31287 - Unable to build with --enable-boehm on Linux: mulitple definition errors in gc_cpp.cc

Status: RESOLVED FIXED

(SeaMonkey :: Build Config, defect, P3)

Description

[hacksaw]

* Overview Description:

Compiling with the --enable-boehm option on gcc 2.95.2 reports 3 seperate

"multiple definition" errors in "gc_cpp.cc".



* Steps to Reproduce:

1) compile with ----enable-boehm on Linux



* Actual Results:

See attached file for a snippet of (g)make's output.  In short, the following

sybols were apparently redefined: GC_push_all_stacks, GC_stop_world, and

GC_start_world.



Expected Results:

A complete build resulting in a browser that never leaks memory and saves the

world.  ;-)



* Reproducibility:

I'm currently only able to test this on my Debian (Potato) GNU/Linux (2.2.13)

machine.



* Build Date & Platform Bug Found:

Mar. 7-9 CVS HEAD



* Additional Information:

The problem seems to be in generic_threads.c vs.  linux_threads.c:  both contain

different definitions of GC_stop_world, GC_start_world etc.

Comment 1

[hacksaw]

Attachment: make output snippet.

Comment 2

[cls]

Ok, I was able to get gc/boehm to build but viewer & mozilla segfaulted in
xpcom::nsLeakDetector.  I played with it some more and was able to get all of
the components registered before it segfaulted.  Looking at
gc/boehm/README.linux, I'm wondering if there's not a less invasive gc we could
use.  For using boehm with threads, it appears to expect you to include "gc.h"
in every file that makes a thread call.  Or is that only necessary if you're
doing some sort of gc on threads?  I'm guessing that would include NSPR as well.

Beard, any suggestions?

Comment 3

[cls]

Attachment: Build changes to gc/boehm

Comment 4

[cls]

Attachment: Build changes outside of gc/boehm

Comment 5

[cls]

The patches have been checked in but if I understand correctly, there are still
some changes that need to be made to at least NSPR for this option to work
correctly.

Comment 6

[hacksaw]

> The patches have been checked in [...]

Great work!  I have it running now, though for  some strange reason I can't

start a second instance or restart after exiting (I had another build do this

earlier today).

Comment 7

[hacksaw]

>I have it running now [...]
Just an update--it's now back to segfaulting as cls described.  Is there any
particular aspect of the NSPR stuff I could help look into?

Comment 8

[cls]

The build option compiles again but some changes remain to allow gc work with
nspr on linux. Reassigning to beard as he's the one actually doing the
integration.

Comment 9

[Chris Yeh]

mass re-assign of all bugs where i was listed as the qa contact

Comment 10

[Patrick C. Beard]

Working on it...

Comment 11

[Patrick C. Beard]

I have changes in my tree for this.

Comment 12

[Patrick C. Beard]

Wade Hennessey is working on this, and it's almost ready. Nominating for beta3 
checkin. This code will go in under #ifdef, so it's safe.

Comment 13

[Patrick C. Beard]

I think we can close this bug now, given wade's recent checkins.