| Age | Commit message (Collapse) | Author |
|---|
|
|
|
Dialyzer sometimes output warnings like
Attempt to test for inequality between a term of type 'false' and a
term of opaque type 'false' | gb_set()
The reason was that erl_types:t_inf/3 when called from
erl_types:t_find_unknown_opaque() did not return 'false' but found the
mismatch. It should not return the mismatch unless the intersection is
empty ('none').
Thanks to Shayan Pooya [[email protected]] for pointing out the bug.
|
|
Two steps are needed to make this work:
1) Avoid generating the additional "apply_constraint" in dialyzer_typesig by
reporting every function argument as a potential external function (patch on
dialyzer_dep).
This will produce correct success typings for all functions in the test case,
but dataflow would miss the key warnings that help identify the bugs.
2) Patch dialyzer_dataflow so that it uses the "handle just external" path as a
fallback whenever there are any external calls. As a result, if we have info
about some paths, then:
a) use the old "handle known apply" code to mark these functions as used and
b) ignore the generalized result and use the one found by typesig for the
return value of the apply itself.
|
|
* aronisstav/hipe/opaque_fix:
Don't 'opaque-decorate' a success typing using an incompatible spec
|
|
Without this patch Dialyzer crashes when analyzing the supplemented test case.
|
|
|
|
Guard constraints used to be limited to a certain depth, which handled
mutually depending constraints safely, but also sometimes introduced
unnecessary generalizations.
This patch puts no explicit limit upon guard constraints (other than
those that already exist in erl_types), but breaks cycles by replacing
variables with the any() type.
In some cases the old method resulted in more warnings, but since the
limit was quite arbitrary and mutually depending guard constraints are
(very) rare, the new method should been seen as an improvement since
it handles cases that used to make Dialyzer loop or miss warnings.
|
|
|
|
* hb/dialyzer/fix_on_load/OTP-11743:
FIx handling of 'on_load' attribute
|
|
[pull request from Kostis Sagonas]
The handling of functions appearing in an 'on_load' attribute was wrong.
Instead of considering the functions specified in these attributes as
escaping from the module and performing a full analysis starting from
them, the code just bypassed this analysis and only suppressed unused
warning messages for these functions. This worked for most of the cases
but resulted in functions (directly or indirectly) called by 'on_load'
functions being reported as not called by the module.
Such a case existed in the code of the 'crypto' application.
To solve these issues the initialization code for functions escaping
from the module was changed and the test for the on_load functionality
was appropriately extended.
|
|
The types array(), dict(), digraph(), gb_set(), gb_tree(), queue(),
set(), and tid() have been deprecated. They will be removed in OTP 18.0.
Instead the types array:array(), dict:dict(), digraph:graph(),
gb_set:set(), gb_tree:tree(), queue:queue(), sets:set(), and ets:tid()
can be used. (Note: it has always been necessary to use ets:tid().)
It is allowed in OTP 17.0 to locally re-define the types array(), dict(),
and so on.
New types array:array/1, dict:dict/2, gb_sets:set/1, gb_trees:tree/2,
queue:queue/1, and sets:set/1 have been added.
|
|
|
|
|
|
|
|
Hans Bolider reported a dialyzer crash when using a remote type in
the tail position of a maybe_improper_list() declaration. A test
was created (by extending an existing module of the testsuite)
and erl_types was modified to expand the remote type and not pass
it unexpanded to subsequent phases in the processing.
|
|
|
|
|
|
|
|
|
|
|
|
This changed the lines so the results now differ.
|
|
It makes no sense to be able to do `<<...,Rest/binary>> <= ...` in a
comprehension. The related Dialyzer test is removed.
|
|
|
|
Dialyzer was constraining bitstring data used in the construction of other
bitstring values too much. These constraints have now been relaxed.
|
|
|
|
* sa/dialyzer-unknown-arity-funs/OTP-10772:
Update one more unused funs result as per ffe582b
Remove pubsub and comm_layer tests from testsuite
Properly support functions with arbitrary arity in type specs.
