--- layout: fc_discuss_archives title: Message 34 from Frama-C-discuss on February 2014 ---
> I wonder if you could comment under what circumstances I should use > this --max-split option. > Unfortunately, -max-spit is just a heuristic. We are investigating the improvement of our "case split analysis" method described in [1] to get rid of this "frustrating" limitation. - Mohamed. > *From:*frama-c-discuss-bounces at lists.gforge.inria.fr > [mailto:frama-c-discuss-bounces at lists.gforge.inria.fr] *On Behalf Of > *Mohamed Iguernelala > *Sent:* Monday, February 17, 2014 1:42 AM > *To:* frama-c-discuss at lists.gforge.inria.fr > *Subject:* Re: [Frama-c-discuss] Frama-c: WP issues with Alt-Ergo (but > works with Z3) > > Hello, > > > Is there any possibility to make this program verifiable with Alt-Ergo? > > Yes ... with option *-max-split -1* > > More precisely, the decision procedure of linear integer arithmetic in > Alt-Ergo uses bounds inference and interval calculus [1]. > It is based on the following claims: > (a) If all the affine forms of the problem are unbounded, we are sure > that this problem has a solution. > (b) If some affine form has a lower (integer) bound *L* and an upper > bound *U*, Alt-Ergo may try all the possible integer values in [| L , > U|] to see if there exists a solution that satisfies these bounds. > > In practice, when the intervals *[| L , U|]* are huge, Alt-Ergo may > spend a lot of time in (b), which makes it "unfair". So, we put a > "max-split" limit in this part of Alt-Ergo to prevent it from > "diverging". The option -max-split -1 deactivates this limit. > > [1] A Simplex-Based Extension of Fourier-Motzkin for Solving Linear > Integer Arithmetic: F. Bobot, S. Conchon, E. Contejean, M. > Iguernelala, A. Mahboubi, A. Mebsout, G. Melquiond > > - Mohamed. > > ---- > Senior R&D Engineer, OCamlPro > Research Associate, VALS team, LRI. > http://www.iguer.info <http://cp.mcafee.com/d/FZsS921J5AQsFK8CzBBZATsSDtV5VxxZ5cSDtV5VxxZNASDtV5VxxZ5wSDtVdOX2rOpJ0zIfFI0kXoKGxPqG7uwSrtInlgVJl3LgrdzDTthKO_R-hK-qekQT-LsKCPtDAmjhOPtBzBHEShhlKYzOEuvkzaT0QSyrjdTV5xdVYsyMCqejtPo0ah5LUZa3JjrItlQLoKxaBCWkbAaJMJZ0kvaAWsht00_QEhBLeNdEI6zBdYS2_id41Fr5UWHFuNt2lbdQEqnjh0bNRniZAzh00Emd44Gkr-xEwciCjd40wvcKdh8Rd41EE8dPd41yIiJ3h0oBcCq809pd59IsMryffg> > > > > Le 16/02/2014 22:01, Dharmalingam Ganesan a ?crit : > > Hi, > > > > For the attached program, all the properties were proved by Z3 but Alt-Ergo timeout/Unknown for loop invariants. > > > > If I replace UINT_MAX by a small constant, say 100, then Alt-Ergo was also able to prove if I substitute UINT_MAX by 1000. > > > > Is there any possibility to make this program verifiable with Alt-Ergo? > > > > Here is my log for Z3 and Alt-Ergo: > > > > > > [formal_verification]$ frama-c -wp -wp-rte -pp-annot -wp-timeout 25 -wp-proof=z3 c_nullify.c > > [kernel] preprocessing with "gcc -C -E -I. -dD c_nullify.c" > > [wp] Running WP plugin... > > [wp] Collecting axiomatic usage > > [rte] annotating function C_nullify > > [rte] annotating