This would end up taking the successors_count=2 case, even though
we need to treat SWITCH and MATCH differently. This incorrectly
marked a block as FOLLOW, resulting in incorrect block pass
optimization.
Fixes oss-fuzz #39380.
This was doing a plain copy of JMPZNZ, even though it encodes
offsets relative to the opline. As such, the offsets would be
relative to target, while they should be relative to opline.
Fix this by recomputing them.
Fixes oss-fuzz #39295.
This ensures that code directly before the loop var free is
separated out (and will generally be eliminated as unreachable).
This fixes some assumptions we have that unreachable loop var free
blocks start with the loop var free.
Fixes oss-fuzz #39395.
We can't remove a trivial phi of the form x = phi(x), because we
don't have a replacement value. We could drop the whole block
though. SCCP would normally do this, but in this particular case
we only determine non-reachability based on type information.
Fixes oss-fuzz #39316.
If we're removing a predecessor because it already exists during
replacement, we should also drop pi nodes for that predecessor.
Fixes oss-fuzz #39276.
We were not inferring anything for the ASSIGN_STATIC_PROP_REF
result type at all, leaving it as an empty type. In the test
case this results in a live range being incorrectly eliminated,
but this could break in all kinds of other ways as well..
We have an invariant that an array value type is set if and only
if an array key type is set, which is violated for the case of
undef values for always invalid keys.
While technically legal, this may cause unexpected situations
(in this example, setting an FE_FREE operand to constant null)
and is suboptimal anyway. It's better to preserve the vacuous type
and drop it later (though we currently don't implement this).
The evaluation of the initializer may throw. This could be refined
by checking whether the initializer is a constant AST. For now
just fix the miscompile.
We'd have usually converted it into a PRE_INC if there is no use,
but that's not guaranteed. If there is no use at this point, make
sure we don't try to use the sentinel value.
This issue is properly fixed by GH-7121 on master. For older
branches, disable the use of range information in SCCP, to
reduce impact of potentially incorrect ranges.
Literal compaction was incorrectly assuming that literals with
the same base literal and the same number of related literals
would be equal. Maybe that was the case historically, but at
least it isn't true in PHP 8, where FETCH_CONSTANT and INIT_METHOD
have distinct literals at the second position.
Fix this by making the cache key a concatenation of all literals,
rather than just the base literal. We still distinguish the number
of related literals based on a bias added to the string hash.
Make sure that the previous opline is part of the same block,
otherwise it may be non-dominating.
The test case does not fail on PHP-7.4, but I think the general
problem can appear on 7.4 as well, so I'm applying the patch to
that branch.
The ASSERT_CALLBACK value is not validated at all -- it's possible
to set it to an arbitrary value. As such, the function can also
return any value or type (even without outright abuse, the opcache
func info was wrong in that the return can be rcn, and the array
can be array_of_ref).
Function info for curl_exec() incorrect specified that the
function cannot return true. This is already fixed in PHP 8,
as the func info entry was removed there.
