[2] [gclist] GC & profiling

Dave Berry daveb@tardis.ed.ac.uk
Mon, 23 Apr 2001 22:08:33 +0100


At 04:13 24/04/2001, Fergus Henderson wrote:
>It's nice to know from Nick Barnes' post that the Harlequin MLWorks
>profiler got this right.  But since Harlequin went out of business,
>I'm not sure what that tells us :-(  Perhaps the only thing to be learnt
>from this is another one of those depressing worse-is-better lessons:
>maybe the problem is just not sufficiently important in practice to
>be worth worrying about.

I disagree with that.  The main reason that MLWorks failed was the classic
one: lack of business management.  The MD had faith in ML, but the
organisation as a whole did not define a market, nor a strategy for
reaching that market.  It's quite possible that no such market exist[s/ed],
although some of us did have ideas that would have been worth pursuing.
The technology itself had good points and bad points.  Many of the bad
points followed from the lack of clear direction.  

The profiler was definitely one of the good parts of the system.  We found
it invaluable in improving the performance of the system, and of customers'
programs.  The only downside was that it increased the work involved in
porting the system, as the profiler had to be aware of the stack layout on
each architecture.

In addition to the time and heap profilers that Nick mentioned, MLWorks
also had a call counting option.  We tended not to use it so much, possibly
because it required recompiling the code and was therefore more of a hassle
to use. 

By the way, Harlequin is still in business, as two firms.  Harlequin still
produce their postscript interpreter and related add-ons for the high-end
printer marker.  Xanalys [eXpert ANALYSts] are still active with Lisp, and
of course memory management -- your cue, Pekka?

Dave.

Dave.