function main > > [wp] 13 goals scheduled > > [wp] [Qed] Goal typed_C_nullify_loop_inv_established : Valid > > [wp] [Qed] Goal typed_C_nullify_loop_assign_part1 : Valid > > [wp] [Z3] Goal typed_C_nullify_assert_rte_unsigned_overflow : Valid (10ms) > > [wp] [Z3] Goal typed_C_nullify_assert_rte_mem_access : Valid (10ms) > > [wp] [Z3] Goal typed_C_nullify_loop_inv_preserved : Valid (10ms) > > [wp] [Z3] Goal typed_C_nullify_loop_assign_part2 : Valid (10ms) > > [wp] [Qed] Goal typed_C_nullify_assign_part1 : Valid > > [wp] [Qed] Goal typed_C_nullify_loop_term_decrease : Valid > > [wp] [Qed] Goal typed_C_nullify_loop_term_positive : Valid > > [wp] [Qed] Goal typed_main_call_C_nullify_pre : Valid > > [wp] [Qed] Goal typed_main_call_C_nullify_pre_2 : Valid > > [wp] [Z3] Goal typed_C_nullify_assign_part2 : Valid > > [wp] [Z3] Goal typed_C_nullify_loop_assign_part3 : Valid (20ms) > > > > > > > > [formal_verification]$ frama-c -wp -wp-rte -pp-annot -wp-timeout 100 -wp-proof=alt-ergo c_nullify.c > > [kernel] preprocessing with "gcc -C -E -I. -dD c_nullify.c" > > [wp] Running WP plugin... > > [wp] Collecting axiomatic usage > > [rte] annotating function C_nullify > > [rte] annotating function main > > [wp] 13 goals scheduled > > [wp] [Qed] Goal typed_C_nullify_loop_inv_established : Valid > > [wp] [Qed] Goal typed_C_nullify_loop_assign_part1 : Valid > > [wp] [Alt-Ergo] Goal typed_C_nullify_assert_rte_unsigned_overflow : Valid (13) > > [wp] [Alt-Ergo] Goal typed_C_nullify_assert_rte_mem_access : Valid (20ms) (22) > > [wp] [Qed] Goal typed_C_nullify_assign_part1 : Valid > > [wp] [Alt-Ergo] Goal typed_C_nullify_loop_assign_part2 : Valid (20ms) (23) > > [wp] [Qed] Goal typed_C_nullify_loop_term_decrease : Valid > > [wp] [Alt-Ergo] Goal typed_C_nullify_loop_assign_part3 : Valid (Qed:4ms) (24ms) (31) > > [wp] [Qed] Goal typed_C_nullify_loop_term_positive : Valid > > [wp] [Alt-Ergo] Goal typed_C_nullify_assign_part2 : Valid (16ms) (16) > > [wp] [Qed] Goal typed_main_call_C_nullify_pre : Valid > > [wp] [Qed] Goal typed_main_call_C_nullify_pre_2 : Valid > > [wp] [Alt-Ergo] Goal typed_C_nullify_loop_inv_preserved : Unknown (Qed:4ms) (12s) > > > > > > > > Thanks, > > Dharma > > > > > _______________________________________________ > > Frama-c-discuss mailing list > > Frama-c-discuss at lists.gforge.inria.fr <mailto:Frama-c-discuss at lists.gforge.inria.fr> > > http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/frama-c-discuss <http://cp.mcafee.com/d/avndz8w939J5AQsFK8CzBBZATsSDtV5VxxZ5cSDtV5VxxZNASDtV5VxxZ5wSDtVdOX2rOpJ0zIfFI0kXoKGxPqG7uwSrtInlgVJl3LgrdzDTthKO_R-hK-qekQT-LsKCPtDAmjhOPtBzBHEShhlKYzOEuvkzaT0QSCrjdTV5xdVYsyMCqejtPo0c-l9QUyW01_Fgzbutzw25O4Eo9GX33VkDa3JstzGKBX5Q9kITixJSeGWnIngBiPta5O5mUm-wafBite8Kw0vWk8OTDoCQm3hOC-r1vF6y0QJyYtlQLoKxaBCWkdbFEw5UWHFuOhEw0kb6y2lad_gQg69j9Cy0gfCn6EAqCy0Qk46VCy0Nm9mxEwciCjd404ICyASeodCp5eP5cM> > -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.gforge.inria.fr/pipermail/frama-c-discuss/attachments/20140217/aec93f00/attachment-0001.html>