Segfaults in version 4.3

[jamesjer]

I am the Fedora maintainer of the gnofract4d package. I am working on updating our package to gnofract4d 4.3, but am seeing segfaults in the test suite. You can see an example here:

https://koji.fedoraproject.org/koji/taskinfo?taskID=48696332

The segfaults appear to be random; i.e., on this run, it was the arm and ppc64le builders that hit the issue. On other runs, it is a different architecture or architectures. The only commonalities I have noted are:

• The segfaults always happen while running fract4dgui/tests/test_browser.py
• The backtrace always points to somewhere in fract4d_compiler, although different places on different runs

[jamesjer]

FYI, I tried adding -DDEBUG_ALLOCATION as suggested in #109, but it looks like an #include is missing:

fract4d/c/fract_stdlib.cpp: In function ‘bool arena_add_page(arena_t)’:
fract4d/c/fract_stdlib.cpp:174:13: error: ‘stderr’ was not declared in this scope
  174 |     fprintf(stderr, "%p: ARENA PAGE : CTOR\n", newalloc);
      |             ^~~~~~
fract4d/c/fract_stdlib.cpp:7:1: note: ‘stderr’ is defined in header ‘<cstdio>’; did you forget to ‘#include <cstdio>’?
    6 | #include "fract_stdlib.h"
  +++ |+#include <cstdio>
    7 | 
fract4d/c/fract_stdlib.cpp:174:5: error: ‘fprintf’ was not declared in this scope; did you mean ‘rintf’?
  174 |     fprintf(stderr, "%p: ARENA PAGE : CTOR\n", newalloc);
      |     ^~~~~~~
      |     rintf
fract4d/c/fract_stdlib.cpp: In function ‘s_arena* arena_create(int, int)’:
fract4d/c/fract_stdlib.cpp:195:13: error: ‘stderr’ was not declared in this scope
  195 |     fprintf(stderr, "%p: ARENA : CTOR(%d,%d)\n", arena, page_size, max_pages);
      |             ^~~~~~
fract4d/c/fract_stdlib.cpp:195:13: note: ‘stderr’ is defined in header ‘<cstdio>’; did you forget to ‘#include <cstdio>’?
fract4d/c/fract_stdlib.cpp:195:5: error: ‘fprintf’ was not declared in this scope; did you mean ‘rintf’?
  195 |     fprintf(stderr, "%p: ARENA : CTOR(%d,%d)\n", arena, page_size, max_pages);
      |     ^~~~~~~
      |     rintf
fract4d/c/fract_stdlib.cpp: In function ‘void* arena_alloc(arena_t, int, int, int*)’:
fract4d/c/fract_stdlib.cpp:244:13: error: ‘stderr’ was not declared in this scope
  244 |     fprintf(stderr, "%p: ALLOC : (req=%d,esize=%d,nelem=%d)\n",
      |             ^~~~~~
fract4d/c/fract_stdlib.cpp:244:13: note: ‘stderr’ is defined in header ‘<cstdio>’; did you forget to ‘#include <cstdio>’?
fract4d/c/fract_stdlib.cpp:244:5: error: ‘fprintf’ was not declared in this scope; did you mean ‘rintf’?
  244 |     fprintf(stderr, "%p: ALLOC : (req=%d,esize=%d,nelem=%d)\n",
      |     ^~~~~~~
      |     rintf
fract4d/c/fract_stdlib.cpp: In function ‘void arena_delete_page(allocation_t*)’:
fract4d/c/fract_stdlib.cpp:259:13: error: ‘stderr’ was not declared in this scope
  259 |     fprintf(stderr, "%p: ARENA PAGE : DTOR\n", alloc);
      |             ^~~~~~
fract4d/c/fract_stdlib.cpp:259:13: note: ‘stderr’ is defined in header ‘<cstdio>’; did you forget to ‘#include <cstdio>’?
fract4d/c/fract_stdlib.cpp:259:5: error: ‘fprintf’ was not declared in this scope; did you mean ‘rintf’?
  259 |     fprintf(stderr, "%p: ARENA PAGE : DTOR\n", alloc);
      |     ^~~~~~~
      |     rintf
fract4d/c/fract_stdlib.cpp: In function ‘void arena_delete(arena_t)’:
fract4d/c/fract_stdlib.cpp:272:13: error: ‘stderr’ was not declared in this scope
  272 |     fprintf(stderr, "%p: ARENA : DTOR()\n", arena);
      |             ^~~~~~
fract4d/c/fract_stdlib.cpp:272:13: note: ‘stderr’ is defined in header ‘<cstdio>’; did you forget to ‘#include <cstdio>’?
fract4d/c/fract_stdlib.cpp:272:5: error: ‘fprintf’ was not declared in this scope; did you mean ‘rintf’?
  272 |     fprintf(stderr, "%p: ARENA : DTOR()\n", arena);
      |     ^~~~~~~
      |     rintf

[mindhells]

Not sure if it has something to do, but as I can see here, the tests are running on Python 3.9, which is not even an official release:

platform linux -- Python 3.9.0b5, pytest-5.4.3, py-1.9.0, pluggy-0.13.1

The DEBUG_ALLOCATION macro it's meant to print some debug information from the C++ fract4dc extension, but I think it won't help here since the errors you are getting are not related to that extension (apparently). Nevertheless we need to review those missing includes.

[josebailo]

Hey @jamesjer, do you have any kind of metrics about the use of the package like how many downloads/installations has?

[jamesjer]

Sorry for the delay. No, I'm afraid the Fedora project doesn't have any reliable way of collecting such metrics. It would be interesting information to have, but I'm afraid I can't provide it.

I will try building with python 3.8 and see what happens.

[jamesjer]

A python 3.8 build finished flawlessly on all architectures. I'll see if I can figure out why python 3.9 is having issues, but this does seem to point away from gnofract4d and towards python as the cause of the problem.

[jamesjer]

This turns out to be a gnofract4d bug after all. See #166 for an explanation.

[mindhells]

@jamesjer #166 has been merged, please check if you can now create the package and close this issue.