charm AT lists.siebelschool.illinois.edu

Subject: Charm++ parallel programming system

List archive

Re: [charm] vector< bool >

From: Sam White <white67 AT illinois.edu>
To: Jozsef Bakosi <jbakosi AT lanl.gov>
Cc: "charm AT lists.cs.illinois.edu" <charm AT lists.cs.illinois.edu>, "Pandare, Aditya Kiran" <apandare AT lanl.gov>
Subject: Re: [charm] vector< bool >
Date: Thu, 12 Dec 2019 10:41:57 -0600
Authentication-results: illinois.edu; spf=pass smtp.mailfrom=samt.white AT gmail.com; dkim=pass header.d=illinois-edu.20150623.gappssmtp.com header.s=20150623; dmarc=none header.from=illinois.edu

Hi Jozsef,

Thanks for pointing this out. We already had a specialization in place for vector<bool> in pup_stl.h, but it was buggy.
Here's a fix: https://github.com/UIUC-PPL/charm/pull/2650

-Sam

On Thu, Dec 12, 2019 at 9:53 AM Jozsef Bakosi <jbakosi AT lanl.gov> wrote:

Hi folks,

We have recently run into a very nasty problem that was pretty hard to
diagnose: serializing a std::vector< bool > corrupts its contents and
will yield dissimilar data on different PEs. We fixed the problem by
using std::vector< int > instead, which serializes correctly.

The problem with std::vector< bool > is somewhat well-known, see e.g.,
https://stackoverflow.com/questions/17794569/why-is-vectorbool-not-a-stl-container/17794965#17794965,
and this is probably the cause of the erroneous behavior with the
serialization machinery.

It would be very user-friendly if Charm++ handled this special case
correctly.

Thanks,
Jozsef

[charm] vector< bool >, Jozsef Bakosi, 12/12/2019
- <Possible follow-up(s)>
- Re: [charm] vector< bool >, Sam White, 12/12/2019
  - Re: [charm] vector< bool >, Jozsef Bakosi, 12/12/2019