gdb: how do you get the max value of a std::vector?

Question

I am debugging some C++ code and I have a really large std::vector<floating_point_type> (float or double). In C++ code when I want the max value of the std::vector I can just use std::max_element(). So in gdb I tried to use the same code, but I got an error.

In gdb I used call std::max_element(x) and got an error message No symbol "max_element" in namespace "std".

Is there any way to get the max value of a vector in gdb?

I would also appreciate an explanation for why my attempt at using std::max_element was not working (perhaps std::max_element a header-only or inline function).

`std::max_element()` is a template function so I am afraid gdb is not clever enough to instantiate it on the fly. — Slava, Jun 18 '18 at 13:28
Gdb lets you write functions (in it's own scripting language or python) that you can call to display your vector: they can trivially iterate the elements to find a maximum then display that. [this answer](https://stackoverflow.com/a/25499805/410767) should help you get started. — Tony Delroy, Jun 18 '18 at 13:52
@TonyDelroy i purposely wrote the question open ended because i was thinking of writing my own for loop to find the max element. your comment could easily be a solution (whether in gdb language or python language.) — Trevor Boyd Smith, Jun 18 '18 at 16:59

score 14 · Accepted Answer · answered Jun 18 '18 at 13:29

14

std::max_element is a function template, not a function. You are asking GDB to do template argument deduction and the whole shebang involved with calling a template function without specifying arguments. It can't do that, naturally, it's not a full fledged compiler.

As far as I know any solution, from the simplest to the most complex, will require of you to modify your source in such a way that std::max_element is instantiated for the iterator types of your vector. So you may as well add "debug only" code that computes the maximum element and stores it into a local variable.

answered Jun 18 '18 at 13:29

StoryTeller - Unslander Monica

165,132
21
377
458

for most cases, i think compiling/instantiating a "debug only" function is the most straight forward solution to get access to `std::max_element` from `gdb`. for some minority cases, i think @n.m. has a good solution (although for my eyes @n.m.'s looks difficult/very-complex (his solution probably only seems difficult to me because I'm a newbie at C++ with only 17+ years experience)). – Trevor Boyd Smith Jun 19 '18 at 13:27
[here is some good example code that someone created after your solution to demonstrate what you are describing](https://stackoverflow.com/a/50912371/52074). – Trevor Boyd Smith Jun 19 '18 at 13:28
@TrevorBoydSmith - Well. To be fair, n.m.'s solution is "the most complex solution" I had in mind. It just brilliantly manages to avoid restarting the debugging session. – StoryTeller - Unslander Monica Jun 19 '18 at 13:29

score 6 · Answer 2 · answered Jun 18 '18 at 15:03

(gdb) p std::max⭾⭾⭾

(No response from gdb)

 (gdb) p std::max_element(v.begin(), v.end())
 No symbol "max_element" in namespace "std".

Here is how to call an uninstantiated function template from gdb without breaking your session and recompiling your program.

Open an editor and create a C++ source file with an explicit instantiation of the needed function. For example:
```
// /tmp/tmpsource.cpp
#include <algorithm>
#include <vector>
using T = std::vector<int>::iterator;
template T std::max_element<T>(T, T);
```
This can probably be automated somehow with a shell script. Give it a name of a function and template arguments, and it will generate you a complete buildable C++ source. Probably not worth the trouble though.

Build a shared library/DLL from the source.

g++ -fPIC -shared -ggdb -O0 -o /tmp/libtmpsource.so /tmp/tmpsource.cpp

Load the library in your debugging session.

load /tmp/libtmpsource.so
# if this doesn't work
p dlopen("/tmp/libtmpsource.so", 2)
# or perhaps even
p LoadLibraryA("c:/temp/libtmpsource.so")

Let's try the function now.

(gdb) p std::max_element(v.begin(), v.end())
No symbol "max_element" in namespace "std".

What? No worries, everything is under control. gdb is not a C++ compiler and it cannot do the template deduction thing. You need to specify your <...> yourself. Fortunately, autocompletion works (sometimes).

(gdb) p std::max⭾
(gdb) p std::max_element<__gnu_cxx::__normal_iterator<int*, std::vector<int, 
 std::allocator<int> > > >(__gnu_cxx::__normal_iterator<int*, std::vector<int, 
 std::allocator<int> > >, __gnu_cxx::__normal_iterator<int*, std::vector<int, 
 std::allocator<int> > >)

Guess I was wrong. There is a sufficiently complex solution that doesn't involve changing the debugging sessions' source code :D +1 — StoryTeller - Unslander Monica, Jun 18 '18 at 15:08
**starts reading... gets to part about "just open an editor, compile, then go to gdb and load the shared object library. laughs out loud** — Trevor Boyd Smith, Jun 18 '18 at 16:54
**starts to read section on you need to specify the template parameters because gdb can not deduct... reads the horrifyingly long template parameters. laughs out loud.** — Trevor Boyd Smith, Jun 18 '18 at 16:56
@TrevorBoydSmith yes they are horribly long but at least gdb tab-completes them for you. — n. m. could be an AI, Jun 18 '18 at 17:19
@TrevorBoydSmith Would you still be laughing if you discovered that you have to leave gdb and recompile after a 100+ hours session of chasing an elusive bug? — n. m. could be an AI, Jun 18 '18 at 17:22

score 3 · Answer 3 · answered Jun 18 '18 at 14:59

As std::max_elemente is a function template, you can not use it directly in gdb. What you can do, create a wrapper over it.

float my_max_element(std::vector<float>& vec) 
{
    return *(std::max_element(vec.begin(), vec.end());
}

Now you can call my_max_element in gdb.

gdb: how do you get the max value of a std::vector?

3 Answers3