Review Request 121831: libksysguard: process.h: encapsulate private fields

[Gregor Mi]

> On Jan. 21, 2015, 10:10 a.m., Dominik Haumann wrote:
> > processcore/processes_linux_p.cpp, line 171
> > <https://git.reviewboard.kde.org/r/121831/diff/3/?file=343197#file343197line171>
> >
> >     Don't you change behavior here?
> >     
> >     Before, we just wrote into the varialbes.
> >     
> >     Now, we use the setters, which also sets 'changes |= Process::Gids;'
> >     
> >     Is that maybe an issue? I myself don't know the code well enough to see this here.
> 
> Gregor Mi wrote:
>     True. Thanks for noticing. The changes will be read in ProcessModelPrivate::processChanged when a change is signaled by KSysGuard::Process::processChanged. Then GUI model updates are triggered. So probably the worst that can happen is one additional GUI update when processChanged is emitted.

I did some analysis. There is no visible change in behaviour ksysguard.

ProcessesLocal::Private::readProcStatus where the change in discussion was made (setter method with change detection instead of direct assignment).

is called by ProcessesLocal::updateProcessInfo in the same file:

```
bool ProcessesLocal::updateProcessInfo( long pid, Process *process)
{
    bool success = true;
    QString dir = "/proc/" + QString::number(pid) + '/';
    if(!d->readProcStat(dir, process)) success = false;
    if(!d->readProcStatus(dir, process)) success = false;  <------------
   ...
```

Then there is

```
bool Processes::updateProcess( Process *ps, long ppid)
    ...
    ...

    bool success = updateProcessInfo(ps); <--------
    emit processChanged(ps, false);  <-------

    return success;
}
```

```
bool Processes::updateOrAddProcess( long pid)
...
...
    Process *ps = d->mProcesses.value(pid);
    if(!ps)
        return addProcess(pid, ppid);
    else
        return updateProcess(ps, ppid); <---------
}
```

which is called e.g. every second.


The emitted processChanged signal is connected here:

```
connect( mProcesses, SIGNAL(processChanged(KSysGuard::Process*,bool)), this, SLOT(processChanged(KSysGuard::Process*,bool)));
```

void ProcessModelPrivate::processChanged(KSysGuard::Process *process, bool onlyTotalCpu):

Each change on a field that was recored in ProcessesLocal::Private::readProcStatus will emit dataChanged signal, e.g.:

```
        if(process->changes() & KSysGuard::Process::Uids) {
            totalUpdated++;
            QModelIndex index = q->createIndex(row, ProcessModel::HeadingUser, process);
            emit q->dataChanged(index, index);
        }
```

The dataChanged signal is from QAbstractItemModel. So as far as I can see the worst thing could happen that the GUI is updated at more places than needed.

When I start ksysguard the processChanged method is called about 300 times (1 time for each process in the list) in the update interval (every second).

I tried to update everything at once and circument the dataChanged of individual items with
```
        QModelIndex index1 = q->createIndex(row, 0, process);
        QModelIndex index2 = q->createIndex(row, q->columnCount() - 1, process);
        emit q->dataChanged(index1, index2);
        return;
```
which had not visible effect.

So what now? Leave the setters and have maybe a performance penalty? Remove the setters again and use references getters? Change the Process API otherwise?


- Gregor


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://git.reviewboard.kde.org/r/121831/#review74463
-----------------------------------------------------------


On Jan. 25, 2015, 10:27 p.m., Gregor Mi wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://git.reviewboard.kde.org/r/121831/
> -----------------------------------------------------------
> 
> (Updated Jan. 25, 2015, 10:27 p.m.)
> 
> 
> Review request for KDE Base Apps, Dominik Haumann, Eike Hein, and John Tapsell.
> 
> 
> Repository: libksysguard
> 
> 
> Description
> -------
> 
> In process.h there are several public fields. This RR introduces a d-ptr.
> 
> 
> (In a separate commit I would add the .reviewboardrc file)
> 
> What is the current policy on using small C++ macros as done in this RR? Use it (code is more compact and readable) or don't use it (better for debugging)?
> 
> 
> Diffs
> -----
> 
>   processui/scripting.h 2445c0ab0d81af3283c0f6e9c5f349a3d70b0de9 
>   processui/scripting.cpp 76291b0ae0a26e486aa81a4ca3976ff4a47cb3c0 
>   processcore/processes_solaris_p.cpp f054df4b1e762e9cbec1ff8dea78f467b878bee0 
>   processui/ProcessFilter.cpp ec520593fb67c777d56817f2493d40dc5ade0347 
>   processui/ProcessModel.cpp 53bc041110c9cdb686fef783895104969b661889 
>   processui/ksysguardprocesslist.cpp 4dc142e864d8353ceafc3a6735ffa81e48291420 
>   processcore/processes.cpp 580df8db152040f1ad075430fdce08fe7ad4ae2d 
>   processcore/processes_atop_p.cpp 24c76e3e35f62bd8e9e705ad32cc11cbd3662601 
>   processcore/processes_base_p.h 71b8a9cc6ee14bf7934a0a9d3199b257b5ce1be7 
>   processcore/processes_linux_p.cpp 898d4fa491873fe95a8b32a5c1b85642b2e46ad5 
>   processcore/processes.h d09c3265333fe7e2702deaa910c5fbe4bc3ac9e6 
>   tests/processtest.cpp f9b36e9a3a3c2048b51f1d935f8c40de2ad8a9b8 
>   .reviewboardrc PRE-CREATION 
>   CMakeLists.txt cefc86f12be684e195bd148641483e9e1734e636 
>   processcore/process.h b6695c0ed301dc5f0fad8ba847da811f19ebfd9a 
>   processcore/process.cpp a38b8be71da1a51cb87f636664ebac817b1d20ab 
> 
> Diff: https://git.reviewboard.kde.org/r/121831/diff/
> 
> 
> Testing
> -------
> 
> Compiles and runs. Data is still shown; no visible error. Unit tests succeed.
> 
> 
> Thanks,
> 
> Gregor Mi
> 
>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.kde.org/pipermail/kde-core-devel/attachments/20150130/faece34a/attachment.htm>