alt-ergo goals generated directly / via why3 differ in provability
ID0002353: This issue was created automatically from Mantis Issue 2353. Further discussion may take place here.
|ID0002353||Frama-C||Plug-in > wp||public||2018-02-01||2019-10-17|
|Reporter||Jochen||Assigned To||correnson||Resolution||no change required|
|Platform||Sulfur-20171101||OS||-||OS Version||Ubuntu 17.04|
|Product Version||Frama-C 16-Sulfur||Target Version||-||Fixed in Version||-|
I generated an Alt-Ergo proof obligation for the ensures clause "xxx" (line 1140) of the attached program "search_n_standalone.c" (which was obtained from "search_n" in "Acsl by example" by "gcc -E -C"). When I did this directly (see attached script "fd"), it was proven without problems. However, when I did this via Why3 (see script "fw"), Alt-Ergo responded "Don't know" after 0.1 seconds.
In order to compare both mlw files given to Alt-Ergo, I moved the goal of the via-Why3 file into the direct file; it became provable there. Removing as many axioms and definitions as possible while keeping the (non-)provability, I obtained the files "xxx_direct.mlw" and "xxx_via_why3.mlw". It seems that the former defines and uses "shift_sint32", while the latter does not (it uses "shift" instead). Probably, this is the reason for the different outcomes of Alt-Ergo.
Although the observed issue may not be bug (both mlw files may be semanically equivalent), it might be desirable to unify the proof obligations given to Alt-Ergo on different pathways.