[procps] Re: A rainy days effort
- From: Jim Warner <james.warner@xxxxxxxxxxx>
- To: procps@xxxxxxxxxxxxx
- Date: Sun, 4 Mar 2012 04:22:10 -0600
On Mar 3, 2012, at 3:16 PM, Craig Small wrote:
>> I've got nothing right now but what is your release target?
> For Debian there is no hurry as there nothing too critical. I got
> one important bug around dejagnu because I change the w option.
> Hmm, there is also #627257 about top leaking memory but that's a 3.2.8
> bug though valgrind is still having a complain about it:
Hi Craig,
That bug resulted from a patch to readproc.c supporting supgid addition to ps
in response to #506303. Whenever /proc/#/status was accessed, memory was
acquired and never freed.
I think the current "memory leaks" have something to do with library hashing
but they never grow and are so small I've never bothered to track them down.
Below is a summary of all top bugs and their status.
> Maybe, but that's about 3 hours drive away. Minneapolis vs Duluth with
> a large mountain range in between.
Hey, Sydney or Canberra -- what's 3 hours drive to an Aussie. And despite the
common misconception, I'm pretty sure there's no Duluth in Minnesota, where we
only have big hills.
Regards,
Jim
-----------------------------------------------------------
---------------------------------------------- PROPOSED FIX
#84992, should have (better) column for page faults
Will add page fault deltas in libprocps1 branch.
----------------------------------------------- PENDING FIX
#651213, top: incompatible rcfile
#651863, top: incompatible upgrade, missing instructions
Will be fixed in pending 3.3.3 release.
-------------------------------------------- POSSIBLY FIXED
#225542, top (procps: top(1)): ``Unknown command...'' message blocks
further commands
With procps-ng, the time a message is shown was reduced to a more tolerable
1.25 seconds.
#567509, procps: top 'i'idle command not working for threaded programs
With 'i', top now shows all tasks/threads that have used some cpu, not just
those running.
----------------------------------------------------- FIXED
#322984, top: repeating "%" wastes precious space
Fixed (sort of) with procps-ng. Now top just shows a single '%'.
#351934, procps: Make 'top' aware of resources hotplug
Fixed with procps-ng.
#410292, top: interface error
This was really a problem with termios and emulation. Now that top defaults to
using his own line input logic (TERMIOS_ONLY not defined), it can't happen.
#441166, top (procps): Weired CPU utilization when toggling thread
display ("H")
fixed with commit b73a423cc2d3426f4820d519acb95adaac7d5444.
#454786, top: Use M, G or even T instead of K for better readability
Fixed (sort of) with procps-ng. Now top just uses a single instance of the
memory suffix.
#459287, top: doesn't handle hot-added, hot-removed CPUs
Addressed in procps-ng, refined in commit
f348575edc915db9df89acfb3f292920726ffe9a.
#623200, top -U parses user name error
This was caused by the patch submitted by Justin Pryzby, in response to
#358724. Fixed with procps-ng.
#627257, top: memory leaks
Caused by a patch to readproc.c supporting the supgid addition to ps in
response to #506303. Whenever /proc/#/status was accessed, memory was acquired
and never freed. Fixed with procps-ng.
#628462, top (linux-2.6): reports impossible swap statistics for a
process
Fixed with procps-ng.
#650864, top forest view segfault
Fixed with procps-ng-3.3.2.
------------------------------------------ SHOULD BE CLOSED
#320501, top: default behavior on a pipe
Even the user points out how to satisfy the pipe need, with the -b and -n
switches.
#340751, procps: top session: which hostname?
Craig's reply already pointed out the symlink solution to the hostname need.
or the user could just make copies of the top program as hostname/hostname-top.
#470911, top: inconsistent column headers in multi-window mode
This kind of aberration would be impossible to duplicate or track down. It's
why top provides a refresh capability (enter or space). Changing the delay
interval was unnecessary.
#489305, incorrect 'top' behavior (memory for threads)
This is actually a bug in the kernel since the data for a thread is just copied
from /proc/#/statm to /proc/#/task/#/statm. And contrary to the author's
assertion, ps displays the same distorted values.
#586497, top: graceful degradation on small screen sizes
The symlink approach, with separate configuration files, provides the means to
achieve the user's objective.
#596131, top: on kfreebsd: /proc/uptime: Input/output error
Craig's reply should have ended the matter/
------------------------------------------------- UNCERTAIN
#270987, procps: top and ps should use hour:min:sec for CPU time
#641905, top & ps (procps): CPU usage reporting for very long running
jobs broken
Other related posts:
