[Bug c++/29918] New: GDB cannot step or break into template functions compiled with g++ -Og

[Bug c++/29918] New: GDB cannot step or break into template functions compiled with g++ -Og

[vimacs.hacks at gmail dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=29918

            Bug ID: 29918
           Summary: GDB cannot step or break into template functions
                    compiled with g++ -Og
           Product: gdb
           Version: 12.1
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: c++
          Assignee: unassigned at sourceware dot org
          Reporter: vimacs.hacks at gmail dot com
  Target Milestone: ---

This is another issue posted on
https://sourceware.org/pipermail/gdb/2022-November/050431.html

For the following C++ code, I find that GDB (both 12.1 and
gdb-13-branchpoint-5-g2a517ca94d3) cannot step or break into C++ template
function t0<int> compiled with g++ -Og (GCC version 12.2.0), while LLDB from
LLVM 14.0.6 can step into this function:

template <typename T = int>
int t0(int n)
{
     int sum = 0;
     for (int i = 0; i < n; ++i) {
         sum += i * i;
     }
     return sum;
}

int t1(int n)
{
     int sum = 0;
     for (int i = 0; i < n; ++i) {
         sum += i * i;
     }
     return sum;
}

int main()
{
     int a = t0(5);
     int b = t1(5);
     return a + b;
}

The behavior in GDB looks like this:

Breakpoint 1, main () at template_func.cc:21
21      {
(gdb) n
22           int a = t0(5);
(gdb) s
t0<int> (n=n@entry=5) at template_func.cc:2
2       int t0(int n)
(gdb) 
main () at template_func.cc:23
23           int b = t1(5);
(gdb) 

The assembly output of "g++-g -Og" and "g++ -g -O0" are both attached in the
attachment of the following comments, which can be assembled and reproduce this
issue.

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug c++/29918] GDB cannot step or break into template functions compiled with g++ -Og

[vimacs.hacks at gmail dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=29918

--- Comment #1 from vimacs.hacks at gmail dot com <vimacs.hacks at gmail dot com> ---
Created attachment 14523
  --> https://sourceware.org/bugzilla/attachment.cgi?id=14523&action=edit
Assembly output with -g -O0

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug c++/29918] GDB cannot step or break into template functions compiled with g++ -Og

[vimacs.hacks at gmail dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=29918

--- Comment #2 from vimacs.hacks at gmail dot com <vimacs.hacks at gmail dot com> ---
Created attachment 14524
  --> https://sourceware.org/bugzilla/attachment.cgi?id=14524&action=edit
Assembly output with -g -Og

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug c++/29918] GDB cannot step or break into template functions compiled with g++ -Og

[tromey at sourceware dot org]

https://sourceware.org/bugzilla/show_bug.cgi?id=29918

Tom Tromey <tromey at sourceware dot org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
     Ever confirmed|0                           |1
                 CC|                            |tromey at sourceware dot org
   Last reconfirmed|                            |2022-12-19
             Status|UNCONFIRMED                 |NEW

--- Comment #3 from Tom Tromey <tromey at sourceware dot org> ---
I was able to reproduce with the Fedora 36 system gcc.
However, the line table that is emitted is weird.

First, consider the line table for lines 12-18.  This
corresponds to the function t1:

File name                            Line number    Starting address    View   
Stmt
prog.cc                                       12            0x400576           
   x
prog.cc                                       13            0x400576       1   
   x
prog.cc                                       14            0x400576       2   
   x
prog.cc                                       14            0x400576       3
prog.cc                                       13            0x40057b        
prog.cc                                       14            0x400580        
prog.cc                                       15            0x400582           
   x
prog.cc                                       15            0x400582       1
prog.cc                                       15            0x400587        
prog.cc                                       14            0x400589           
   x
prog.cc                                       14            0x40058c           
   x
prog.cc                                       17            0x400590           
   x
prog.cc                                       18            0x400590       1
prog.cc                                       18            0x400592        

Not that many lines are represented and multiple lines in this range have the
statement marker.

Now consider the line table for lines 2-9, aka t0<>:

File name                            Line number    Starting address    View   
Stmt
prog.cc                                        2            0x4005ae           
   x
prog.cc                                        5            0x4005ae       1
prog.cc                                        4            0x4005b3        
prog.cc                                        5            0x4005b8        
prog.cc                                        6            0x4005ba        
prog.cc                                        6            0x4005bf        
prog.cc                                        5            0x4005c1           
   x
prog.cc                                        5            0x4005c4           
   x
prog.cc                                        9            0x4005c8        
prog.cc                                        9            0x4005ca        

This seems weird.

Perhaps gdb can work around it, after all lldb does.
I tend to suspect one of the line table patches that went in in the 12 era,
though I haven't tried bisecting or anything.

Anyway, while I agree gdb can be better, I think there's also a compiler bug.

-- 
You are receiving this mail because:
You are on the CC list for the bug.