# Configure GDB in a reasonable way

set breakpoint pending on

set pagination off

# Suppress "Missing separate debuginfos" message:

## This is now done in 'mcmini forward'

## set build-id-verbose 0

# GDB will track both parent and child

set detach-on-fork off

set print pretty

set print address off

# In McMini, parent sends SIGUSR1 to child on exit.

handle SIGUSR1 nostop noprint pass

handle SIGUSR2 nostop noprint pass

# Allow the other inferior to continue to execute if not at breakpoint

set schedule-multiple on

## Optional for additional modes for threads/inferios:

## SEE: https://stackoverflow.com/questions/27140941/preventing-debugging-session-from-pausing-after-each-inferior-exits

# set non-stop on

# set target-async on

# FIXME:

## Newer GDBs can use "tui disable" to undo "layout src"

## We should test if "tui disable" works, and then use it.

# source [$MCMINI_ROOT/]gdbinit_commands.py"

python import os, sys

python DIR = ""

python if os.environ.get("MCMINI_ROOT"): DIR = os.environ["MCMINI_ROOT"]+"/"

python if DIR: gdb.execute("dir " + DIR)

python if os.path.isfile("NO-GDB-G3"): \

print("\n*** Not compiled with -g3. Please call:\n" + \

"*** make clean && make -j9 debug"); \

sys.exit(1)

python print("source " + DIR + "gdbinit_commands.py")

python gdb.execute("source " + DIR + "gdbinit_commands.py")

# Stop at main in scheduler.

# Later, whenever we fork into target process, stop at main.

break main

run

tbreak execvp

continue

python if (not gdb.selected_inferior().threads()): sys.exit(1)

# Stop if about to exit:

break _exit

set variable $bpnum_exit = $bpnum

tbreak 'mcmini_main()'

continue

## WE WANT TO DO: break __real_sem_init

## Unfortunately, it's a macro. So, it's tricky to set breakpoint on it.

## In next version, let's make it a function,

## with a local variable:

## __real_sem_init(...) {

## static typeof(&sem_init) sem_init_ptr = NULL;

## if (sem_init_ptr == NULL) { dlsym(RTLD_NEXT, "sem_init"); }

## (*sem_init_ptr)(...);

## }

#### This variation will actually put a breakpoint on sem_init, since

#### *sem_init_ptr and sem_init are the same address in memory. :-(

## We need to break at sem_init only for debugging McMini

### FIXME: If I comment these lines out, I see a bug when doing:

### (gdb) finishTrace

### (gdb) nextTrace

# break mc_load_intercepted_symbol_addresses

# continue

# finish

# break *&sem_init_ptr

### The above works with gcc-4.8. Higher versions of gcc need following:

### See abvoe for how to rewrite sem_init_ptr, etc., to avoid this issue.

## break *(long *)&sem_init_ptr

## break *(void **)&sem_init_ptr

# continue

# disable $bpnum

# break 'mc_initialize_trace_sleep_list()'

# break *sem_init_ptr

# tbreak fork

# continue

# continue

# Continue through 'fork' and into 'main' for target (child) process

tbreak main

continue

## Python nextTransition will set this:

## break mc_new_trace_at_current_state()

## break mc_shared_sem_wait_for_scheduler

# break thread_await_scheduler_for_thread_start_transition

# NOT NEEDED: set follow-fork-mode child

# We 'set detach-on-fork off'

# So, we can set breakpoints in parent or child process, and they take effect.

# Print Python-based GDB commands:

help user-defined

echo \n\ \ *** Type 'mcmini help' for usage. ***\n

echo \ \ (Do 'set print address on' for more verbose outpus.)\n\n