test: suppress ld leak

common: fix persistent domain termination

Ref: pmem/issues#1098

Ref: pmem/issues#1093
Ref: pmem/issues#1096
Ref: pmem/issues#1097

test: remove invalid expect_normal_exit

expect_normal_exit is incompatible with copy_file_to_node and is
causing trouble when used with valgrind.

Ref: pmem/issues#1092

test: remove old log files for windows

Ref: pmem/issues#1013

test: adjust matchfiles in vmem_valgrind_region test

test: require enoug space for obj_zones/TEST1

obj: fix recycler not locating unused chunks

In scenarios where free() is done very rarely the memory in the recycler
would not be looked at because the estimates would indicate that no memory
was freed and there's no reason to search the tree. Normally that's OK,
because eventually the heap would trigger a forced search of the entire
recycler for the allocation class, locating all the memory that might have
been missed. But, the forced recalculation erronously looked at the
estimates and required that at least some memory were freed since the
last recycler run - which might have not been the case.
This patch makes the forced recycling run without looking at the estimates.

Reported-by: Denny Zhao <denny.zhao@memverge.com>

examples: fix two strncpy issues

test: delete vmem delete

Ref: pmem/issues#1074

We decided to remove this tests,
because it checks undefined behavior;
operations on deleted pool.

common: implement correct / robust device_dax_alignment

The original implementation was looking at an NVDIMM specific attribute
that is not generally applicable to all device-dax instances. It was
also fragile in the face of the device-dax instances transitioning
between 'dax-bus' and 'dax-class' device types.

Re-implement device_dax_alignment() to be compatible with either device
model. The implementation walks the sysfs path hierarchy until it
reaches the device hosting the "dax_region".

Link: https://github.com/pmem/issues/issues/1071


Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Signed-off-by: Łukasz Plewa <lukasz.plewa@intel.com>

We fix up the missing null-terminator in the next line.

As usually, any use of strncpy() for a C string is a bug.  In this
case, it will never null-terminate the string, then if the malloced
buffer isn't zeroed, the strncat() will skip over some garbage.

Found by gcc-9's new warnings, just enabled in dpkg-buildflags.

doc: automatic stable-1.5 docs update

common: fix path to valid-branches script

common: add autodoc to Travis

vecq: fix a pointer-to-struct aliasing violation

The compiler assumed that pointers to differently-sized structs can't
alias each other, and reordered accesses to vec->back in some cases.

Thus, access members via the real-typed struct; this requires putting
this part into a macro.

tools: fix dynamic link line in Makefile.inc

This is an obvious thinko.  The result is that debug objects are
linked into the production build of the library.  Fix it.

Signed-off-by: Jeff Moyer <jmoyer@redhat.com>

obj: don't overwrite errno when palloc_heap_check_remote fails

Found using fault injection.

doc: update manpage

test: libpmempool_feature require debug