This commit changes ProcessList_scan to lazily remove Processes by
index, which is known, instead of performing a brute-force search by
pid and immediately reclaiming the lost vector space via compaction.
Searching by pid is potentially quadratic in ProcessList_scan because
the process we are searching for is always at the back of the vector
(the scan starts from the back of the vector). Additionally, removal
via Vector_remove immediately reclaims space (by sliding elements
down).
With these changes process removal in ProcessList_scan is now linear.
Changes:
* ProcessList: add new ProcessList_removeIndex function to remove
by index
* Vector: add Vector_softRemove and Vector_compact functions to
support lazy removal/deletion of entries Vector_softRemove
Vector_compact
* Vector: replace Vector_count with Vector_countEquals since it only
used for consistency assertions.
- Move swap() macro to source file and implement as function
- Implement Vector_get() and Vector_size() as inline functions
to make them type safe and avoid lhs usage
- Comment comparison statistics, they are only needed for performance
testing
Doing a quick check with callgrind this gives
an average reduction from 1804 cycles/call
down to 491 cycles/call on my test system.
The average was taken over about 40k calls.
Use NDEBUG conditional instead of DEBUG.
Do not call static functions in extern inline ones.
Vector.c:67:11: error: static function 'Vector_isConsistent' is used in an inline function with external linkage [-Werror,-Wstatic-in-inline]
PR htop-dev/htop#70 got rid of the infrastructure for generating header
files, but it left behind some code duplication.
Some of cases are things that belong in the header file and don't need
to be repeated in the C file. Other cases are things that belong in the
C file and don't need to be in the header file.
In this commit I tried to fix all of these that I could find. When given
a choice I preferred keeping things out of the header file, unless they
were being used by someone else.
Reasoning:
- implementation was unsound -- broke down when I added a fairly
basic macro definition expanding to a struct initializer in a *.c
file.
- made it way too easy (e.g. via otherwise totally innocuous git
commands) to end up with timestamps such that it always ran
MakeHeader.py but never used its output, leading to overbuild noise
when running what should be a null 'make'.
- but mostly: it's just an awkward way of dealing with C code.
* Performance improvements
* Support for splitting CPU meters into two or four columns
(thanks to Wim Heirman)
* Switch from PLPA, which is now deprecated, to HWLOC.
* Bring back support for native Linux sched_setaffinity,
so we don't have to use HWLOC where we don't need to.
* Support for typing in user names and column fields in selection panels.
disable useless code in release builds such as runtime type-checking on
dynamic data structures and process fields that are not being computed,
faster(?) method for verifying the process owner (still need to ensure
correctness), don't destroy and create process objects for hidden kernel
threads over and over. Phew. I shouldn't be doing all this today, but I
could not resist.
Benny Baumann