Don't leak the last argument expanded by expandNextArg()

[vathpela]

While using POPT_ARG_ARGV, I noticed this in valgrind's leak checker:

==1738== HEAP SUMMARY:
==1738== in use at exit: 8 bytes in 1 blocks
==1738== total heap usage: 94 allocs, 93 frees, 42,319 bytes allocated
==1738==
==1738== 8 bytes in 1 blocks are definitely lost in loss record 1 of 1
==1738== at 0x4C2EB6B: malloc (vg_replace_malloc.c:299)
==1738== by 0x4E3DF47: expandNextArg (popt.c:699)
==1738== by 0x4E3F681: poptGetNextOpt (popt.c:1501)
==1738== by 0x401F72: main (bingrep.c:433)
==1738==
==1738== LEAK SUMMARY:
==1738== definitely lost: 8 bytes in 1 blocks
==1738== indirectly lost: 0 bytes in 0 blocks
==1738== possibly lost: 0 bytes in 0 blocks
==1738== still reachable: 0 bytes in 0 blocks
==1738== suppressed: 0 bytes in 0 blocks

My command line argument is a 7-byte string, and on first glance, it
appears this is because both expandNextArg() and poptSaveString()
duplicate the string. The copy from poptSaveString() is the consuming
program's responsibility to free, but the intermediate pointer is popt's
responsibility.

Upon further examination, it appears popt normally does free this
string, but it only does it on the next entry to poptGetNextOpt(), and
on cleanOSE() in the case if we're not already at the bottom of
con->OptionStack.

This patch modifies poptResetContext() to ensure we'll always attempt to
free con->os->nextArg regardless of our position in the OptionStack, and
removes the duplicate free of con->os->argb in poptFreeContext(), as
it's called unconditionally by the poptResetContext() call on the
previous line.

This ensures that if poptGetNextOpt() isn't re-intered, poptFreeContext()
will free the memory that was allocated. Now valgrind tells me:

==31734== HEAP SUMMARY:
==31734== in use at exit: 0 bytes in 0 blocks
==31734== total heap usage: 94 allocs, 94 frees, 42,319 bytes allocated
==31734==
==31734== All heap blocks were freed -- no leaks are possible

Signed-off-by: Peter Jones pjones@redhat.com