[boost] assert II: a summary

From: Valentin Bonnard (Bonnard.V_at_[hidden])
Date: 1999-08-31 12:43:48

The debate about assert ran very quickly: I got answers 
to previous messages in my mailbox before I even had time 
to type the message 9there were more than 16 messages sent 
today about assertions !).

With so many contributions, things tend to get confusing 
for people who haven't followed all the debate with a 
lot of care.

So, let me summarize:

General comments:

- no one here wants to discourage the use assertions

- boost might provide an alternative to the standard 
  assert macro

The macro issue:

- in order to report file/line information, the 
  assertion has to be a macro

- macros are not good citizens, and are  
  often best avoided in favor of inline 
  function/constants/templates

The behaviour issue:

- throwing an exception (logic_error) is more C++-ish

- halting is harder to ignore

- producing a core dump helps debugging

The granularity issue:

- we would want to be able to turn assertion 
  on or off more precisely than per TU

- the TU granularity often breaks the ODR (see 
  below)

The ODR:

- the ODR requires that inline/template functions are 
  defined the same way in all TU

- it means that for assertions to be usable in 
  inline/template functions, then must be TU-independant

- for <cassert> it means that NDEBUG should get the 
  same value in all TUes

- easily enforcable policies are preferred, so

- being able to specify in exactly one place the 
  activation/desactivation of assertions is prefered

-- 
Valentin Bonnard