| Age | Commit message (Collapse) | Author |
|---|
|
erlang:get_stacktrace/0 returns the stacktrace for the latest
exception. The problem is that the stacktrace is kept until the next
exception occurs. If the last exception was a 'function_clause' or a
'badarg', the arguments for the call are also kept forever. The
arguments can be terms of any size (potentially huge).
In a future release, we would like to only allow
erlang:get_stacktrace/0 from within a 'try' expression. That would
make it possible to clear the stacktrace when the 'try' expression is
exited.
The 'catch' expression has no natural end where the stacktrace could
be cleared. The stacktrace could be cleared at the end of the function
that the 'catch' occurs in, but that would cause problems in the
following scenario (from real life, but simplified):
try
...
catch _:_ ->
io:format(...),
io:format("~p\n", [erlang:get_stacktrace()])
end.
%% In io.erl.
format(Fmt, Args) ->
Res = case ... of
SomePattern ->
catch...
...;
SomeOtherPattern ->
%% Output the formatted string here
...
end,
clear_stacktrace(), %% Inserted by compiler.
Res.
The call to io:format() would always clear the stacktrace before
it could be retrieved.
That problem could be solved by tightning the scope in which the
stacktrace is kept, but the rules for how long erlang:get_stacktrace/0
would work would become complicated.
Therefore, the solution we suggest for a future major release of
OTP is that erlang:get_stacktrace/0 will return [] if it is called
outside the 'catch' part of a 'try' expression.
To help users prepare, introduce a warning when it is likely that
erlang:get_stacktrace/0 will always return an empty list, for example
in this code:
catch error(foo),
Stk = erlang:get_stacktrace()
or in this code:
try Expr
catch _:_ -> ok end,
Stk = erlang:get_stacktrace()
|
|
As of the introduction of Unicode characters in atoms, the control
sequences 'w' and 'W' can return non-Latin-1 characters, unless some
measure is taken.
This commit makes sure that '~w' and '~W' always return Latin-1
characters, or bytes, which can be output to ports or written to raw
files.
The Unicode translation modifier 't' is needed to return non-Latin-1
characters.
|
|
The same checks are also performed by the Dialyzer.
|
|
In OTP 20, all Unicode characters are allowed in atoms.
However, we have decided that we will not allow non-latin1
characters in module names in OTP 20. This restriction may
be lifted in a future major release of OTP.
Enforce the restriction in erl_lint, so that it will be a
compilation error to have a module name containing non-latin1
characters. That means that tools such as debugger, xref,
dialyzer, syntax_tools, and so on, do not have to be modified
to handle non-latin1 module names.
|
|
|
|
|
|
The filters in a list comprehension can be guard expressions or
an ordinary expressions.
If a guard expression is used as a filter, an exception will basically
mean the same as 'false':
t() ->
L = [{some_tag,42},an_atom],
[X || X <- L, element(1, X) =:= some_tag]
%% Returns [{some_tag,42}]
On the other hand, if an ordinary expression is used as a filter, there
will be an exception:
my_element(N, T) -> element(N, T).
t() ->
L = [{some_tag,42},an_atom],
[X || X <- L, my_element(1, X) =:= some_tag]
%% Causes a 'badarg' exception when element(1, an_atom) is evaluated
It has been allowed for several releases to override a BIF with
a local function. Thus, if we define a function called element/2,
it will be called instead of the BIF element/2 within the module.
We must use the "erlang:" prefix to call the BIF.
Therefore, the following code is expected to work the same way as in
our second example above:
-compile({no_auto_import,[element/2]}).
element(N, T) ->
erlang:element(N, T).
t() ->
L = [{some_tag,42},an_atom],
[X || X <- L, element(1, X) =:= some_tag].
%% Causes a 'badarg' exception when element(1, an_atom) is evaluated
But the compiler refuses to compile the code with the following
diagnostic:
call to local/imported function element/2 is illegal in guard
|
|
The compiler would crash in v3_codegen when trying to compile the
following code:
is_port(_) -> false.
foo(P) when port(P) -> ok.
