Hi Takashi,

On Apr 16 11:16, Corinna Vinschen wrote:
> On Apr 16 09:41, Takashi Yano wrote:
> > On Mon, 15 Apr 2019 10:38:32 +0200 Corinna Vinschen wrote:
> > > This is supposed to be combined with the preceeding code to a single
> > > expression in the end, right?
> > 
> > Do you mean this code and the preceding code should be executed
> > exclusively? If so, the answer is no. If the stderr is pty slave,
> > the message should be written to both output_handle and
> > output_handle_cyg. The reason will be described later.
> > 
> > > I don't quite understand why writing
> > > to STD_ERROR_HANDLE doesn't work.  Is STD_ERROR_HANDLE not connected
> > > to the ConPty?
> > 
> > You are right. STD_ERROR_HANDLE (output_handle) is connected to
> > the ConPTY. The data written to output_handle is fowarded to
> > to_master_cyg in the master, then displayed. to_master_cyg is
> > duplicated to output_handle_cyg in the slave. Therefore, both
> > writing to output_handle and writing to output_handle_cyg seem
> > to work, at first glance.
> > 
> > However, in the fact, the code below stops forwarding if the
> > process is a cygwin process. 
> > +	  /* Avoid duplicating slave output which is already sent to
> > +	     to_master_cyg */
> > +	  if (!get_ttyp ()->switch_to_pcon)
> > +	    continue;
> > 
> > The role of the code above is as follows. ConPTY sometimes redraws
> > screen based on its own screen buffer. The screen buffer has only
> > the data written to output_handle. Basically, cygwin apps use
> > output_handle_cyg and native apps use output_handle. Terefore,
> > outputs from cygwin apps are disappeared at that time. To avoid
> > this, the data write to output_handle_cyg is pushed into the ConPTY
> > screen buffer by WriteFile() to output_handle. To prevent the data
> > from being displayed twice, forwarding is stopped by the code above
> > if the process is a cygwin app.
> > 
> > In strace.cc, writing to output_handle is necessary for pushing
> > the data into ConPTY screen buffer, and writing to output_handle_cyg
> > is necessary for displaying it.
> 
> Huh, crazy.  Thanks for explaining the context.
> 
> > 
> > > > +      fhandler_base *fh = ::cygheap->fdtab[2];
> > > > +      if (fh && fh->get_major () == DEV_PTYS_MAJOR)
> > > 
> > > For a quick test if the process is connected to a pty, you can
> > > use `if (istty_slave_dev (myself->ctty))'
> > 
> > Thanks for advice. My code is checking if the stderr is pty or
> > not, and write the message to output_handle_cyg if the stderr
> > is pty. If the stderr is redirected to another device, writing
> > is not done. Does istty_slave_dev() checking work same for
> > stderr if it is redirected? How can I get output_handle_cyg
> > after checking using istty_slave_dev()?
> 
> Well... I didn't really think about it.  YOu would have to
> fetch ::cygheap->fdtab[2] anyway so checking myself->ctty
> doesn't really help.

Any news on this?  Do you consider the latest state from April
stable enough for master?


Thanks,
Corinna

-- 
Corinna Vinschen
Cygwin Maintainer