rustc fails with memory corruption when --disable-jemalloc is used

[mneumann]

If I build rust with --disable-jemalloc, I am getting a strange memory issue somewhere in stage1 (see below for how to reproduce it on Linux). I also reproduced the same issue during porting rust to DragonflyBSD (which I initially built w/o jemalloc). There I noticed for some input files that it wants to mmap an anonymous region of size 0x800_000_000 or 0x1000_000_000, probably due to a mal-formed malloc call (just a wild guess). And then more or less stalls the system, probably trying to zero that region and swapping to disk. Changing a single letter in the file subject for compilation makes rustc behave well again. With jemalloc, these issues are not existent (both on Dragonfly and Linux)!

# uname -a
Linux linux-vbox 3.11.0-12-generic #19-Ubuntu SMP Wed Oct 9 16:20:46 UTC 2013 x86_64 x86_64 x86_64 GNU/Linux
# ./configure --disable-jemalloc
# make

results in:

cp: x86_64-unknown-linux-gnu/stage1/lib/librustc_llvm
cp: x86_64-unknown-linux-gnu/stage1/lib/librustc_back
cp: x86_64-unknown-linux-gnu/stage1/lib/librustc
cp: x86_64-unknown-linux-gnu/stage1/lib/libnative
cp: x86_64-unknown-linux-gnu/stage1/bin/rustc
cp: x86_64-unknown-linux-gnu/stage1/lib/rustlib/x86_64-unknown-linux-gnu/lib/libmorestack.a
cp: x86_64-unknown-linux-gnu/stage1/lib/rustlib/x86_64-unknown-linux-gnu/lib/libcompiler-rt.a
rustc: x86_64-unknown-linux-gnu/stage1/lib/rustlib/x86_64-unknown-linux-gnu/lib/libcore
*** Error in `x86_64-unknown-linux-gnu/stage1/bin/rustc': free(): invalid next size (fast): 0x00002b032802af30 ***
*** Error in `x86_64-unknown-linux-gnu/stage1/bin/rustc': malloc(): memory corruption: 0x00002b032802af60 ***

[alexcrichton]

if you run valgrind, does it turn up anything suspicious?

[mneumann]

hm, is it only me seeing this failure? I thought it is somehow reproducable everywhere :). I haven't tried valgrind yet.

[alexcrichton]

Ah I was just hoping you still had a build lying around, I hadn't personally tried it out just yet, but I shall do so now!

[thestinger]

@mneumann: Is this still an issue with master?

[thestinger]

AFAIK this has been fixed. I can't duplicate it on Linux or Windows.