[mercury-users] .par grades of Mercury: are they safe?

Ralph Becket rafe at cs.mu.OZ.AU
Mon Nov 22 12:04:14 AEDT 2004

Dominique de Waleffe, Sunday, 21 November 2004:
> I spotted the following paragraphs in the NEWS file of the current ROTD.
> >>>
> Changes to the Mercury compiler:
> * Due to personnel changes and semantic problems in the design of
>  POSIX threads and problems in the implementations thereof that we have 
>  tried,
>  we cannot guarantee that we will be able to fix low-level concurrency bugs 
>  in
>  .par grades.
> <<<
> That makes me worried. 
> We are using the hlc.par.gc grade for a big application for a 
> commercial customer (objective is 300 concurrent users) and I would 
> like to know what are those bugs being talked about and 
> in which kind of use of threads in Mercury we would be likely to
> encounter or be hit by those bugs.
> Can anyone comment?

Apart from the program termination "feature" that Fergus pointed out, we
are unaware of any bugs with the .par grades.  The message in the ROTD
NEWS file, which I need to amend in the light of recent discussion (*),
was prompted by the amount of effort it took to find a fix for the
hlc.par.gc grades which weren't working in the development version of
the compiler (all is well now!).  We do not plan on dropping support for
.par grades.  What I was trying to say in the NEWS file is that IF a
user discovers a bug with .par grades, THEN it may take us longer to
find a fix, ALSO that it MAY be the case that some bugs we can't fix.
That said, Fergus understands that code pretty well and we guess he'll
be helping us out from Seattle one way or another for the forseeable

I hope this puts your mind at rest,
-- Ralph

(*) It's not clear exactly where we should say this, if at all.
mercury-users mailing list
post:  mercury-users at cs.mu.oz.au
administrative address: owner-mercury-users at cs.mu.oz.au
unsubscribe: Address: mercury-users-request at cs.mu.oz.au Message: unsubscribe
subscribe:   Address: mercury-users-request at cs.mu.oz.au Message: subscribe

More information about the users mailing list