OpenBSD · brlcad · Zulip Chat Archive

Stream: brlcad

Topic: OpenBSD

starseeker (Dec 18 2020 at 04:33):

No wonder I was never able to do anything with that core file on OpenBSD - it needs lldb, not gdb: image.png

starseeker (Dec 18 2020 at 04:39):

Not sure what to make of it, but finally got it...

Sean (Dec 18 2020 at 04:51):

looks like it's actually a pointer to mged_curr_dm->dm_mged_variables

Sean (Dec 18 2020 at 05:09):

I bet it's the offset getting calculated incorrectly.

loc is set to base + sdp->sp_offset
sdp is a line in the parsetab with sp_offset defined in the struct. the struct is mged_curr_dm->dm_mged_variables so it could be that mged_curr_dm is invalid or dm_mged_variables isn't initialized or dm_mged_variables is initailized but has bad offset values.

Sean (Dec 18 2020 at 05:16):

or mged_curr_dm is random garbage

Sean (Dec 18 2020 at 05:22):

next step would probably be to identify which mged_variables is being used/set there, check dm_curr-dm for validity, and find where it's initalizing the fields of dm_mged_variables.

starseeker (Dec 18 2020 at 13:31):

I wish I knew which test was doing that - "make check" produces it, but despite making a core file it doesn't halt the test. Suggests something isn't checking a test return code somewhere, unless it's possible to dump a core and still complete "successfully"...

starseeker (Dec 18 2020 at 13:33):

Hmm... wonder if lldb can tell me from the core file...

starseeker (Dec 18 2020 at 13:54):

It's a little hard to follow, but it looks like it's running "vars " and when I do that in isolation the output does look strange: image.png

starseeker (Dec 18 2020 at 15:08):

This is handy: https://www.nesono.com/sites/default/files/lldb%20cheat%20sheet.pdf

starseeker (Dec 18 2020 at 18:30):

Whee! valgrind doesn't like the vars command run, even on Linux.

Sean (Dec 18 2020 at 18:34):

yeah, that looks like memory corruption.

starseeker (Dec 18 2020 at 18:35):

I wonder if I updated mged_variables but never updated mged_vparse...

Sean (Dec 18 2020 at 18:36):

looks like mged_curr_dm has to be involved, so maybe related to the dm init logic and the specific invocation mode of the script (like running with redirected input or something)

starseeker (Dec 18 2020 at 19:44):

Wow. "mged> vars" shows valgrind issues at least as far back as 70555...

starseeker (Dec 18 2020 at 19:50):

Make that r62647 - getting old enough it's hard to test...

starseeker (Dec 18 2020 at 19:54):

OK, clearly not a release blocker...

starseeker (Dec 18 2020 at 19:58):

/me wonders if "vars" would be better called "settings" - for what it seems to be intended to be, I wouldn't have thought to look first for a "vars" command...