Horton

Re: Remove last traces of HPPA support

Andres Freund <andres@anarazel.de>

View thread View on postgr.es
From: Andres Freund <andres@anarazel.de>
To: Thomas Munro <thomas.munro@gmail.com>
Cc: Tom Lane <tgl@sss.pgh.pa.us>, Noah Misch <noah@leadboat.com>, pgsql-hackers@lists.postgresql.org
Date: 2024-07-31T19:07:30Z
Lists: pgsql-hackers 
Hi,

On 2024-07-31 17:52:34 +1200, Thomas Munro wrote:
> 2.  The pg_atomic_unlocked_test_flag() function was surprising to me:
> it returns true if it's not currently set (according to a relaxed
> load).  Most of this patch was easy, but figuring out that I had
> reverse polarity here was a multi-coffee operation :-)  I can't call
> it wrong though, as it's not based on <stdatomic.h>, and it's clearly
> documented, so *shrug*.

I have no idea why I did it that way round. This was a long time ago...


> 4.  Another reason would be that you need it to implement
> SpinLockFree() and S_LOCK_FREE().  They don't seem to have had any
> real callers since the beginning of open source PostgreSQL!, except
> for a test of limited value in a new world without ports developing
> their own spinlock code.  Let's remove them!  I see this was already
> threatened by Andres in 3b37a6de.

Note that I would like to add a user for S_LOCK_FREE(), to detect repeated
SpinLockRelease():
https://postgr.es/m/20240729182952.hua325647e2ggbsy%40awork3.anarazel.de

Greetings,

Andres Freund

Commits

  1. Require memory barrier support.

  2. Require compiler barrier support.

  3. Remove --disable-atomics, require 32 bit atomics.

  4. Remove --disable-spinlocks.

  5. Remove support for HPPA (a/k/a PA-RISC) architecture.

  6. Add a basic atomic ops API abstracting away platform/architecture details.