
>> by both GDB and GDBserver, with the MinGW-specific logic added by
>> commit 559e7e5056 ("Improve process exit status macros on MinGW")
>> moved there too.
> 
> AFAIU, this basically adds a Cygwin-specific branch to the code that
> determines the terminating signal and status of a program, leaving the
> code for the native Windows and MinGW programs intact.  

Yes.  For v2, I split the refactor to its own patch, which I think makes
that more obvious.

> I suggest to
> say this in the commit log message, because as written, it sounds like
> it does something for Cygwin that is not done for MinGW.  Which is not
> true.

Done this too, thanks.

> 
>> +#ifdef __CYGWIN__
>> +  /* /usr/include/cygwin/wait.h explains that a wait status is 16
>> +     bits, and looks like:
>> +
>> +       "<1 byte info> <1 byte code>
>> +       <code> == 0, child has exited, info is the exit value
>> +       <code> == 1..7e, child has exited, code is the signal number.
>> +       <code> == 7f, child has stopped, info was the signal number.
>> +       <code> == 80, there was a core dump."
>> +
>> +     However, when passing the wait status to native ExitProcess as a
>> +     native exit code, cygwin1.dll swaps the <info>/<code> bytes.
>> +     Swap them back into a wait status here.  */
>> +  int wstatus = ((exit_code & 0xff) << 8) | ((exit_code >> 8) & 0xff);
> 
> Should the commentary say something about _why_ we swap the bytes
> here?  I know very little about Cygwin, but my naïve view is that when
> a Cygwin program is run from another Cygwin program, no such swapping
> should be needed, is that right?  One should just use the macros from
> the sys/wait.h header, right?  If so, why do we need to swap the bytes
> here, and how is "passing the wait status to native ExitProcess as a
> native exit code" relevant to what this part of GDB needs to do?

Thanks for raising the question, it made me look deeper, dig into the Cygwin code
in mode detail, and realize that in the attach case, we shouldn't do the swap.
v2 has more and better comments explaining all this, as well as handling the
attach case, and testing it too.  The new comments will implicitly answer your
questions.  I'll be sending v2 shortly to the list.

Pedro Alves

> 
> Other than these questions, the MinGW part of the code looks okay to
> me, thanks.
> 
> Reviewed-By: Eli Zaretskii <eliz@gnu.org>
>