Boost logo

Boost :

From: Marcin Kalicinski (kalita_at_[hidden])
Date: 2005-09-23 13:13:03


Hi Everybody,

I see there are mixed opinions on use of __assume in BOOST_ASSERT. The chief
concern is there are people using BOOST_ASSERT in "uncommon" (but still
valid) ways, such as:

- to show dialog box, but only in debug mode
- using assert handler to throw during unit testing
- possibly more

These types of usage would be prohibited if we introduced __assume into
BOOST_ASSERT, so basically this is a no-no from stability point of view
(nobody wants his code broken). Btw. the second way can be dealt with with
disabling __assume if assertion handler is defined.

On the other hand some people (including me) consider __assume to be nice,
helping optimization and suppressing annoying warnings. I see 3 options:

1. BOOST_ASSUME
===============================================

Put __assume in BOOST_ASSUME and leave BOOST_ASSERT unmodified.

Pros:
- does not disturb existing code in any way

Cons:
- no existing code uses it, i.e. all existing libraries would have to be
modified to take advantage of it; this will probably never happen so most of
the code will not benefit
- obscure: most of the people would be confused about it when seeing it for
the first time

2. BOOST_ASSERT + define to enable assume (default: on)
===============================================

Pros:
- does enable assume on all the boost code, some user programs get somewhat
smaller and faster without any intervention from the programmer

Cons:
- breaks user code which uses BOOST_ASSERT in "uncommon" way. Would require
a change in docs that executing BOOST_ASSERT(false) is undefined behavior
(i.e. different than std::assert)

BOOST_ASSERT + define to enable assume (default: off)
===============================================

Pros:
- does enable assume on all the boost code, some user programs get smaller
and faster with a simple -D"BOOST_ASSERT_USE_ASSUME" on command line.

Cons:
- only small number of users will know about that #define, so probably most
of the code will not benefit.

What do you think?

Marcin


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