GNU bug report logs - #64504
[PATCH] gnu: Add unsio.

Previous Next

Full log

Message #14 received at 64504 <at> debbugs.gnu.org (full text, mbox):

From: Andreas Enge <andreas <at> enge.fr>
To: Iliya Tikhonenko <tikhonenko <at> mpe.mpg.de>
Cc: 64504 <at> debbugs.gnu.org, Sharlatan Hellseher <sharlatanus <at> gmail.com>,
 Efraim Flashner <efraim <at> flashner.co.il>, Eric Bavier <bavier <at> posteo.net>
Subject: Re: [bug#64504] [PATCH] gnu: Add unsio.
Date: Tue, 8 Aug 2023 17:49:50 +0200

[Message part 1 (text/plain, inline)]

Am Tue, Aug 08, 2023 at 01:37:14PM +0200 schrieb Iliya Tikhonenko:
> On 8/8/23 11:29, Andreas Enge wrote:
> > For the #:build-type, the default value (release with debug information) also works and results in a package of the same size. Is there a reason to change it? Modified patch attached.
> The debug build type is unnecessary verbose (see src/uns.cc line 71-73 for
> an example) to the point that it impacts performance sometimes, which is
> important considering the usecase of the library.

Okay, I have reinstated your line and added a comment.
 
> > I can also compile the package without the configure flags, are they useful?
> Checked again, I still cannot compile it without -DCMAKE_CXX_STANDARD=14

I can... Patch attached.

> > And there are lots of warnings about 34 bit shifts in a 32 bit type.
> > But I suppose you tested that the library works.
> I have not noticed any issues using it during the past two years, but this
> cannot be considered a proper test because I am actually using a small
> subset of the library. Perhaps, this warning should be reported upstream,
> thanks!

/tmp/guix-build-unsio-1.3.3-0.25e5246.drv-0/source/src/snapshotinterface.cc:125:45: warning: result of ‘(-2147483648 << 2)’ requires 34 bits to represent, but ‘int’ only has 32 bits [-Wshift-overflow=]
  125 |       req_bits = (unsigned int) (( 1 << 31 )<<2)-1;
      |                                 ~~~~~~~~~~~~^~~~

Here is the offending line. This is indeed a bit surprising, as "int"
usually has 32 bits. The 1 to be shifted is a signed int; 1*2^31 is not
representable in the type, so if I understand the C standard correctly,
the result is undefined. Then I have seen a Stackoverflow comment that
it is okay in C++20. And indeed:
   #:configure-flags #~(list "-DCMAKE_CXX_STANDARD=20")
passes without these warnings. Should we add this?
Although such constructs look a bit brittle to me, and it is not
suggested by the author.

There is still the question about the lib64/ installation directory,
I think this should be changed.

Andreas

[0001-gnu-Add-unsio.patch (text/plain, attachment)]

Previous Next

GNU bug tracking system
Copyright (C) 1999 Darren O. Benham, 1997,2003 nCipher Corporation Ltd, 1994-97 Ian Jackson.