getpass regressions on Solaris/Illumos - 3.6 and master.

Jeremy Allison jra at samba.org
Mon Jan 30 18:18:54 MST 2012


On Mon, Jan 30, 2012 at 05:09:47PM -0800, Jeremy Allison wrote:
> On Mon, Jan 30, 2012 at 10:27:39AM -0500, Ira Cooper wrote:
> > In looking at this issue in more detail:
> > 
> > I've heard the argument: Putting all the header files in replace.h will
> > increase build time.  So I did the experiment.  I put all the system/*
> > include files in replace.h except for system/kerberos.h, because
> > system/kerberos.h broke my build.  So far the difference doesn't look very
> > big at all, at least on my machine!  Now, I'll admit my development machine
> > is a really bad place to benchmark builds, so I'd like others to chime in
> > with their results, if there is interest in this approach.  This patch is
> > include_it_all.patch.
> > 
> > (make -j64 of our autoconfed build; on my dev box.)
> > 
> > Without the extra headers:
> > 
> > real    0m47.438s
> > user    8m38.416s
> > sys     1m17.440s
> > 
> > With them:
> > 
> > real    0m49.108s
> > user    8m55.999s
> > sys     1m23.193s
> > 
> > There is a second viable approach that metze pointed at, but this is NOT
> > what he suggested directly.  Pull the getpass section of system/passwd.h
> > into replace.h  Including the definition of rep_getpass.  If you do it
> > without that definition, you'll break platforms without a good getpass and
> > getpassphrase.  :(  This patch is add_getpass.patch.
> > 
> > Diffs for both approaches are included.
> > 
> > The add_getpass.patch is what I consider the "absolute minimum" to close
> > this issue.  But I thought the first approach was interesting enough, and
> > the result interesting enough that I've included both.  I'd like to see
> > other developer's times on the "include_it_all.patch".  Please be careful
> > of caching effects.  Build both ways a few times.
> 
> Ok, I like the add_getpass.patch as it seems smaller and cleaner to
> me.
> 
> Metze, or anyone else with an interest can you please comment on
> this asap ?
> 
> Otherwise I think I'll just add add_getpass.patch into master and
> drive it through the 3.6.next process.

Or optionally, what about (for master at least) simply
requiring getpassphrase in Samba ?

Change the replacement rep_getpass to be rep_getpassphrase
for systems that don't support rep_getpass.

Comments ?

Jeremy.


More information about the samba-technical mailing list