We *could* have the compiler interpret the code as:
is_port(_) -> false.
foo(P) when erlang:is_port(P) -> ok.
But that would encourage using the obsolete form of the guard tests.
Note that the following code is illegal:
is_port(_) -> false.
foo(P) when is_port(P) -> ok.
It produces the following diagnostic:
call to local/imported function is_port/1 is illegal in guard
Therefore, we should refuse to compile the code.
|
|
|
|
|
|
Map keys cannot be bound and then used in parallel matching.
Example:
#{ K := V } = #{ k := K } = M.
This is illegal if 'K' is not already bound.
|
|
|
|
|
|
|
|
|
|
* maint:
xmerl: Remove 'no_return' Dialyzer warnings
xmerl: Add suppression of Dialyzer warnings
eunit: Add suppression of Dialyzer warnings
debugger: Add suppression of Dialyzer warnings
kernel: Add suppression of Dialyzer warnings
mnesia: Add suppression of Dialyzer warnings
observer: Add suppression of Dialyzer warnings
runtime_tools: Add suppression of Dialyzer warnings
stdlib: Add suppression of Dialyzer warnings
test_server: Add suppression of Dialyzer warnings
tools: Add suppression of Dialyzer warnings
Conflicts:
lib/stdlib/src/erl_lint.erl
lib/stdlib/src/otp_internal.erl
|
|
|
|
Calls to map_anno(), fold_anno(), and mapfold_anno() with lists of
erl_parse trees have been replaced. Those functions accept lists of
erl_parse trees, but it was not the intention when the functions were
introduced, and it is not documented.
|
|
* maint:
stdlib: Fix linter crash due to missing -module declaration
Conflicts:
lib/stdlib/test/erl_lint_SUITE.erl
|
|
The Erlang Code Linter no longer crashes if there is a -deprecated()
attribute but no -module() declaration.
See also ERL-62 at bugs.erlang.org.
|
|
The syntax -spec/callback F/A :: FunctionType; has been removed.
No deprecation was deemed necessary.
|
|
* maint:
Extend erl_lint:format_error/1 to handle bittype mismatches
erl_lint_SUITE: Add smoke test of format_error/1
|
|
erl_lint:format_error/1 would crash with a function error if
conflicting types were given. That was most easily noticed in the
shell:
Eshell V7.0.3 (abort with ^G)
1> <<0/integer-binary>>.
*** ERROR: Shell process terminated! ***
Noticed-by: Aleksei Magusev
|
|
This type modifier was missed in 90efeaf21147505b1e8207822e606027f94183cc.
|
|
The recently added module erl_anno can no longer handle
negative line numbers.
|
|
|
|
* josevalim/jv-annotate-form-type:
Annotate used types in erl_lint
OTP-12800
|
|
This is a follow up to commit b5ee5c6.
Similar to function calls, we also need to annotate used types
on call site since this information is used later on when showing
undefined type errors.
Notice we don't need to annotate the type export because
it is an attribute and those are always properly annotated.
|
|
* josevalim/jv-annotate-form:
Only annotate forms when linting in the compiler
OTP-12772
|
|
A lot of time spent on linting is due to eval_file_attribute/2
function that recursively traverses all the AST, annotating
the file name on the AST nodes. This traversal happens so
it is easier to add errors and warnings later on by simply
introspecting the node.
This patch changes eval_file_attribute/2 to only annotate
forms (i.e. attributes and functions) and rely on the #lint
record to keep the proper file information (which it already
did before this patch).
To summarize, both pre scan and pos scan will use the annotated
file attribute in forms, however form/2 already maintains the
proper file in #lint, which we pass around when retrieving the
location information.
After this patch, linting of a regular erlang module with 500LOC
became twice faster when measured with eprof.
|
|
Parameterized modules were removed in cdf8060868575, but a few
vestiges still remained in erl_lint.
|
|
When compiling parser files, because they rely heavily
on inline annotations, retrieving the nowarn_bif_clash
information from the compiler options is expensive.
