Frama-C Bug Tracking System - Frama-C
View Issue Details
0002312Frama-CPlug-in > wppublic2017-06-15 13:392017-06-15 15:14
Jochen 
correnson 
normaltweakalways
assignedopen 
Phosphorus-20170501-beta1xubuntu
Frama-C 15-Phosphorus 
 
0002312: false postcondition shouldn't be verified in default memory-model setting
Running "frama-c -wp -wp-rte memmodel_default.c" verifies all 11 proof obligations, while the assert clase in line 21 is obviously violated. The reason for Frama-C's behavior is that it assumes the "Hoare Variables mixed with Pointers" memory model as a default (in accordance with WP manual sect.3.4, p.45) without checking its preconditions, viz. the absence of any address-taking of a variable.

A novice user who doesn't yet know about the subtleties of memory models will assume after the above Frama-c run that the program is ok, as is has been formally verified. This may build up unjustified trust in the program, and discredit Frama-C (or even the whole field of formal methods) once the bug is detected at runtime, possibly causing severe damage.

I suggest to either
1. check the applicability of the "Hoare Variables mixed with Pointers" model (this should be easily achievable, if only the source code needs to be scanned for a unary "&"), or
2. use another, less restrictive, model as default.
No tags attached.
c memmodel_default.c (273) 2017-06-15 13:39
https://bts.frama-c.com/file_download.php?file_id=1192&type=bug
Issue History
2017-06-15 13:39JochenNew Issue
2017-06-15 13:39JochenStatusnew => assigned
2017-06-15 13:39JochenAssigned To => correnson
2017-06-15 13:39JochenFile Added: memmodel_default.c
2017-06-15 15:10JochenNote Added: 0006410

Notes
(0006410)
Jochen   
2017-06-15 15:10   
Running "frama-c -wp -wp-rte memmodel_default.c -wp-model ref" on the above program verifies all obligations, too.

This contradicts the WP manual Sect.3.5 "Hoare Variables for Reference Parameters", p.46, which says "necessary separation conditions are generated on-the-fly".

Apparently, no condition "requires \separated(&g,x);" is genarated for foo. If that condition is added manually, it can't be verified (as expected).