BUG: Memory leak when the module created by MOD_DEF is not decreased on failure (static analyzer report)

[Snape3058]

Macro MOD_DEF will create a module PyObject with a new reference returned and assigned to the first argument.

numba/numba/_pymodule.h

Line 16 in 468647d

ob = PyModule_Create(&moduledef); }

The following places invoke the macro but do not decrease the refcnt of the module PyObject on failure.

---

Invoked here:

numba/numba/_dispatcher.cpp

Line 1211 in 468647d

MOD_DEF(m, "_dispatcher", "No docs", ext_methods)

Function returns without decreasing the refcnt.

numba/numba/_dispatcher.cpp

Line 1217 in 468647d

return MOD_ERROR_VAL;

Internal Report ID: 0dc3d1

---

Invoked here:

numba/numba/_dynfuncmod.c

Line 66 in 468647d

MOD_DEF(m, "_dynfunc", "No docs", ext_methods)

Function returns without decreasing the refcnt.

numba/numba/_dynfuncmod.c

Line 71 in 468647d

return MOD_ERROR_VAL;

Internal Report ID: 235f50

---

Invoked here:

numba/numba/_helpermod.c

Line 288 in 468647d

MOD_DEF(m, "_helperlib", "No docs", ext_methods)

Function returns in macro import_array without decreasing the refcnt.

numba/numba/_helpermod.c

Line 292 in 468647d

import_array();

Internal Report ID: f3d5d2

---

Invoked here:

numba/numba/core/runtime/_nrt_pythonmod.c

Line 193 in 468647d

MOD_DEF(m, "_nrt_python", "No docs", ext_methods)

Function returns in macro import_array without decreasing the refcnt.

numba/numba/core/runtime/_nrt_pythonmod.c

Line 196 in 468647d

import_array();

Internal Report ID: 724a48

---

Invoked here:

numba/numba/experimental/jitclass/_box.c

Line 156 in 468647d

MOD_DEF(m, "_box", "No docs", ext_methods)

Function returns without decreasing the refcnt.

numba/numba/experimental/jitclass/_box.c

Line 162 in 468647d

return MOD_ERROR_VAL;

Internal Report ID: 31f898

---

Invoked here:

numba/numba/mviewbuf.c

Line 356 in 468647d

MOD_DEF(module, "mviewbuf", "No docs", core_methods)

Function returns without decreasing the refcnt.

numba/numba/mviewbuf.c

Line 362 in 468647d

return MOD_ERROR_VAL;

Internal Report ID: f68b83

---

Invoked here:

numba/numba/np/ufunc/_internal.c

Line 762 in 468647d

MOD_DEF(m, "_internal", "No docs",

Function returns without decreasing the refcnt.

numba/numba/np/ufunc/_internal.c

Line 769 in 468647d

return MOD_ERROR_VAL;

Internal Report ID: ed7b7c

[stuartarchibald]

Thanks for the report.

Macro MOD_DEF will create a module PyObject with a new reference returned and assigned to the first argument.

numba/numba/_pymodule.h

Line 16 in 468647d

ob = PyModule_Create(&moduledef); }

The following places invoke the macro but do not decrease the refcnt of the module PyObject on failure.

Invoked here:

numba/numba/_dispatcher.cpp

Line 1211 in 468647d

MOD_DEF(m, "_dispatcher", "No docs", ext_methods)

Function returns without decreasing the refcnt.

numba/numba/_dispatcher.cpp

Line 1217 in 468647d

return MOD_ERROR_VAL;

Internal Report ID: 0dc3d1

On inspection, this analysis looks valid.

Invoked here:

numba/numba/_dynfuncmod.c

Line 66 in 468647d

MOD_DEF(m, "_dynfunc", "No docs", ext_methods)

Function returns without decreasing the refcnt.

numba/numba/_dynfuncmod.c

Line 71 in 468647d

return MOD_ERROR_VAL;

Internal Report ID: 235f50

On inspection, this analysis looks valid.

Invoked here:

numba/numba/_helpermod.c

Line 288 in 468647d

MOD_DEF(m, "_helperlib", "No docs", ext_methods)

Function returns in macro import_array without decreasing the refcnt.

numba/numba/_helpermod.c

Line 292 in 468647d

import_array();

Internal Report ID: f3d5d2

On inspection, this analysis looks valid. Note that the macro import_array comes from NumPy, it contains the return NULL.

Invoked here:

numba/numba/core/runtime/_nrt_pythonmod.c

Line 193 in 468647d

MOD_DEF(m, "_nrt_python", "No docs", ext_methods)

Function returns in macro import_array without decreasing the refcnt.

numba/numba/core/runtime/_nrt_pythonmod.c

Line 196 in 468647d

import_array();

Internal Report ID: 724a48

On inspection, this analysis looks valid. Note that the macro import_array comes from NumPy, it contains the return NULL.

Invoked here:

numba/numba/experimental/jitclass/_box.c

Line 156 in 468647d

MOD_DEF(m, "_box", "No docs", ext_methods)

Function returns without decreasing the refcnt.

numba/numba/experimental/jitclass/_box.c

Line 162 in 468647d

return MOD_ERROR_VAL;

Internal Report ID: 31f898

On inspection, this analysis looks valid.

Invoked here:

numba/numba/mviewbuf.c

Line 356 in 468647d

MOD_DEF(module, "mviewbuf", "No docs", core_methods)

Function returns without decreasing the refcnt.

numba/numba/mviewbuf.c

Line 362 in 468647d

return MOD_ERROR_VAL;

Internal Report ID: f68b83

On inspection, this analysis looks valid.

Invoked here:

numba/numba/np/ufunc/_internal.c

Line 762 in 468647d

MOD_DEF(m, "_internal", "No docs",

Function returns without decreasing the refcnt.

numba/numba/np/ufunc/_internal.c

Line 769 in 468647d

return MOD_ERROR_VAL;

Internal Report ID: ed7b7c

On inspection, this analysis looks valid. There's also further places in that MOD_INIT(_internal) which have the same issue. Does the tool being used pick these up, or just the first case?

[Snape3058]

For the return NULL in import_array, we are discussing a solution in SciPy by moving the invocation of import_array to the very beginning of the function.
You can have a try on this solution to see whether it works for this project.