Possible regression on gdb.gdb/complaints.exp (was: Re: [RFA 0/9] Radically simplify the complaint system)

[Sergio Durigan Junior]

On Tuesday, May 22 2018, Tom Tromey wrote:

> I must confess, I've long disliked the complaint system.  It got in
> the way a little bit, ages ago, when I tried to multi-thread the
> psymtab reader.  And, it got in the way a bit more during one of my
> attempts to format the "Reading symbols" output more nicely (spoiler:
> I have another approach to this in the works, but I took this detour
> first).  Aside from these things, the code also seemed unusually
> complex for the task it performed.
>
> This series radically simplifies the complaint system.  It removes
> most of the code -- which, I think, has never really been used.
>
> Tested by the buildbot.  I also locally tested complaints.exp with
> each patch in the series.

Hi Tom,

While preparing a Fedora GDB release for rawhide, I stumbled upon a
failure on gdb.gdb/complaints.exp.  Here's what I see here:

  (gdb) PASS: gdb.gdb/complaints.exp: run until breakpoint at captured_command_loop
  set stop_whining = 2
  (gdb) PASS: gdb.gdb/complaints.exp: set stop_whining = 2
  set var $cstr = "Register a complaint"
  (gdb) PASS: gdb.gdb/complaints.exp: set var $cstr = "Register a complaint"
  call complaint_internal ($cstr)
  free(): invalid pointer
  
  Thread 1 "xgdb" received signal SIGSEGV, Segmentation fault.
  0x00007ffff5c535ce in raise () from /lib64/libc.so.6
  The program being debugged was signaled while in a function called from GDB.
  GDB remains in the frame where the signal was received.
  To change this behavior use "set unwindonsignal on".
  Evaluation of the expression containing the function
  (complaint_internal(char const*, ...)) will be abandoned.
  When the function is done executing, GDB will silently stop.

While trying to reproduce it on my Fedora 27, I wasn't able to trigger
the failures.  Anyway, after a lot of time investigating this issue, I
found that you have to use a recent GCC (git HEAD as of 2018-06-04, for
example), and you have to compile GDB using -O2 -g.

I'm not sure if this is a regression introduced by your patch, so I also
ran a git-bisect on GCC and found the following possible culprit:

ea5d398198b93e37e9a343dfdb7660f71fdca404 is the first bad commit
commit ea5d398198b93e37e9a343dfdb7660f71fdca404
Author: hubicka <hubicka@138bc75d-0d04-0410-961f-82ee72b054a4>
Date:   Thu Oct 19 20:19:15 2017 +0000                                                         

Not sure if it makes sense...

Anyway, I'm reporting because I thought it was the right thing to do.
As per IRC discussions recently, I know gdb.gdb/ is fragile and one idea
(by Tom) is to replace it by selftests.

Thanks,

-- 
Sergio
GPG key ID: 237A 54B1 0287 28BF 00EF  31F4 D0EB 7628 65FC 5E36
Please send encrypted e-mail if possible
http://sergiodj.net/