Tweaks:
* --headers command line option, which can be used to
enable/disable the automatic headers, and the
automatic empty lines between samples.
* --show-timestamps, which changes the #ts line with
a timestamp in the HH:MM:SS format.
* Always use ReadKeyMini, even when we have Term::ReadKey.
In the latter case, use the Term::ReadKey functions,
but if they fail, try again with our own versions.
Fix:
Looks like I forgot to actually enable the "save the current, first
and last" samples. This is now on. It should mean faster and better
output when changing --group-by modes.
This fixes two bugs -- One a rounding error, one a desyncronization
between the main program and the data gathering daemon.
Additionally, it changes how the daemon prints the very first
sample. Before it would unconditionally wait until the time was
%$sample-time, which meant that in a worst case scenario,
a user could be left waiting for a while before seeing a line
of output.
This commit changes that first wait to only sleep if less
than 20% away from %$sample-time.
Originally, DiskstatsMenu differentiated between an empty pattern
entered by the user (which in the shell version meant "match all")
and any other valid pattern by replacing it with /(?=)/.
However, Diskstats now carries the original string, inside the
OptionParser object, so the above was redundant.
/proc/diskstats actually has two formats, one for devices, one for
partitions. The latter wasn't handled by the original pt-diskstats,
but one of the previous commits introducing some minimal handling
for that. This commit removes that, since the partition format
doesn't have enough information to be useful.
Still TODO: Attributes still have a single method that doubles
as a getter and setter. The constructor for Diskstats is still
weird -- A tad more stricter than it should be, if anything.
->print_rest is still rest, even though that's hardly
memorable, mostly because of a lack of ideas on what to
rename it. The main loop in the Menu is still a while (1).
As a nice perk, it's nearly twice as fast now! It also adds a
_very_ experimental --memory-for-speed argument, which
turns on memoization for the current biggest bottleneck.
Missing a doc overhaul and tests, but this deals with most of the
feature upgrades in the blueprint.
The shell version of pt-diskstats is now in bin has pt-diskstats-shell,
in case that's needed.
