Bug #9534: Incorrect printing of Linux BUILD_BUG_ON - C Instrumentation Framework - Open-Source Projects

Actions

Copy link

Bug #9534

closed

Incorrect printing of Linux BUILD_BUG_ON

Added by Evgeny Novikov almost 6 years ago. Updated almost 6 years ago.

Status:

Rejected

Priority:

Urgent

Assignee:

Category:

C back-end

Start date:

03/13/2019

Due date:

% Done:

Estimated time:

Detected in build:

svn

Platform:

Published in build:

Description

At the moment C back-end prints:

void __kfree_rcu(struct rcu_head *head, unsigned long offset)
{
...
((void)sizeof(char[1 - 2*!!(!__builtin_constant_p(offset))]));
...
}

as:

void __kfree_rcu(struct rcu_head *head, long unsigned int offset)
{
  ...
  __builtin_constant_p ( offset ) == 0 ? -1 : 1, ( void ) ( long unsigned int ) ( __builtin_constant_p ( offset ) == 0 ? -1 : 1 );
  ...
}

This is definitely wrong.

Related issues 2 (0 open — 2 closed)

Actions

Copy link

Updated by Evgeny Novikov almost 6 years ago

Blocks Bug #6629: Used CIL is outdated added

Actions

Copy link

Updated by Evgeny Novikov almost 6 years ago

Blocks Feature #9536: Update CIF added

Actions

Copy link

Updated by Evgeny Novikov almost 6 years ago

Blocks deleted (Bug #6629: Used CIL is outdated)

Actions

Copy link

Updated by Evgeny Novikov almost 6 years ago

Related to Bug #9546: Frama-C (CIL) for GCC does not behave like GCC when casting ternary operators added

Actions

Copy link

Updated by Evgeny Novikov almost 6 years ago

Related to Bug #5407: Duplicated labels are generated added

Actions

Copy link

Updated by Evgeny Novikov almost 6 years ago

Indeed, the situation is not so bad, since there are no bugs in produced code. There is a bug in Frama-C (CIL) that can't parse it (#9536).

GCC calculates sizeofs internally, so, there are always some values and there are no sizeof expressions like in C. In this particular case the compiler could detect that there are no build bugs, i.e. "__builtin_constant_p ( offset ) 0" is always false at a compile time ("__builtin_constant_p(offset)" is always true) and we always have "sizeof(char[1])" that is 1. But there is still original expression calculating array size (it can be removed from the internal representation at all, but this does not happen).

I just dislike the outputted code since it includes "__builtin_constant_p ( offset ) 0 ? -1 : 1" 2 times that is not necessary at all. Perhaps, this causes #9534.

Actions

Copy link