disabling jemalloc on bitrig

[dhuseby]

it turns out that jemalloc doesn't behave well on bitrig. with jemalloc enabled i get some kernel errors related to sbrk failures. with jemalloc disabled, the errors go away. i am investigating, but in the mean time, we should just disable jemalloc by default on bitrig.

[rust-highfive]

r? @nikomatsakis

(rust_highfive has picked a reviewer for you, use r? to override)

[nikomatsakis]

@bors r+ 074f094 rollup

[dhuseby]

While compiling Rust from source on Bitrig using a Rust snapshot built with jemalloc enabled, I get a lot of errors that look like this:

sbrk: grow 12850741248 failed, error = 12
sbrk: grow 8743018496 failed, error = 12
sbrk: grow 12560965632 failed, error = 12
sbrk: grow 12030480384 failed, error = 12
sbrk: grow 9737023488 failed, error = 12
sbrk: grow 11791466496 failed, error = 12
sbrk: grow 11408199680 failed, error = 12
sbrk: grow 10830114816 failed, error = 12
sbrk: grow 10409455616 failed, error = 12
sbrk: grow 9631027200 failed, error = 12
sbrk: grow 10049781760 failed, error = 12

It doesn't seem to affect Rust functionality but it sure looks scary. Like I said, I'm disabling jemalloc on Bitrig until I know what is causing this and I upstream the patches to jemalloc.

[semarie]

@dhuseby I don't known if it is related, but under openbsd I use the following patch https://github.com/jasperla/openbsd-wip/blob/master/lang/rustc/patches/patch-src_jemalloc_configure for building jemalloc.

The latest version of jemalloc proper detect the lake of tls (thread-local-storage), but not the version used by rust.

I hope this could help you.

[dhuseby]

@semarie that might fix the problem. i'll give it a try soon. i did notice that without jemalloc, several of my test apps ran noticeably (wall clock) faster. But that may have been because jemalloc was broken.

[dhuseby]

@nikomatsakis sorry for confusing things a bit, a new snapshot for bitrig was just uploaded and i added it to the snapshots.txt file and bumped llvm-auto-clean-trigger too.

[dhuseby]

@nikomatsakis both disabling jemalloc and the new snapshot are needed to get the bitrig buildbot working properly.

[alexcrichton]

You should be able to leave out the llvm-auto-clean-trigger modification as this PR didn't change LLVM itself (so it shouldn't need to be rebuilt)

[dhuseby]

@alexcrichton aha! ok, I see.

[dhuseby]

OK, I reverted the bump in llvm-auto-clean-trigger and rebased to the current master HEAD.

[alexcrichton]

Could you peel off the last two commits as well? Otherwise r=me, thanks!

[dhuseby]

@alexcrichton sure

[alexcrichton]

@bors: r+ c005f8a2f1022679ab3fb4353e12cc795f04a9a0

[semarie]

the name is invalid: it should be bitrig (missing t)

[dhuseby]

@semarie good catch! corrected.

[dhuseby]

rebased to current master HEAD.

[alexcrichton]

@bors: r+ 33751ff