Re: [AMBER-Developers] Sander parallel build broken by PBSA updates

From: Jason Swails <>
Date: Thu, 18 Feb 2016 22:12:20 -0500

On Thu, Feb 18, 2016 at 2:23 PM, Ray Luo <> wrote:

> Thanks, I'm downloading your fix ...
> Also, the python-related installation seems to mess up the dependence
> issue whether I use single or multi-thread make ... I don't know why
> my changes in pbsa and sander have caused this to break.

‚ÄčAll of the Python build commands are single rules incapable of threading
-- they cannot bring down a parallel build due to botched dependencies
(since there are none). Quite often the last thing you see when parallel
make dies is *not* the thread that died. Once a thread dies, it waits
until all other threads are done doing whatever they're doing, then they
all exit with an error code. So it may look like the last thing "make" did
was try to build a Python package before dying, but it was almost certainly
a different thread that died doing something else.

You can see which make thread died by looking through the output of make --
the thread number responsible for informational messages are typically
labeled with something like "make [#]", where # is the thread number
responsible for that line.

But it sounds like everything is fixed now -- thanks Gerald! FTR, the fix
that Gerald did here was to label several of the new Fortran modules as
having separate LIBPBSA objects. Any object file that has an #ifdef
LIBPBSA or something inside of it needs to be added to the LIBPBSA hash map
in the makedepend script (and that makedepend script needs to be rerun to
regenerate an updated depend file). That will keep the dependencies for
PBSA in order.


Jason M. Swails
Rutgers University
Postdoctoral Researcher
AMBER-Developers mailing list
Received on Thu Feb 18 2016 - 19:30:05 PST
Custom Search