Re: [PATCH 1/5] Cygwin: Make 'ulimit -c' control writing a coredump

[Corinna Vinschen <corinna-cygwin@cygwin.com>]

On Jan 15 13:27, Jon Turney wrote:
> On 15/01/2024 09:46, Corinna Vinschen wrote:
> > On Jan 13 14:20, Jon Turney wrote:
> > > On 12/01/2024 14:09, Jon Turney wrote:
> > > > +
> > > > +  PWCHAR cp = dumper_command;
> > > > +  cp = wcpcpy (cp, L"\"");
> > > > +  cp = wcpcpy (cp, dll_dir);
> > > > +  cp = wcpcpy (cp, L"\\dumper.exe");
> > > > +  cp = wcpcpy (cp, L"\" ");
> > > > +  cp = wcpcpy (cp, L"\"");
> > > > +  cp = wcpcpy (cp, global_progname);
> > > 
> > > I wonder if this should be program_invocation_short_name, so that the
> > > coredump is created in the cwd, rather than next to the executable.
> > 
> > program_invocation_short_name would be nice, but does it really matter?
> > 
> > Because...
> > 
> > > But then, there's then no way to get similar behaviour if you decide you
> > > want to use minidumps instead (by setting CYGWIN="error_start=minidumper"),
> > > as the first argument to dumper/minidump is the full path to the program (to
> > > match the 'prog procID' style of invoking gdb), but they only use it to add
> > > an .core/.dmp extension to name the file to write.
> > > 
> > > I guess that could by fixed by adding an option to the dumpers to strip
> > > paths, or more control about how the JIT command is formatted.
> > 
> > dumper/minidumper are both called with the current working directory set
> > to the ... current working directory, right?  With the full pathname as
> > input, and the CWD already set the same as the dumped application, they
> > can easily generate any target path for the corefile they like.
> > 
> > Given the actual path of the corefile can be generated by the dumpers,
> > the question is how to specify where to store the corefile. For instance
> > 
> > - no option: CWD
> > - some option -c/--coredir for anywhere else
> > 
> > Under Linux versions using systemd, corefiles are by default not stored
> > in the CWD anymore, but to /var/lib/systemd/coredump, so there is a
> > use case for arbitrary corefile paths.
> 
> Yeah, I guess an option to the dumper to control where the file is written
> is probably the best way to address this, which is something which can
> perhaps be added later...

Yeah. In that case we should write the coredump to CWD for the time
being and improve that for 3.6, ok?


Corinna