This test triggers narrowing for two ops: first ZEND_ADD_ARRAY_ELEMENT,
and then ZEND_ASSIGN.
The type inference happens in the following order:
1) The ZEND_ADD_ARRAY_ELEMENT infers type 0x40e04080 (packed flag is set),
arr_type=0 at this point because it hasn't been set by ZEND_INIT_ARRAY yet.
2) The ZEND_INIT_ARRAY infers type 0x40804080
3) The ZEND_ADD_ARRAY_ELEMENT infers type 0x40e04080, arr_type=0x40804080,
which does not have the packed flag set while the existing result of
ZEND_ADD_ARRAY_ELEMENT has the packed flag set.
This seems to occur because of the phi node introduced by the while
loop. If I remove the loop the problem goes away.
As Arnaud noted, this seems to be caused by a too wide type inference
for arr_type==0. We should keep the invariant that if x>=y then
key_type(x) >= key_type(y).
If we write the possible results down in a table we get:
```
arr_type resulting key type
--------------- --------------------------------------------------------------------------
HASH_ONLY -> MAY_BE_ARRAY_NUMERIC_HASH
PACKED_ONLY -> MAY_BE_ARRAY_NUMERIC_HASH | MAY_BE_ARRAY_PACKED (== MAY_BE_ARRAY_KEY_LONG)
HASH || PACKED -> MAY_BE_ARRAY_NUMERIC_HASH | MAY_BE_ARRAY_PACKED (== MAY_BE_ARRAY_KEY_LONG)
0 -> MAY_BE_ARRAY_NUMERIC_HASH | MAY_BE_ARRAY_PACKED (== MAY_BE_ARRAY_KEY_LONG)
```
As we can see, `HASH_ONLY > 0` but
`MAY_BE_ARRAY_NUMERIC_HASH < MAY_BE_ARRAY_NUMERIC_HASH | MAY_BE_ARRAY_PACKED`,
which violates the invariant.
Instead if we modify the zero case to have MAY_BE_ARRAY_NUMERIC_HASH instead,
we get the following table which satisfies the invariant.
```
arr_type resulting key type
--------------- --------------------------------------------------------------------------
HASH_ONLY -> MAY_BE_ARRAY_NUMERIC_HASH
PACKED_ONLY -> MAY_BE_ARRAY_NUMERIC_HASH | MAY_BE_ARRAY_PACKED (== MAY_BE_ARRAY_KEY_LONG)
HASH || PACKED -> MAY_BE_ARRAY_NUMERIC_HASH | MAY_BE_ARRAY_PACKED (== MAY_BE_ARRAY_KEY_LONG)
0 -> MAY_BE_ARRAY_NUMERIC_HASH
```
Broke in 1ffbb73.
Closes GH-10294.
The block optimizer pass allows the use of sources of the preceding
block if the block is a follower and not a target. This causes issues
when trying to remove FREE instructions: if the source is not in the
block of the FREE, then the FREE and source are still removed. Therefore
the other successor blocks, which must consume or FREE the temporary,
will still contain the FREE opline. This opline will now refer to a
temporary that doesn't exist anymore, which most of the time results in
a crash. For these kind of non-local scenarios, we'll let the SSA
based optimizations handle those cases.
Closes GH-11251.
There is a typo which causes the AND and OR range inference to infer a
wider range than necessary. Fix this typo. There are many ranges for
which the inference is too wide, I just picked one for AND and one for
OR that I found through symbolic execution.
In this example test, the previous range inferred for test_or was [-27..-1]
instead of [-20..-1].
And the previous range inferred for test_and was [-32..-25]
instead of [-28..-25].
Closes GH-11170.
Fixes GH-10801
Named arguments are not supported by the constant evaluation routine, in
the sense that they are ignored. This causes two issues:
- It causes a crash because not all oplines belonging to the call are
removed, which results in SEND_VA{L,R} which should've been removed.
- It causes semantic issues (demonstrated in the test case).
This case never worked anyway, leading to crashes or incorrect behaviour,
so just prevent CTE of calls with named parameters for now.
We can choose to support it later, but introducing support for this in
a stable branch seems too dangerous.
This patch does not change the removal of SEND_* opcodes in remove_call
because the crash bug can't be triggered anymore with this patch as
there are no named parameters anymore and no variadic CTE functions
exist.
Closes GH-10811.
The code fetched the class entry into ce for objects and static
properties. However, when the actual update needs to take place (when
result_def exists), the class entry in ce was reset to NULL. So the SSA
object type update never happened. Fetch the class entry in the
result_def>=0 case instead after the reset of ce to NULL.
The "nothing to do" case would never be hit because the switch block
would execute if the opcode is ZEND_ASSIGN_STATIC_PROP_OP,
not ZEND_ASSIGN_STATIC_PROP. This meant that we were falling through to
the else block. Fix this by correcting the check condition.
We're in the case of ZEND_JMPZ_EX or ZEND_JMPNZ_EX. The opcode gets
overwritten and only after the overwriting gets checked if we're in a
JMPZ or JMPNZ case. This results in a wrong optimization.
Close GH-10329
I don't think this is strictly necessary, but I think it makes
sense to check this before interpreting opline->op2.num as an
argument number.
This also adds one more has_known_send_mode() check that I had
missed before.
