Re: Memory leak

[Paul D. Smith]

%% Mark Bucciarelli <address@hidden> writes:

  mb> I've been seeing garbage output to the console when running a set
  mb> of tests with make check.

  mb> I ran valgrind on make check and it indicated two memory leaks 
  mb> in /usr/bin/make.

There are a few known, but innocuous memory leaks in make.  I've
investigated them in the past but decided that the work needed to fix
them made the code really gross, and they were just one-time allocations
for dealing with command line arguments, etc., so I didn't bother.

  mb> You can find all the details, including the valgrind output, here:

  mb> http://bugs.kde.org/show_bug.cgi?id=95237 

  mb> When I put the tests in a bash script and run that script
  mb> repeatedly, I got clean output every time.  make check was
  mb> producing garbage every time.

I don't see any way that a memory LEAK could cause that behavior: it's
just memory that no one wants anymore but that hasn't been freed.  The
only problem it could cause would be if make lost memory constantly and
eventually ran out or used up too much on the system.  Losing 500 bytes
or so is not even noticeable.


I really don't see, offhand, how it could be make's fault that the
regression tests don't run.  Make just invokes the commands you ask it
to.

One possible difference is that make always runs its commands in
/bin/sh, not /bin/bash.  Also, make might be exporting different values
for the environment than you get when you run it from the command line.

Try this: change the command that invokes the script to run "env" so you
know what your environment is, then put that output into a script
starting with #!/bin/sh and then add the command to invoke your test,
and run that script from your shell command line.  If it fails, then you
know it's not make.  If it works, then it is looking more like something
to do with make (although I'm still confused); in that case please post
again with at least the version of make you're using, plus your
platform, and, if you can manage it, a reproducible test case.


In any event, the valgrind issue is a red herring I believe.

-- 
-------------------------------------------------------------------------------
 Paul D. Smith <address@hidden>          Find some GNU make tips at:
 http://www.gnu.org                      http://make.paulandlesley.org
 "Please remain calm...I may be mad, but I am a professional." --Mad Scientist