BUG: Memory leak bugs due to returned new reference is not decreased on failure (static analyzer reports)

[Snape3058]

Returning a new reference here:

pyxattr/xattr.c

Line 632 in c3466e7

my_tuple = Py_BuildValue("yy#", name, buf_val, nval);

Variable my_tuple goes out of scope without decreasing the refcnt.

pyxattr/xattr.c

Line 639 in c3466e7

goto free_buf_val;

Internal Report ID: 19f80b

---

Returning a new reference here:

pyxattr/xattr.c

Line 1185 in c3466e7

PyObject *m = PyModule_Create(&xattrmodule);

Error handling code after goto target does not decrease the refcnt. (also other gotos in this function)

pyxattr/xattr.c

Line 1201 in c3466e7

goto err_out;

Internal Report ID: 1eed62

[iustin]

Thanks for the report, much appreciated! I'll take a look in some while, right now away from proper machine to test.

Just for my curiosity, which analyser did you use? I'd like to integrate it in GitHub hooks or at least release process, if it's freely available.

[Snape3058]

It is an experimental analyzer developed on the top of Clang Static Analyzer. When the tool is publicly available, I will reply to this issue to tell you. Currently, I am busy writing the paper for this research.

[iustin]

Sounds good, thank you - and thanks for finding the bugs!

[iustin]

I've committed a fix for the first issue, which was very clear. For the second batch of issues, in the module initalisation, I'm not entirely sure what's the correct way of handling initialisation error - is deconstructing (via ref count decrease) the module the right thing? It seems so, but testing behaviour here is hard.

[iustin]

Ah, actually the tutorial shows exactly this. Then it's a clear and simple fix.