On Tue, 3 Sep 2002, Bastian Kleineidam wrote: > Package: dselect > Version: 1.10.5 > Severity: normal > > Hi, > > alarmed by the segfaults in dselect I ran valgrind on the code and found > leaks. Eeeek!
I've been using valgrind alot lately. The next version of dpkg and dselect will have lots of memory fixes. The segfault is fixed in 1.10.6. > Start dselect, hit "update" and exit. > ==19937== 20 bytes in 1 blocks are definitely lost in loss record 5 of 36 > ==19937== at 0x400436E8: __builtin_vec_new (vg_clientfuncs.c:156) > ==19937== by 0x804D568: refreshmenu(void) > (/home/calvin/src/dpkg-1.10.5/dselect/main.cc:392) > ==19937== by 0x804D612: urq_menu(void) > (/home/calvin/src/dpkg-1.10.5/dselect/main.cc:406) > ==19937== by 0x804DA25: main > (/home/calvin/src/dpkg-1.10.5/dselect/main.cc:496) > ==19937== > ==19937== 143 bytes in 1 blocks are definitely lost in loss record 20 of 36 > ==19937== at 0x400436E8: __builtin_vec_new (vg_clientfuncs.c:156) > ==19937== by 0x804EFE0: readmethods(char const *, dselect_option **, int > *) (/home/calvin/src/dpkg-1.10.5/dselect/methparse.cc:76) > ==19937== by 0x804E674: ensureoptions(void) > (/home/calvin/src/dpkg-1.10.5/dselect/method.cc:89) > ==19937== by 0x804EBF1: runscript(char const *, char const *) > (/home/calvin/src/dpkg-1.10.5/dselect/method.cc:209) > ==19937== > > The first one is easy: lockfile is never delete[]d. So a simple > "delete[] lockfile" before the return should suffice. Already know of this one. > > The second one in methparse: pathbuf is not delete[]d in all > possible code paths, especially in the return on ENOENT. I dont know > what the ohshite function is doing - possibly bailing out, so thats > another candidate. This one too. ohshit* and friends are error functions. You don't need to worry about freeing memory if one of these are called. > After fixing the above two bugs the "update and exit" had no more memory > leaks (at least in my case). > I am trying the "update, select, exit" procedure now, but it is very slow, > eating a lot of memory. > Filing this bug for now. it'll take 1.1 gigs of memory.
