Memory leak in CRFSuite::Tagger

[nikolsky]

I use crfsuite tagger in simple test program:

CRFSuite::Tagger* newTagger = new CRFSuite::Tagger();
newTagger->open("test_model.model");
...
tagVec = newTagger->tag(items);
delete newTagger;

Running the program in valgrind would return an error such as:

==11280== 7,519,904 bytes in 1 blocks are definitely lost in loss record 2 of 2
==11280==    at 0x4C2DBB6: malloc (vg_replace_malloc.c:299)
==11280==    by 0x42041A: crf1dm_new (in /home/artem/projects/test_crfsuite/dist/Debug/GNU-Linux/test_crfsuite)
==11280==    by 0x415B8F: crf1m_create_instance_from_file (in /home/artem/projects/test_crfsuite/dist/Debug/GNU-Linux/test_crfsuite)
==11280==    by 0x403551: CRFSuite::Tagger::open(std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&) (crfsuite.hpp:289)
==11280==    by 0x403D1A: main (main.cpp:26)
==11280== 
==11280== LEAK SUMMARY:
==11280==    definitely lost: 7,519,904 bytes in 1 blocks
==11280==    indirectly lost: 0 bytes in 0 blocks
==11280==      possibly lost: 0 bytes in 0 blocks
==11280==    still reachable: 72,704 bytes in 1 blocks
==11280==         suppressed: 0 bytes in 0 blocks
==11280== 
==11280== For counts of detected and suppressed errors, rerun with: -v
==11280== ERROR SUMMARY: 1 errors from 1 contexts (suppressed: 0 from 0)

"Definitely lost" here is equal .model file size.
The place that valgrind points:

...
buffer = buffer_orig = (uint8_t*)malloc(size + 16);
if (buffer_orig = NULL) {
    goto error_exit;
}
...
return crf1dm_new_impl(buffer_orig, buffer, size);

where:

static crf1dm_t* crf1dm_new_impl(uint8_t* buffer_orig, const uint8_t* buffer, uint32_t size)
{
...
model->buffer_orig = buffer_orig;

then in "destructor" memory will never free:

void crf1dc_delete(crf1d_context_t* ctx)
{
...
    if (model->buffer_orig != NULL) {
        free(model->buffer_orig);
        model->buffer_orig = NULL;
    }
    model->buffer = NULL;
...

After my fix #74 valgrind shows that memory doesn't leak any more:

==27439== LEAK SUMMARY:
==27439==    definitely lost: 0 bytes in 0 blocks
==27439==    indirectly lost: 0 bytes in 0 blocks
==27439==      possibly lost: 0 bytes in 0 blocks
==27439==    still reachable: 72,704 bytes in 1 blocks
==27439==         suppressed: 0 bytes in 0 blocks
==27439== 
==27439== For counts of detected and suppressed errors, rerun with: -v
==27439== ERROR SUMMARY: 0 errors from 0 contexts (suppressed: 0 from 0)

[TrinhKien94]

I have same problem when i run crfsuite version 0.12:
==25384==
==25384== HEAP SUMMARY:
==25384== in use at exit: 1,320 bytes in 1 blocks
==25384== total heap usage: 1,139 allocs, 1,138 frees, 62,120,243 bytes allocated
==25384==
==25384== 1,320 bytes in 1 blocks are definitely lost in loss record 1 of 1
==25384== at 0x4C2FA50: calloc (vg_replace_malloc.c:711)
==25384== by 0x40E9A3: crf1dc_set_num_items (in /home/kl/gr/crfsuite-0.12/frontend/build/crfsuite)
==25384== by 0x40821B: tagger_set (in /home/kl/gr/crfsuite-0.12/frontend/build/crfsuite)
==25384== by 0x4053A9: tag (tag.c:321)
==25384== by 0x40584C: main_tag (tag.c:433)
==25384== by 0x404221: crfsuite (main.c:149)
==25384== by 0x40401F: main (main.c:105)
==25384==
==25384== LEAK SUMMARY:
==25384== definitely lost: 1,320 bytes in 1 blocks
==25384== indirectly lost: 0 bytes in 0 blocks
==25384== possibly lost: 0 bytes in 0 blocks
==25384== still reachable: 0 bytes in 0 blocks
==25384== suppressed: 0 bytes in 0 blocks
==25384==
==25384== For counts of detected and suppressed errors, rerun with: -v
==25384== ERROR SUMMARY: 1 errors from 1 contexts (suppressed: 0 from 0)
Valgrind show that some thing in fucntion crf1dc_set_num_items have leak. Then i add this line: free(ctx->state); after line 99 (after line: free(ctx->alpha_score);) of file crfsuite-0.12/lib/crf/src/crf1d_context.c then i make && make install again. Run valgrind with my program again and have no leak as below:
==31489==
==31489== HEAP SUMMARY:
==31489== in use at exit: 0 bytes in 0 blocks
==31489== total heap usage: 1,139 allocs, 1,139 frees, 62,120,243 bytes allocated
==31489==
==31489== All heap blocks were freed -- no leaks are possible
==31489==
==31489== For counts of detected and suppressed errors, rerun with: -v
==31489== ERROR SUMMARY: 0 errors from 0 contexts (suppressed: 0 from 0)
I think it's helpful for fix this issue.

[usptact]

@chokkan bump!

[usptact]

So this is is why we don't hear back from @chokkan :) He is playing soccer! :)