[Bug symtab/29105] new DWARF reader still slow

["simon.marchi at polymtl dot ca" <sourceware-bugzilla@sourceware.org>]

https://sourceware.org/bugzilla/show_bug.cgi?id=29105

Simon Marchi <simon.marchi at polymtl dot ca> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |simon.marchi at polymtl dot ca

--- Comment #4 from Simon Marchi <simon.marchi at polymtl dot ca> ---
Hi Tom,

I'm not sure why I am starting to see this only today, but I see a lot of tests
timing out.  gdb.base/sizeof.exp is one of them.  It looks like it's due to a
slowdown that started with the new DWARF reader patches.

My build uses -O0 and all the sanitizer and glibcxx debug stuff, so it doesn't
help, but I see the same relative slowdown without that.

I copied the test program of gdb.base/sizeof.exp to /tmp/sizeof so it could be
reused easily across builds.

The command I use is:

  time ./gdb -q --data-directory=data-directory -nx /tmp/sizeof -ex "start" 
-iex "set debug dwarf-read 0" -ex "p/d sizeof(short)" -batch

This replicates the first steps of the test.

Using 3d20b8d99a5:
./gdb -q --data-directory=data-directory -nx /tmp/sizeof -ex "start" -ex  -e 
1.18s user 0.17s system 116% cpu 1.157 total

Using 3d20b8d99a5^:
./gdb -q --data-directory=data-directory -nx /tmp/sizeof -ex "start" -ex  -e 
0.40s user 0.07s system 109% cpu 0.427 total

By toggling "set debug dwarf-read" to 1, I noticed that we are now expanding a
lot of CUs from libc's debug info when doing the "print sizeof":

[dwarf-read] process_queue: Expanding one or more symtabs of objfile
/usr/lib/debug/lib/x86_64-linux-gnu/libc-2.31.so ...
[dwarf-read] process_queue: Expanding symtab of CU at offset 0xd3cbe
[dwarf-read] process_queue: Done expanding CU at offset 0xd3cbe
[dwarf-read] process_queue: Done expanding symtabs of
/usr/lib/debug/lib/x86_64-linux-gnu/libc-2.31.so.
[dwarf-read] process_queue: Expanding one or more symtabs of objfile
/usr/lib/debug/lib/x86_64-linux-gnu/libc-2.31.so ...
[dwarf-read] process_queue: Expanding symtab of CU at offset 0x22bf6f
[dwarf-read] process_queue: Done expanding CU at offset 0x22bf6f
[dwarf-read] process_queue: Done expanding symtabs of
/usr/lib/debug/lib/x86_64-linux-gnu/libc-2.31.so.
[dwarf-read] process_queue: Expanding one or more symtabs of objfile
/usr/lib/debug/lib/x86_64-linux-gnu/libc-2.31.so ...
[dwarf-read] process_queue: Expanding symtab of CU at offset 0x7ef56e
[dwarf-read] process_queue: Done expanding CU at offset 0x7ef56e
[dwarf-read] process_queue: Done expanding symtabs of
/usr/lib/debug/lib/x86_64-linux-gnu/libc-2.31.so.
...

We expand ~1500 CUs.

Before, we didn't do that.  So having debug info for libc is important for
reproducing this, and a GDB configured with --prefix/usr or some other switch
that makes it find separate debug info in /usr/lib/debug (by default a local
build would look in /usr/local/lib/debug).  I am reproducing this on Ubuntu
20.04.

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