This patch stores nowarn_bif_clash in the lint record.
By using erlc +'{eprof,lint_module}' when compiling the
erlang parser, we noticed the time spent on
nowarn_function/2 reduced from 30% to 0.01%.
|
|
|
|
* nox/stdlib/erl_lint-maps/OTP-12515:
Properly lint map expressions in erl_lint
|
|
* nox/stdlib/erl_lint-expr_list:
Remove a few superfluous vt operations in erl_lint
|
|
Given any Vt1 and Vt2 values, vtmerge(vtnew(Vt1, Vt2), vtold(Vt1, Vt2)) is always
equal to Vt1.
|
|
The returned variable table when linting a map expression shouldn't include
variables that didn't appear in the expression.
Reported-By: Alexei Sholik
|
|
The -dialyzer() attribute can be used for suppressing warnings in a
module by specifying functions or warning options. It can also be used
for requesting warnings in a module.
|
|
Robert has OK'ed the removal of the token ':-'.
|
|
* egil/maps/variable-keys/OTP-12218: (22 commits)
compiler: Update test for Maps aliasing
compiler: Properly support Map aliasing
compiler: Refactor Map pairs aliasing
compiler: Fix harmless need_heap error for Maps
stdlib: Update Map tests
stdlib: Use environment bindings for Maps keys in erl_eval matching
debugger: Update Map tests
compiler: Update Map tests
compiler: Fix v3_core Maps pair chains
compiler: Use expressions in core patterns
compiler: Use variables in Map cerl inliner
compiler: Reintroduce binary limit for Map keys
compiler: Shameless v3_core hack for variables
compiler: Use variables in Map beam assmebler
compiler: Use variables in Map kernel pass
compiler: Use variables in Map core pass
compiler: Normalize unary ops on Maps key literals
stdlib: Update Map tests
stdlib: erl_lint Map key variables
compiler: Maps are always patterns never values in matching
...
|
|
* nox/fix-exporting-rules/OTP-12186:
Rewrite merge of clause variable tables (in case, try, etc)
|
|
|
|
erl_lint:icrt_export/4 has been rewritten to make the code really
follow the scoping rules of Erlang, and not just in most situations
by accident.
* The function should not depend on calling unused_vars/3 because that
function does not return variables which begins with an underscore,
something that only matters when emitting warnings. This could cause
a compiler crash if such a variable was reused afterwards.
* The variable tables from each clause are first merged together,
lists:merge/1 is safe to use because they are orddicts and thus
already sorted. This list is then traversed parallelly to the old
variable table, again taking advantage of their sorted order.
* The function does not emit warnings itself, there is no need to pass
around the lint state. In the same vein, vtunsafe/3 has been rewritten
to do more things by itself, given that all of its calls were similar.
Finally, compiled-out code has been removed.
* This reverts the code in 9ce148b1059e4da746a11f1d80a653340216c468,
which fixed the compiler crash and made erl_lint remember unsafe
variables, but forget about unused variables in the process.
* Other places of the code which relied on the old clunky behaviour were
also updated: unused and unsafe old variables are forgotten when
merging fun clauses and boolean shortcircuiting operators do not rely
on icrt_export/3 anymore.
|
|
The pre-defined types array(), dict(), digraph(), gb_set(), gb_tree(),
queue(), set(), and tid() have been removed.
|
|
Use F/A rather than M:F/A for local functions.
|
|
Types are represented by quadruples {type, LINE, Name, Args},
but maps were represented by five-tuples
{type, LINE, map_field_assoc, Dom, Range}.
Note: this is *not* about the quadruples used for representing
expressions, {map_field_assoc,L,K,V}.
|
|
sys_pre_expand used to crash. There is no known reason to
allow -callback attributes with explicit module.
|
|
product/_, union/_, range/2 as well as tuple/N (N > 0), map/N (N > 0),
atom/1, integer/1, binary/2, record/_, and 'fun'/_ can now be used as
type names.
|
|
|
|
When redefining and exporting the type map() erl_lint erroneously
emitted an error. This bug has been fixed.
|
