Re: LLVM 18 - Mailing list pgsql-hackers

From Thomas Munro
Subject Re: LLVM 18
Date
Msg-id CA+hUKGKv5i1AMGcPB0Yov76UVuGJsNDkQEUHZoC6=7g6cAhDFA@mail.gmail.com
Whole thread Raw
In response to LLVM 18  (Thomas Munro <thomas.munro@gmail.com>)
Responses Re: LLVM 18
List pgsql-hackers
On Wed, Jan 3, 2024 at 6:04 PM Thomas Munro <thomas.munro@gmail.com> wrote:
> LLVM 16 provided a new function name[1], and LLVM 18 (not shipped yet)
> has started complaining[2] about the old spelling.
>
> Here's a patch.

And pushed.

Just in case anyone else is confused by this, be aware that they've
changed their numbering scheme.  The 18.1 schedule visible on llvm.org
doesn't imply that 18.0 has already shipped, it's just that they've
decided to start at X.1.

By the way, while testing on my Debian system with apt.llvm.org
packages, I discovered that we crash with its latest llvm-18 package,
namely:

llvm-18_1%3a18~++20240122112312+ad01447d30ed-1~exp1~20240122112329.478_amd64.deb

Program terminated with signal SIGSEGV, Segmentation fault.
#0  0x00007f033e73f5f8 in llvm::InlineFunction(llvm::CallBase&,
llvm::InlineFunctionInfo&, bool, llvm::AAResults*, bool,
llvm::Function*) () from /lib/x86_64-linux-gnu/libLLVM-18.so.1

... so I re-confirmed that I wasn't hallucinating and it did work
before I disappeared for the holidays by downgrading to the one before
that from my /var/cache/apt/archives, namely:

llvm-18_1%3a18~++20231218112348+a4deb14e353c-1~exp1~20231218112405.407_amd64.deb

So I built the tip of their release/18.x branch so I could try to get
some more information out of my debugger and perhaps their assertions,
but it worked.  So I have to assume that something was broken at their
commit ad01447d30ed and has been fixed in the past few days, but I
didn't have time to dig further, and will re-check a bit later when a
fresh package shows up.



pgsql-hackers by date:

Previous
From: James Coleman
Date:
Subject: Re: Teach predtest about IS [NOT] proofs
Next
From: Peter Smith
Date:
Subject: Re: Synchronizing slots from primary to standby