Re: More: [1.7] packaging problem? Both /usr/bin/ and /usr/lib/ are non-empty

[Charles Wilson <cygwin at cwilson dot fastmail dot fm>]

In followup to this thread:
http://cygwin.com/ml/cygwin-developers/2009-05/msg00007.html
which began as a discussion of default mount behavior of the cygwin DLL,
but veered off into a discussion of setup.exe behavior. It's now almost
entirely about setup.exe, so I'm replying here.

Eric Blake wrote:
> But since something is still
> interactive, it seems like it must be the outer invocation of
> c:\cygwin-1.7\bin\bash.exe, where we explicitly passed --norc.  I
> still see no reason why the outer shell should be sourcing these
> files, particularly since we did not ask it to be interactive, but
> since the inner invocation claims to be non-interactive, it seems like
> it is the outer shell doing it.  Maybe it would be worth temporarily
> patching setup.exe to pass -cxv instead of -c to the outer invocation.
>
> Or also try patching it to do c:\cygwin-1.7\bin\bash.exe ... -c "echo
> outer: $- $0; fake-bash.sh", so we can see whether .bashrc is being
> sourced before or after the attempt to start the inner shell.  And
> we're certain that SHELLOPTS, BASH_ENV, and ENV are not set?

Actually, we are NOT invoking with --norc. Here's the most recent patch
to setup.exe:

@@ -223,7 +223,7 @@ Script::run() const
   OutputLog file_out = std::string (mktemp (tmp_pat));
   if (sh.size() && stricmp (extension(), ".sh") == 0)
     {
-      log(LOG_PLAIN) << "running: " << sh << " -c " << scriptName <<
endLog;
+      log(LOG_PLAIN) << "running: " << sh << " --norc --noprofile -c "
<< scriptName << endLog;
       retval = ::run (sh.c_str(), "-c", scriptName.c_str(), file_out);
     }

Notice anything?  I'll try again Monday with the actually ::run command
modified...

--
Chuck