On Thu, 18 Apr 2013, D. Michael McIntyre wrote:
> On 04/18/2013 11:41 AM, Holger Marzen wrote:
>
> > trying to understand the data structures, failing. But I noticed an
> > excessive use of ++j in for-loops instead the C-style j++.
>
> Probably because there are tons and tons of situations in STL and Qt
> where standard idiom is to iterate through container classes with
> iterators, and the standard language is
>
> for (some_class_iterator i = some_class.begin();
> i < some_class.end(); ++i) { ...
Yes, that's what I think. But then we have tons of logical inaccuracies
because we don't see "i < some_class.end()" but "i != some_class.end()".
That's less safe.
> > So loops
> > don't start with a zero offset. Bug 1358 is an off-by-one bug, so I
> > tried
> >
> > pluginLatency +=
> > m_instrumentMixer->getPluginLatency((unsigned int) *
> > connections.begin() - 1);
> >
> > and it looks like the bug has been gone.
>
> Nah, trying to start iterating from before begin() can't be right, and
> is likely to end badly.
I agree that it leaves a bad smell. But I guess the QT-style offsets
(starting by 1 because off ++i) don't apply correctly in other
structures. Maybe the code for subgroups was added by a different
programmer.
> I looked around in the code and thought about where an "off by one"
> problem could be coming from, and I don't see it.
My big problem: I don't find the code of getPluginLatency. I only find a
declaration in sound/AudioProcess.h:
size_t getPluginLatency(unsigned int id);
In sound/AudioProcess.cpp there is
1305 size_t
1306 AudioInstrumentMixer::getPluginLatency(unsigned int id)
1307 {
1308 // Not RT safe
1309
1310 size_t latency = 0;
1311
1312 RunnablePluginInstance *synth = m_synths[id];
1313 if (synth)
1314 latency += m_synths[id]->getLatency();
1315
1316 for (PluginList::iterator i = m_plugins[id].begin();
1317 i != m_plugins[id].end(); ++i) {
1318 RunnablePluginInstance *plugin = *i;
1319 if (plugin)
1320 latency += plugin->getLatency();
1321 }
1322
1323 return latency;
1324 }
But my lack of C++-knowledge prevents me of mapping this to
sound/JackDriver.cpp:
2086 } else if (!empty) {
2087 pluginLatency +=
2088 m_instrumentMixer->getPluginLatency((unsigned int) *
connections.begin());
2089 }
And even then the off-by-one-problem (experimentally proven and
reproducable) could be in the mixer's data structures. My stomach tells
me that the problem is located in connections.begin and
m_instrumentMixer in line 2088.
It's not satisfacting to simply subtract 1 without understanding if it
opens the door to other collateral damages.
Maybe I'll try hightech-debugging with printf to see if there are zeroes
returned in real life.
Regards
Holger
------------------------------------------------------------------------------
Precog is a next-generation analytics platform capable of advanced
analytics on semi-structured data. The platform includes APIs for building
apps and a phenomenal toolset for data science. Developers can use
our toolset for easy data analysis & visualization. Get a free account!
http://www2.precog.com/precogplatform/slashdotnewsletter
_______________________________________________
Rosegarden-devel mailing list
[email protected] - use the link below to unsubscribe
https://lists.sourceforge.net/lists/listinfo/rosegarden-devel
