[m-rev.] for review: dumping trace counts

[Julien Fischer]

On Mon, 25 Sep 2006, Ian MacLarty wrote:

> On Mon, Sep 25, 2006 at 05:23:41PM +1000, Julien Fischer wrote:
>>
>> On Mon, 25 Sep 2006, Ian MacLarty wrote:
>>
>>> On Mon, Sep 25, 2006 at 03:58:05PM +1000, Julien Fischer wrote:
>>>>
>>>> On Mon, 25 Sep 2006, Zoltan Somogyi wrote:
>>>>
>>>>> For review by anyone.
>>>>>
>>>>> Question for reviewers: should the dump_trace_counts_to predicate be
>>>>> moved
>>>>> to a library module to make it available to other programs, and if so,
>>>>> which
>>>>> one and with what name?
>>>>
>>>> Yes, the io module and io.write_trace_counts. (or alternatively you could
>>>> add an impure version to benchmarking.)
>>>>
>>>
>>> Note that the preds in benchmarking can be made pure now that we have
>>> trace goals (by adding io state args).
>>>
>>> I'd therefore say it should be benchmarking.write_trace_counts and be
>>> pure (with io state args).
>>
>> Note that by doing so you've just introduced a cyclical
>> dependency between the benchmarking and io modules (if there wasn't
>> one already ... the library module dependency graph is fairly complicated.)
>>
>
> Is there a problem with that?
>

Only in that introduces more complexity.  (It also makes intermodule
optimization a bit more fiddly.)

(I've just spent a couple of minutes trying to simplify the library
dependency graph - the breakup of std_util a couple of months ago
certainly helped - we should probably consider breaking up the io
module at some point, but how ...?)

Julien.
--------------------------------------------------------------------------
mercury-reviews mailing list
Post messages to:       mercury-reviews at csse.unimelb.edu.au
Administrative Queries: owner-mercury-reviews at csse.unimelb.edu.au
Subscriptions:          mercury-reviews-request at csse.unimelb.edu.au
--------------------------------------------------------------------------