I have encountered reproducible segfaults on OpenBSD's love (version 11.5) when using it to run the game Arco, occurring some way into the second act in the NPC party member inventory:

(gdb) bt
#0  0x00000111a6befa07 in love::graphics::Polyline::render(love::Vector2 const*, unsigned long, unsigned long, float, float, bool) ()
   from /usr/local/lib/liblove-11.5.so.0.0
#1  0x00000111a6bda0d5 in love::graphics::Graphics::polyline(love::Vector2 const*, unsigned long) () from /usr/local/lib/liblove-11.5.so.0.0
#2  0x00000111a6c24ed4 in love::graphics::w_line(lua_State*) ()
   from /usr/local/lib/liblove-11.5.so.0.0
#3  0x0000011115378146 in lj_BC_FUNCC ()
   from /usr/local/lib/libluajit-5.1.so.1.0
#4  0x0000010ecd83f6d8 in ?? ()
#5  0x0000010ecd83f1db in ?? ()
#6  0x0000000000000000 in ?? ()

Adding debug symbols showed it happening at https://github.com/love2d/love/blob/main/src/modules/graphics/Polyline.cpp#L105. When I add a printf right above the line, it reveals that this happens when vertex_count is 0, so this tries to access vertices[-1] in that case.

With the diff in this PR that checks that vertex_count is greater than 0, I've run Arco without any recurrence of the issue, and no new issues with graphics rendering or other. No issues with a few other love 11.5 games that I tested it with either.

I am not sure about the logic behind vertices and vertex_count, so would appreciate a look if this is correct to avoid the array access with negative index...