Boost logo

Boost :

Subject: Re: [boost] [Christian S] Spirit and ANTLR - Request for example
From: Larry Evans (cppljevans_at_[hidden])
Date: 2009-06-09 14:25:57


On 06/09/09 11:46, Christian Schladetsch wrote:
> This exactly corelates to my 1,2,3 steps of Spirite usage.
> But somehow I am the Demon?
[snip]

I think you have some good points, Christian, that you made here:

   http://article.gmane.org/gmane.comp.lib.boost.devel/190361

The slow compile times, IIRC, have been mentioned frequently. The
cryptic error messages have also been mentioned before, but they are
not spirit specific. They are a problem with template metaprogramming
and expression templates debugging in general, as acknowledged near
the beginning of Chap. 8 of:

   http://www.amazon.com/exec/obidos/ASIN/0321227255/billvennersA/

So, the problem, as you noted earlier, but slightly modified, is the
C++ *compiler* is not a *fast* language *development* tool.

However, the one advantage of spirit (that I can think of at the
moment) is that the parser produced *parses faster* than one produced
with antlr or yacc because of the compile-time optimization possible
with expression templates. OK, I'm just guessing here, and I have no
benchmarks to show that. Maybe the spirit folks do?

Now I can see your justification for abandoning spirit because the
slow compile times and cryptic error messages slow down the developer.
On the other hand, the faster parse times saves the end-user time. So
it's a balance between whether one thinks the parse time, which is
experienced *many* times by the end user, is more important than the
developer time, which is experienced probably *fewer* times than the
parse times.

Is that about right?

Anyway, although I think you may have some good points, I also think
you could get more useful responses with more diplomacy.

-Sincerely,
Larry


Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk