46543 - Support XPCOM_BREAK_ON_LOAD for Sun Solaris

Status: RESOLVED INCOMPLETE

(Core :: XPCOM, defect, P3)

Description

[Axel Hecht]

In
http://lxr.mozilla.org/seamonkey/source/xpcom/components/xcDll.cpp#483
there is only support for intel linux and VMS.
Find the debugger trap on Solaris and add the appropriate call there.

For those, that have not completely abandoned the Workshop Compilers, is there
a difference between dbx and gdb?

Axel

Comment 1

[Rich Burridge]

I've just had a quick look at the gdb source code. For SPARC there is
the following in .../gdb-5.0/gdb/sparc-stub.c

/* This function will generate a breakpoint exception.  It is used at the
   beginning of a program to sync up with a debugger and can be used
   otherwise as a quick means to stop program execution and "break" into
   the debugger. */

void
breakpoint()
{
  if (!initialized)
    return;

  asm(" .globl _breakinst

    _breakinst: ta 1
      ");
}

The equivalent code in .../gdb-5.0/gdb/i386-stub.c is:

#define BREAKPOINT() asm("   int $3");

/* This function will generate a breakpoint exception.  It is used at the
   beginning of a program to sync up with a debugger and can be used
   otherwise as a quick means to stop program execution and "break" into
   the debugger. */

void breakpoint()
{
  if (initialized)
    BREAKPOINT();
}

which kind of nicely ties in with the existing Linux code in
xcDll.cpp

Comment 2

[Axel Hecht]

This is awkward, I can get it to halt on the specified module, but I can't 
resume execution.
Doing a cont in gdb just finds the breakpoint again, and both next and step
hang.
What I found in the /usr/include/sys/signal.h on solaris is
#define SIGTRAP 5       /* trace trap (not reset when caught) */
This seems to be different on linux, at least I don't find a comment like this
there.
Rich, can you ping your channels?

Diff attached

Axel

Comment 3

[Axel Hecht]

Attachment: Adding SPARC Solaris hack to xcDll.cpp [broken]

Comment 4

[Rich Burridge]

Alex, couple of things:

1/ The breakpoint() function in the gdb source included a .globl label as
   well (see my previous comment). Have you tried including that and see
  what happens.

2/ When you create a patch, could you use the cvs diff -u format please, so
  that there is some context for the changes?

Thanks.

Comment 5

[Axel Hecht]

I just tried the full multiline statement. Other than an extra warning, I see
no difference. Tested both against gdb5.0 and 4.17 on Solaris 2.7

richb, do you need an extra diff for this? No wisdom there but cut&paste.

Axel

Comment 6

[Rich Burridge]

Axel, I'm not going to get to this any time soon (we have a PR2 release
coming up), so I'm not likely to be able to try anything for a few weeks.
I suggest getting a copy of the gdb 5.0 source code, compile it up on your
SPARC machine, and by adding debug messages (how do you debug a debugger ;-)
see if you can see how gdb solves this problem.

Comment 7

[leger]

Updating QA Contact.

Comment 8

[Hong Kwon]

changing qa contact to [e-mail]; please reassign as necessary.

Comment 9

[drapeau]

Setting QA contact to Sun QA lead.

Comment 10

[Rich Burridge]

Reassigning to Ed to do with as he wants.

Comment 11

[edburns]

I've asked Dauphin to take a look at this.