[Bug c++/26491] New: second run of program very slow

[Bug c++/26491] New: second run of program very slow

[noelgrandin at gmail dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=26491

            Bug ID: 26491
           Summary: second run of program very slow
           Product: gdb
           Version: 9.1
            Status: UNCONFIRMED
          Severity: minor
          Priority: P2
         Component: c++
          Assignee: unassigned at sourceware dot org
          Reporter: noelgrandin at gmail dot com
  Target Milestone: ---

Created attachment 12784
  --> https://sourceware.org/bugzilla/attachment.cgi?id=12784&action=edit
perf flame graph

version 9.1-5 running on Fedora 32

I am running LibreOffice unit tests.

(e.g. make CppunitTest_chart2_import)

The first time I debug something, I get this

(gdb) r
... starts within 10 seconds....
.. I stop stuff...
... process exits...
(gdb) r
... now it takes 5 mins to get going again ....

This is on a pretty decent box (4Ghz i7)

I attach a flame graph.

I am building with clang, and with 
  --enable-split-debug
  --enable-gdb-index
if that makes any difference.


Funnily, this only seems to apply to the unit tests, the actually main binary
(soffice) is fine.

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

[Bug c++/26491] second run of program very slow

[noelgrandin at gmail dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=26491

--- Comment #1 from Noel Grandin <noelgrandin at gmail dot com> ---
Created attachment 12791
  --> https://sourceware.org/bugzilla/attachment.cgi?id=12791&action=edit
patch to speed up demangling

This patch reduces the time taken to restart a program by 50%

Yes, the buffer really needs to be this big - C++ symbols in Libreoffice
frequently generate very long demangled names :-)

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

[Bug c++/26491] second run of program very slow

[simark at simark dot ca]

https://sourceware.org/bugzilla/show_bug.cgi?id=26491

Simon Marchi <simark at simark dot ca> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |simark at simark dot ca

--- Comment #2 from Simon Marchi <simark at simark dot ca> ---
Hi,

We don't do patch review on Bugzilla.  Would you mind sending the patch to the
relevant mailing lists?

Since this touches code shared by binutils and gcc, please include the
following mailing lists:

- gdb-patches@sourceware.org
- binutils@sourceware.org
- gcc-patches@gcc.gnu.org

Thanks,

Simon

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

[Bug c++/26491] second run of program very slow

[noelgrandin at gmail dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=26491

Noel Grandin <noelgrandin at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         Resolution|---                         |INVALID
             Status|UNCONFIRMED                 |RESOLVED

--- Comment #3 from Noel Grandin <noelgrandin at gmail dot com> ---
looks like I was mistaken, and a big chunk of the CPU time was caused by
unwittingly having  
    MALLOC_PERTURB_ 
    MALLOC_CHECK_ 
active for gdb itself.

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

[Bug c++/26491] second run of program very slow

[noelgrandin at gmail dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=26491

--- Comment #4 from Noel Grandin <noelgrandin at gmail dot com> ---
thanks to djdelorie for helping me debug that

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