Continue to debug after failed assertion on Linux?
Solution 1
You really want to recreate the behavior of DebugBreak. This stops the program in the debugger.
My googling of "DebugBreak linux" has turned up several references to this piece of inline assembly which is supposed to do the same.
#define DEBUG_BREAK asm("int $3")
Then your assert can become
#define ASSERT(TEST) if(!(TEST)) asm("int $3");
According to Andomar int 3 causes the cpu to raise interrupt 3. According to drpepper a more portable way to do this would be to call:
raise(SIGTRAP);
Solution 2
You can configure gdb to handle specific signals in a different way. For example, the following will cause SIGSTOP not to be treated as a stoppable event.
handle SIGSTOP nostop noprint pass
help handle
within gdb will give you more information.
Solution 3
Even better usability is achieved with
/*!
* \file: assert_x.h
* \brief: Usability Improving Extensions to assert.h.
* \author: Per Nordlöw
*/
#pragma once
#include <errno.h>
#include <signal.h>
#include <assert.h>
#ifdef __cplusplus
extern "C" {
#endif
#if !defined(NDEBUG)
# define passert(expr) \
if (!(expr)) { \
fprintf(stderr, "%s:%d: %s: Assertion `%s' failed.", \
__FILE__, __LINE__, __ASSERT_FUNCTION, __STRING(expr)); raise(SIGTRAP); \
}
# define passert_with(expr, sig) \
if (!(expr)) { \
fprintf(stderr, "%s:%d: %s: Assertion `%s' failed.", \
__FILE__, __LINE__, __ASSERT_FUNCTION, __STRING(expr)); raise(sig); \
}
# define passert_eq(expected, actual) \
if (!(expected == actual)) { \
fprintf(stderr, "%s:%d: %s: Assertion `%s' == `%s' failed.", \
__FILE__, __LINE__, __ASSERT_FUNCTION, __STRING(expected), __STRING(actual)); raise(SIGTRAP); \
}
# define passert_neq(expected, actual) \
if (!(expected != actual)) { \
fprintf(stderr, "%s:%d: %s: Assertion `%s' != `%s' failed.", \
__FILE__, __LINE__, __ASSERT_FUNCTION, __STRING(expected), __STRING(actual)); raise(SIGTRAP); \
}
# define passert_lt(lhs, rhs) \
if (!(lhs < rhs)) { \
fprintf(stderr, "%s:%d: %s: Assertion `%s' < `%s' failed.", \
__FILE__, __LINE__, __ASSERT_FUNCTION, __STRING(lhs), __STRING(rhs)); raise(SIGTRAP); \
}
# define passert_gt(lhs, rhs) \
if (!(lhs > rhs)) { \
fprintf(stderr, "%s:%d: %s: Assertion `%s' < `%s' failed.", \
__FILE__, __LINE__, __ASSERT_FUNCTION, __STRING(lhs), __STRING(rhs)); raise(SIGTRAP); \
}
# define passert_lte(lhs, rhs) \
if (!(lhs <= rhs)) { \
fprintf(stderr, "%s:%d: %s: Assertion `%s' <= `%s' failed.", \
__FILE__, __LINE__, __ASSERT_FUNCTION, __STRING(lhs), __STRING(rhs)); raise(SIGTRAP); \
}
# define passert_gte(lhs, rhs) \
if (!(lhs >= rhs)) { \
fprintf(stderr, "%s:%d: %s: Assertion `%s' >= `%s' failed.", \
__FILE__, __LINE__, __ASSERT_FUNCTION, __STRING(lhs), __STRING(rhs)); raise(SIGTRAP); \
}
# define passert_zero(expr) \
if (!(expr == 0)) { \
fprintf(stderr, "%s:%d: %s: Assertion `%s' is zero failed.", \
__FILE__, __LINE__, __ASSERT_FUNCTION, __STRING(expr)); raise(SIGTRAP); \
}
#else
# define passert(expr)
# define passert_with(expr, sig)
# define passert_eq(expected, actual)
# define passert_lt(lhs, rhs)
# define passert_gt(lhs, rhs)
# define passert_lte(lhs, rhs)
# define passert_gte(lhs, rhs)
# define passert_zero(expr)
#endif
#ifdef __cplusplus
}
#endif
Solution 4
Have you tried to send a SIGCONT signal to the process?
kill -s SIGCONT <pid>
drpepper
Updated on June 03, 2022Comments
-
drpepper almost 2 years
When an assertion fails with Visual C++ on Windows, the debugger stops, displays the message, and then lets you continue (or, if no debugging session is running, offers to launch visual studio for you).
On Linux, it seems that the default behavior of assert() is to display the error and quit the program. Since all my asserts go through macros, I tried to use signals to get around this problem, like
#define ASSERT(TEST) if(!(TEST)) raise(SIGSTOP);
But although GDB (through KDevelop) stops at the correct point, I can't seem to continue past the signal, and sending the signal manually within GDB just leaves me hanging, with control of neither GDB nor the debugged process.
-
Andomar over 14 yearsIt will cause the CPU to raise interrupt 3 (faydoc.tripod.com/cpu/int3.htm) The debugger has an interrupt handler registered for interrupt 3, and will break the program.
-
drpepper over 14 yearsperfect! it catches a SIGTRAP event, stops on a dime, and then lets me continue! thanks a lot.
-
drpepper over 14 yearsto make it a bit more portable, i replaced the assembly with the equivalent c code: raise(SIGTRAP); works great.
-
Lightness Races in Orbit almost 10 yearsYour macro is broken in the presence of surrounding
else
-
jheriko almost 9 yearsits worth noting that the int 3 is intel specific, where as raise(SIGTRAP) has worked for me in iOS and Android on ARM32/64 and MIPS, and i suspect works everywhere by virtue of being part of the standard library as well.