[Toybox] weird ps truncation

[enh]

On Tue, Jan 31, 2017 at 12:13 AM, Rob Landley <rob at landley.net> wrote:
>
>
> On 01/30/2017 04:20 PM, enh wrote:
>> On Sat, Jan 28, 2017 at 4:16 PM, Rob Landley <rob at landley.net> wrote:
>>> On 01/27/2017 06:00 PM, enh wrote:
>>>> Android has a lot of long usernames. the bluetooth code, for example,
>>>> runs as "bluetooth", and the camera code as "cameraserver". ideally
>>>> the USER column would grow to fit (especially when the tty is very
>>>> wide), but right now it's kind of narrow, doesn't grow, and truncates
>>>> from left rather than the right:
>>>
>>> The -w flag disables this, it doesn't just ignore screen width but
>>> disables truncation of all fields. (I.E. all strings print at their full
>>> length.)
>>
>> (yes, though it can be hard to read because the columns get shifted about.)
>>
>>> As for changing the default behavior: I already made the numeric fields
>>> leak. Now it sounds like you want the string values to leak and reclaim
>>> space, which is a less bounded problem. (The numbers aren't going to
>>> leak _that_ much, but each string could be wider than the screen.)
>>
>> no, that wasn't actually what i was thinking about. i was thinking
>> about sizing each column to contain its widest value. sort of more
>> like how columnated ls works.
>
> Doable, but expensive. top is already taking 5% cpu on my netbook as
> opposed to ubuntu top's 1.6%. I have a todo item to profile what it's
> spending its time on and trying to streamline that a bit, but it's a
> ways down the todo list. :)
>
>>> So what behavior do you _want_ here: swap in numbers, leak/reclaim, or
>>> preread all preread all the entries and calculate column values like ls
>>> does? Or just truncate from left instead of from right?
>>
>> what you've already checked in, fixing the truncation to be
>> consistent, gets us back to where we were.
>>
>> beyond that i'm not sure...
>
> Me neither.
>
>> the "+" to show we did truncate seems helpful.
>
> Fiddly to implement though. The STIME field is truncated during normal
> display, having that _not_ turn the last digit into + every time
> requires a special case. And that's sounding like I need to audit all
> the fields to see which ones need this...
>
> (I found a bug while I was working on that, it didn't copy the null
> terminator when stripping /dev off the tty name so the display could
> have trailing garbage.)
>
> Also I've now been staring at the code long enough I don't remember why
> I only did the utf8 parsing when we a tty? We jump through hoops to find
> the controlling tty, including checking /dev/tty, because ps without -a
> only shows processes attached to the current terminal. But that doesn't
> involve our OUTPUT being to a tty, ala "ps | cat"... So why was I making
> that distinction again? Hmmm...
>
> It's 2 am, I should go to bed.
>
>> switching from USER to UID seems like a clever special case, .
>
> Yes, but what's expected behavior here?

no idea. i was just giving my opinion of the various choices. i
personally don't really like the switch from USER to UID. that doesn't
seem like it would ever be what i wanted. certainly makes scripting
hard.

>> doing the "+" is at least an obvious signal that something's going on
>> here. i'm surprised no one's noticed before now that "bluetooth" gets
>> truncated. actually, here's the list of usernames that get truncated
>> on a Pixel:
> ...
>> 9 u0_a10000
>>
>> (that last one is the start of a block of generated usernames of
>> length 9, corresponding to installed apps.)
>>
>> maybe Android should just default to 12?
>
> In ps, in top, or in both? And would you still want the +?

good question. but for now i haven't actually had any complaints about
usernames being truncated. only SELinux label truncation has caused
confusion (afaik), and that's already at 30. i think SELinux labels
are the strongest motivating example for some kind of truncation
marker. or column resizing like ls.

some have suggested using fully-qualified app names (such as
"com.android.providers.calendar" or "com.facebook.katana") instead of
the generated u0_a2 style of naming, but -- even though those are
theoretically more helpful -- that just seems like it's going to make
it less likely that folks will see useful names. everything showing up
as "com.and+" isn't much better than everything showing up as
"com.andr" :-)

> Rob



-- 
Elliott Hughes - http://who/enh - http://jessies.org/~enh/
Android native code/tools questions? Mail me/drop by/add me as a reviewer.