> I'd like to hear your thoughts about these logs before trying
> to get into them and understand what's wrong.
The memory losses are sorted by size, so maybe we can try to fix at least the
biggest memory leaks.
For instance:
==22649== 47,920 bytes in 960 blocks are indirectly lost in loss record 920 of
921
==22649== at 0x402471C: operator new(unsigned int) (vg_replace_malloc.c:255)
==22649== by 0x4E75627: std::basic_string<wchar_t,
std::char_traits<wchar_t>, std::allocator<wchar_t> >::_Rep::_S_create(unsigned
int, unsigned int, std::allocator<wchar_t> const&) (in
/usr/lib/libstdc++.so.6.0.15)
==22649== by 0x8182DD8: lyx::Paragraph::asString(int, int, int) const
(basic_string.h:1543)
==22649== by 0x818C5F7: lyx::Paragraph::collectWords() (Paragraph.cpp:3503)
==22649== by 0x818C919: lyx::Paragraph::updateWords() (Paragraph.cpp:3531)
...
Refers to:
void Paragraph::collectWords()
{
// This is the value that needs to be exposed in the preferences
// to resolve bug #6760.
static int minlength = 6;
pos_type n = size();
for (pos_type pos = 0; pos < n; ++pos) {
if (isWordSeparator(pos))
continue;
pos_type from = pos;
locateWord(from, pos, WHOLE_WORD);
if (pos - from >= minlength) {
docstring word = asString(from, pos, AS_STR_NONE);
FontList::const_iterator cit =
d->fontlist_.fontIterator(pos);
if (cit == d->fontlist_.end())
return;
Language const * lang = cit->font().language();
d->words_[*lang].insert(word); // HERE HERE HERE HERE
HERE
}
}
}
We call insert on words_[*lang], that is:
class Paragraph::Private {
[...]
typedef set<docstring> Words;
typedef map<Language, Words> LangWordsMap;
///
LangWordsMap words_;
[...]
};
In the destructor:
Paragraph::~Paragraph()
{
deregisterWords();
delete d;
}
void Paragraph::deregisterWords()
{
Private::LangWordsMap::const_iterator itl = d->words_.begin();
Private::LangWordsMap::const_iterator ite = d->words_.end();
for (; itl != ite; ++itl) {
WordList * wl = theWordList(itl->first);
Private::Words::const_iterator it = (itl->second).begin();
Private::Words::const_iterator et = (itl->second).end();
for (; it != et; ++it)
wl->remove(*it);
}
d->words_.clear();
}
But in wl->remove...
void WordList::remove(docstring const & w)
{
Impl::Words::iterator it = d->words_.find(w);
if (it != d->words_.end()) {
it.data()--;
d->words_.change_weight(it, 0);
// We will not erase here, but instead we just leave it
// in the btree with weight 0. This avoid too much
// reorganisation of the tree all the time.
//if (it.data() == 0)
// d->words_.erase(w);
}
}
If I'm correct we're not removing the words to avoid tree reorganization, but
when we call delete d; in the destructor all the data remains orphan: 47,920
bytes are indirectly lost.
This is a serious memory leak IMHO, as paragraph are created and destroyed
quite frequently.
venom00
