Percolator: input error under win (v 1.17) (Carmen, Thermo) about percolator

[mattiat]

_ the provided input.xml file works in linux but not under windows
_ "something" happens at the level of Caller::xv_step() (but only once the code is compiled for windows??)
_ under windows, the error disappears if and extra "featureDescription" is added in the input.xml

[mattiat]

_last bit of code that gets executed is ssl::CGLS() (line 323), right before releasing some dynamic memory. Seems to be a memory allocation problem

[mattiat]

_ error occurs if input.xml if number of "featureDescription"<=9
_ (on cerr) FeatureNames::getNumFeatures reports the correct value

[mattiat]

_ removing empty "FragmentSpecturmScan"s in input.xml has no effect

[mattiat]

_ cross-debugging: minGW debugger + Wine (on Fedora): -G"Eclipse CDT4 - Unix Makefiles" flag combined with minGW toolchain generates an eclipse project with .exe binaries!

[mattiat]

Reached Wine through remote debugging with gdbserver (http://mingw-cross.sourceforge.net/cross_debug.html). gdbserver.exe distributed with minGW for windows.
_ percolator.exe now crashes at the return 0 statement of Caller::run() (executable shows different behavior on windows and emulated under wine)

[mattiat]

_ Memory allocated by in Caller::readFiles() line 489 by xercesc::XMLPlatformUtils::Initialize() was not being freed: added call to method xercesc::XMLPlatformUtils::Terminate() in destructor of Caller object! -> SOLVED 99% of memory leaks!
http://stackoverflow.com/questions/4680191/when-calling-xmlplatformutilsinitialize-on-xerces

[mattiat]

SOLVED: after svn training, a check of correctness is performed by Caller::pCheck, an object of static type SanityCheck. It should purposively be instantiated according to type of file used in creating the pin file given in input to Percolator. Instead, it is always assigned an object of dynamic type SqtSanityCheck. As a consequence the following implementation of function getDefaultDirection is used which assumes an sqt file was used.

SqtSanityCheck::getDefaultDirection() {
// some code
for (unsigned int ix = 0; ix < numFeatures + 1; ++ix) {
    w[set][ix] = 0;
}
// some more code
w[set][10] = -0.156; // writing to uninitialized vector position

The problem is that the above instructions assume that the pin file contains at least 10 features (as any pin file generate from sqt would). If it does not, an uninitialized position of the std::vector w[set] is accessed and a memory fault might occur (it is up to the operating system to decide how to handle the basic data structure underlying a vector).

The solution I implemented adds a method to the class Caller that takes care of initializing its SanityCheck field to the appropriate value; it does this through a (potentially extendible series of) if statement(s) so that an object of the most suitable dynamic type is instantiated. I decided not to do this inside the SanityCheck class because the information that is needed to decide which subclass to instantiate is available at the level of the Caller object. In case nothing more specific seems to apply, I instantiate an object of base class SanityCheck().

void Caller::initializeSanityCheck() {
if(otherCall.find(SqtSanityCheck::fingerPrint)!= string::npos)
    pCheck = new SqtSanityCheck();
else // if (SomeOtherSanityCheck) 
      // {pCheck = new SomeOtherSanityCheck();} else
    pCheck = new SanityCheck();
}

Resolved with commit 2b79177

EDIT: eventually moved initialization logic inside SanityCheck, commit aac5107