"int pid" replaced with ProcessIdentifier

"int pid" replaced with ProcessIdentifier

[Andrew Cagney]

FYI,

I've pushed a low-level change so that frysk.sys's process code uses a 
ProcessIdentifier object instead of "int pid"s; there is one 
ProcessIdentifier per "pid".  The change is slowly working its way 
through frysk.proc.live; it won't be taken further.

The motivation is simple, while "int pid" looks small and efficient; it 
turns out that the code using it really needed an object (leaking 
something like <<new Integer(pid)>>) to either log or hash the pid.  
With a proper object both of these cases can be greatly simplified.

Andrew

Re: "int pid" replaced with ProcessIdentifier

[Mark Wielaard]

Hi Andrew,

On Thu, 2008-02-07 at 10:51 -0500, Andrew Cagney wrote:
> I've pushed a low-level change so that frysk.sys's process code uses a 
> ProcessIdentifier object instead of "int pid"s; there is one 
> ProcessIdentifier per "pid".  The change is slowly working its way 
> through frysk.proc.live; it won't be taken further.
> 
> The motivation is simple, while "int pid" looks small and efficient; it 
> turns out that the code using it really needed an object (leaking 
> something like <<new Integer(pid)>>) to either log or hash the pid.  
> With a proper object both of these cases can be greatly simplified.

I looked of some of the changes here and it looks somewhat inefficient,
and very verbose, because it basically replaces everything that already
has a pid int with a call to ProcessIdentifierFactory.create(pid). A lot
of these int pids seems to come from Proc.getPid(), which in turn is a
wrapper for ProcId.hashCode() where the hashCode() of a ProcId is indeed
the int representation of the pid of the Proc.

So why not just use ProcId as process identifier for a Proc since that
is already bound to a Proc object? Or the other way around and do away
with ProcId and use ProcessIdentifier instead of ProcId. It seems
confusing to now have 3 ways to identify a process int pid, ProcId and
ProcessIdentifier (4 if you also count TaskId since I saw also code like
ProcessIdentifierFactory.create(this.process.getMainTask().getTid())).

Cheers,

Mark

Re: "int pid" replaced with ProcessIdentifier

[Andrew Cagney]

Mark,

You're right, but we've also not finished working through these 
transformations either.  As code gets finished, the hacks you refer to 
will disappear.

There's one thing that is probably worth clarifying though.  Things 
break down as follows:

ProcessIdentifier -> this is a live running host system dependent value.
Like the rest of frysk.sys, beyond frysk.proc.live and test cases, code 
should not be using those interfaces.  A grep, unfortunately, shows 
otherwise.  We've lots of work to do.

Proc/Task -> this is frysk's internal representation of a process/task 
and what client code manipulate

It's important that, as we push towards having our architecture support 
remote targets, we clarify the separation between these two.

Andrew


Mark Wielaard wrote:
> Hi Andrew,
>
> On Thu, 2008-02-07 at 10:51 -0500, Andrew Cagney wrote:
>   
>> I've pushed a low-level change so that frysk.sys's process code uses a 
>> ProcessIdentifier object instead of "int pid"s; there is one 
>> ProcessIdentifier per "pid".  The change is slowly working its way 
>> through frysk.proc.live; it won't be taken further.
>>
>> The motivation is simple, while "int pid" looks small and efficient; it 
>> turns out that the code using it really needed an object (leaking 
>> something like <<new Integer(pid)>>) to either log or hash the pid.  
>> With a proper object both of these cases can be greatly simplified.
>>     
>
> I looked of some of the changes here and it looks somewhat inefficient,
> and very verbose, because it basically replaces everything that already
> has a pid int with a call to ProcessIdentifierFactory.create(pid). A lot
> of these int pids seems to come from Proc.getPid(), which in turn is a
> wrapper for ProcId.hashCode() where the hashCode() of a ProcId is indeed
> the int representation of the pid of the Proc.
>
> So why not just use ProcId as process identifier for a Proc since that
> is already bound to a Proc object? Or the other way around and do away
> with ProcId and use ProcessIdentifier instead of ProcId. It seems
> confusing to now have 3 ways to identify a process int pid, ProcId and
> ProcessIdentifier (4 if you also count TaskId since I saw also code like
> ProcessIdentifierFactory.create(this.process.getMainTask().getTid())).
>
> Cheers,
>
> Mark
>
>