Conflicts:
lib/hipe/cerl/erl_types.erl
|
|
These tests recently broke because of the deprecation of the package system.
Furthermore, they included copyrighted code.
Finally they were not minimized not targeted to any specific feature and had no
warnings being reported.
|
|
Dialyzer now understands the "fun(...)" syntax in type specs.
|
|
|
|
|
|
* sa/dialyzer-record-field-type/OTP-10681:
Fix precision of record creation violation warnings
|
|
|
|
Before patch Dialyzer was reporting all the fields that were not subtypes
of the declared ones as incorrect. The correct violations are for the fields
whose intersection with the declared ones is empty.
|
|
Originally reported by Joseph Wayne Norton, simplified testcase provided by Hans
Bolinder.
|
|
1. Sometimes the solver forgot that a list had entered the error
state. The bug has been fixed by storing the atom 'error' in
MapDict. An example where the bug occurred is
io_lib_pretty:printable_bin(). The returned spec was weaker than it
should have been, but the fix-point loop hid the bug (in this case).
2. lists:partition() has been substituted for lists:splitwith() in
enumerate_constraints(). This fix together with 3. solves a
problem with long execution times for deeply nested fun:s. An
example which is now much faster is
lib/compiler/test/lc_SUITE:deeply_nested/1
(included as dialyzer/test/small_SUITE_data/src/deep_lc.erl).
3. The calculation of components in enumerate_constraints() has been
simplified and optimized. The important thing here is that _all_ of
the simple constraints have been saturated before entering the
complex part.
4. The pretty printing of constraints has been improved.
|
|
* sa/dialyzer-contract-variables:
Correct handling of type names in contracts
Don't rewrite unchanged PLT
Fix crash related to contract checking
OTP-10083
|
|
Variables in contracts can either be true type variables or simply names for
types thet are defined in the 'when' clauses. Consider the following example:
-spec foo(X, Options) -> {ok, X} | error when Options :: [{atom(), boolean()}].
Here X is a true variable whereas Options is a name for a type that is defined
in the when clause.
'when' clauses may further use names on the right side. These were not treated
properly by Dialyzer and could be generalized to the term() type. This patch
fixes this issue.
A further issue is the treatment of true type variables, but this is left for
another patch.
|
|
When "collapsing" heads of list elements from a list's tail into the list's
head the new head can be different.
|
|
When spawning OS (unix) processes with erlang:open_port, store the
resulting unix pid so that it can be queried later on using
erlang:port_info/1,2.
|
|
Depending of the ordering of the functions during dataflow, a
function with an infinite loop might be identified as one that
always crashes. This is fixed now, by allowing restoration of
the infinitely-looping status.
|
|
|
|
Dialyzer was not reporting unused functions with 0 arity. This was not
a real issue, until we found out that there could be cases where this
could lead to false warnings. This was the case in "no_local_return.erl".
|
|
Variable substitution was not generalizing any unknown variables.
|
|
Whenever a variable that could hold one of two or more possible closures was
used in a particular application, the application was assumed to fail if ONE of
the closures would fail in this application. This has been corrected to infer
failing application if ALL possible closures would fail in the particular
application.
|
|
|
|
* sa/dialyzer-dev:
Add origin information to #fun_var closures
Suppress some warnings about generation of non-returning funs
Enhance Dialyzer's inference on comparisons
Fix infinite loop in dataflow
Minor fix in dead code
Update r9c/{inets,mnesia} results in dialyzer's test suite
OTP-9529
|
|
* rz/add_proplist_type:
Add a proplist() type
OTP-9499
|
|
No warnings are emitted for funs that are non-returning when the
function that generates them has a contract that specifies that
it will return such a non-returning fun.
The actual bug, reported by Tuncer Ayaz and simplified by Maria
Christakis is included in Dialyzer's tests.
|
|
This patch makes Dialyzer aware of Erlang's total ordering of terms,
enabling discrepancy detection in cases where e.g. integer() < tuple()
is treated as a comparison that might also return false (when it is
certain to always return true).
|
|
|
