Hello David,I have tried to apply the same patch to both Emacs22 and Emacs23 but the packages still fail to build on the bytecomp stage.
The patch I've used for Emacs22 is attached.NetBSD 5.99.56 NetBSD 5.99.56 (GENERIC) #2: Mon Oct 17 04:31:52 MSK 2011 dm@vault:/usr/obj/sys/arch/amd64/compile/GENERIC x86_64
Dmitry On Fri, 2 Dec 2011 02:57:09 +0000, David Holland wrote:
So, this behavior (which has been plaguing all the emacs packages on and off) popped up in emacs20 on current with gcc 4.5, and last night I traced it to the compiler optimizing out the changes to __malloc_hook in alloc.c. Adding [the expansion of] __insn_barrier() in between the assignments made the crash go away. A casual inspection suggests that this problem should also affect emacs21 and emacs22 -- that is, the somewhat regrettable structure of malloc calls is unchanged and there's nothing I can see that would lead the compiler to make a different decision from what it did on emacs20. (I didn't look at emacs23 or emacs24 last night, but they're probably either similar or have an overt fix or have completely different code.) Is the behavior we've been seeing actually consistent with it being a gcc 4.5 issue? (What I've seen myself has been. But of course there might be multiple problems with the same symptom.) Can someone(s) who's been seeing this problem building other emacs versions (I haven't been so far) try merging my emacs20 changes and seeing if it helps? The changes are the bottom three hunks of emacs20's patch-bm.
Attachment:
patch-src_alloc.c
Description: Binary data