[Bug analyzer/101503] [12 Regression] ICE in involves_p, at analyzer/svalue.cc:557

["cvs-commit at gcc dot gnu.org" <gcc-bugzilla@gcc.gnu.org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101503

--- Comment #2 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by David Malcolm <dmalcolm@gcc.gnu.org>:

https://gcc.gnu.org/g:a113b14398f2a4ad2742e6e9c87e25cac60f263e

commit r12-2399-ga113b14398f2a4ad2742e6e9c87e25cac60f263e
Author: David Malcolm <dmalcolm@redhat.com>
Date:   Mon Jul 19 15:44:02 2021 -0400

    analyzer: add svalue::can_have_associated_state_p [PR101503]

    PR analyzer/101503 reports an assertion failure due to an unexpected
    "UNKNOWN" value (due to using --param analyzer-max-svalue-depth=0).

    This patch fixes this by rejecting attempts to purge state involving
    unknown/poisoned svalues (in region_model::purge_state_involving),
    as these svalues should not have state associated with them - they
    are singletons w.r.t each type.

    To be more systematic about this, the patch also introduces a new
    svalue::can_have_associated_state_p which returns false for
    unknown/poisoned svalues, so that we can reject adding constraints
    or sm-state on them, or building various kinds of svalue in terms
    of them (e.g. unary ops, binary ops, etc).

    gcc/analyzer/ChangeLog:
            PR analyzer/101503
            * constraint-manager.cc (constraint_manager::add_constraint): Use
            can_have_associated_state_p rather than testing for unknown.
            (constraint_manager::get_or_add_equiv_class): Likewise.
            * program-state.cc (sm_state_map::set_state): Likewise.
            (sm_state_map::impl_set_state): Add assertion.
            * region-model-manager.cc
            (region_model_manager::maybe_fold_unaryop): Handle poisoned
            values.
            (region_model_manager::maybe_fold_binop): Move handling of unknown
            values...
            (region_model_manager::get_or_create_binop): ...to here, and
            generalize to use can_have_associated_state_p.
            (region_model_manager::maybe_fold_sub_svalue): Use
            can_have_associated_state_p rather than testing for unknown.
            (region_model_manager::maybe_fold_repeated_svalue): Use unknown
            when the size or repeated value is "unknown"/"poisoned".
            * region-model.cc (region_model::purge_state_involving): Reject
            attempts to purge unknown/poisoned svalues, as these svalues
            should not have state associated with them.
            * svalue.cc (sub_svalue::sub_svalue): Assert that we're building
            on top of an svalue with can_have_associated_state_p.
            (repeated_svalue::repeated_svalue): Likewise.
            (bits_within_svalue::bits_within_svalue): Likewise.
            * svalue.h (svalue::can_have_associated_state_p): New.
            (unknown_svalue::can_have_associated_state_p): New.
            (poisoned_svalue::can_have_associated_state_p): New.
            (unaryop_svalue::unaryop_svalue): Assert that we're building on
            top of an svalue with can_have_associated_state_p.
            (binop_svalue::binop_svalue): Likewise.
            (widening_svalue::widening_svalue): Likewise.

    gcc/testsuite/ChangeLog:
            PR analyzer/101503
            * gcc.dg/analyzer/pr101503.c: New test.

    Signed-off-by: David Malcolm <dmalcolm@redhat.com>