Re: [PATCH] Fix NO_LIBPCRE1_JIT to fully disable JIT

From: Charles Bailey <charles@hashpling.org>
Date: Nov 13, 2017, 06:54 UTC
Message-ID: <20171113065410.rb43utcbncy7ndrv@hashpling.org>
In-Reply-To: <xmqqmv3qal78.fsf@gitster.mtv.corp.google.com>

On Mon, Nov 13, 2017 at 12:53:15PM +0900, Junio C Hamano wrote:

> 
> Thanks.  This patch needs a sign-off, by the way.

Signed-off-by: cbailey32@bloomberg.net

(I can resend the full patch if required or if anyone requests futher changes.

Show 6 quoted lines

> > But that we should take it anyway regardless of that since it'll *also*
> > work on Linux with your patch, and this logic makes some sense whereas
> > the other one clearly didn't and just worked by pure accident of some
> > toolchain semantics that I haven't figured out yet.
> 
> That is curious and would be nice to know the answer to.

The error that I was getting - if I remember the details of the very brief debugging session that I performed - was an unaligned memory access causing a SIGBUS in PCRE code whose function name contained 'jit' and which was being called indirectly from pcre_study.

My guess is that we are just exposing a pre-existing bug in our Solaris build of libpcre. Unaligned memory accesses on x86 / x86_64 "only" cause performance issues rather than fatal signals so even if the same bug exists on Linux it probably has no noticeable effect (or at least no noticed effect).

Charles.