| Age | Commit message (Collapse) | Author |
|---|
|
OTP-10106
OTP-10107
|
|
|
|
* bjorn/compiler/coverage-and-minor-fixes/OTP-9982:
v3_life: Use common code for guards and bodies
v3_core: Don't put negative line numbers in annotations
v3_kernel: Dig out the line number only when generating a warning
v3_kernel: Clean up handling of guards
v3_kernel: Introduce is_in_guard/1
v3_kernel: Removed unreached clause for #k_bin_int{} in sub_size_var/1
v3_kernel: Remove unreached handling of #k_bin_int{} in arg_con/1
v3_codegen: Eliminate the special case of 'put' without destination
v3_kernel: Don't attempt to share identical literals
v3_kernel: Handle sequences in guards
v3_kernel: Remove clauses that are never executed in arg_val/1
v3_kernel.hrl: Remove unused record #k_string{}
v3_kernel.erl: Remove unused define of EXPENSIVE_BINARY_LIMIT
sys_core_fold: Refactor previous_ctx_to_binary/2 to cover it completely
sys_core_fold: Fix opt_guard_try/1
sys_core_fold: Simplify opt_bool_not() to cover it completely
sys_core_fold: Remove duplicate optimization
|
|
|
|
In Core Erlang and later passes, compiler-generated code can be
indicated in two different ways: by negative line numbers and by
a 'compiler_generated' annotation.
Simplify the code and improve coverage by turning negative line
numbers positive and adding a 'compiler_generated' annotation in
the v3_core pass. That means that Core Erlang and latter passes
do not have deal with negative line numbers.
|
|
|
|
Handling of guards become a mess when andalso/orelse in guards
were introduced. Clean it up by using common code for most
constructs, regardless of where they occur, and when needed use
is_in_guard/1 to handle guard constructs differently.
|
|
* ks/compile_info-fix:
compiler: Fix discrepancy in compile_info
OTP-9917
|
|
The BEAM disassembler used the atom 'none' to signify the absence
of a compile_info chunk in a .beam file. This clashes with the type
declaration of the compile_info field of a #beam_file{} record as
containing a list. Use [] to signify the absence of this chunk.
This simplifies the code and avoids a dialyzer warning.
For fixing a similar problem and for consistency, changed also the
return type of the attributes field of the #beam_file{} record.
This required a change in the beam_disasm test suite.
|
|
The inliner was ignorant of on_load functions and would discard them
(unless they were exported or referenced).
Noticed-by: Yiannis Tsiouris <[email protected]>
|
|
In the following code excerpt, the instruction marked below was
incorrectly removed:
.
.
.
{'try',{y,2},{f,TryCaseLabel}}.
{bif,get,{f,0},[{x,0}],{x,0}}.
{move,{x,1},{y,0}}.
{move,{x,3},{y,1}}. <======= Incorrectly removed
{jump,{f,TryEndLabel}}.
{label,TryEndLabel}.
{try_end,{y,2}}.
{deallocate,3}.
return.
{label,TryCaseLabel}.
{try_case,{y,2}}.
.
.
.
beam_utils indicated that {y,1} was not used at TryEndLabel,
which by itself is correct. But it is still not safe to remove
the instruction, because {y,1} might be used at TryCaseLabel
if an exception occurs.
Noticed-by: Eric Merritt
|
|
|
|
The size field of a #k_bin_int{} is always a literal integer
(#k_int{}); thus there is no way this clause can match.
While we are at it, remove a duplicated call to
select_assert_match_possible/3 in select_bin_int/1.
|
|
arg_con/1 will never be called with #k_bin_int{} argument, since
a #k_bin_int{} is constructed out of #k_bin_seg{} only after
arg_con/1 has been called with a #k_bin_seg{} argument.
|
|
If we let v3_kernel make sure that a 'put' operation always has a
destination register, the special case in v3_codegen is not needed.
|
|
It is unlikely that sharing literals will actually save any time.
Literals that can be shared are usually small. It is more likely
that maintaining the dictionary of literals will waste time.
|
|
A guard with a sequence will cause a crash in v3_codegen. Currently,
it seems that a sequence in a guard will never happen when compiling
from Erlang code, but there are several reasons to fix this problem
anyway:
* There are other compilers that generate Core Erlang code.
* Changes to sys_core_fold (perhaps additional optimizations) may
cause sequences to be generated in guards.
* A previously uncovered line in sys_core_fold:opt_guard_try/1 will
now be covered.
|
|
arg_val/1 will never be called for single-valued types, so there
is no need handle these types.
|
|
|
|
|
|
|
|
opt_guard_try/1 assumed that it was only operating on guards, and
implicitly assumed that any function call was to BIFs without
any side effects.
|
|
In:
case not Expr of ...
we don't have the information to ascertain that Expr is a boolean
expression. Therefore, add the catch-all clause unconditionally.
|
|
The second clause in opt_case_in_let_2/3 attempts to do the same
optimization as opt_bool_case_redundant/1; it will never succeed.
|
|
Also correct the comment in bsm_ensure_no_partition_2/5, and while at
it correct typos in the comment for bsm_nonempty/2.
|
|
In order to save space, rewrite suitable calls to erlang:error/{1,2}
to special BEAM instructions.
This code is probably longer than the code taken out of v3_life and
v3_codegen in the previous commit, but it is much easier to
understand and maintain since the BEAM assembler format is better
understood than the v3_life format.
|
|
In the v3_life pass, it is assumed that a 'match_fail' primop
only occur at the top-level and at the end of a function.
But this code:
do_split_cases(A) ->
case A of
x ->
Z = dummy1;
_ ->
Z = dummy2,
a=b
end,
Z.
will be optimized by sys_core_fold to the following code:
'split_cases'/1 =
fun (_cor0) ->
let <_cor7,Z> =
case _cor0 of
<'x'> when 'true' ->
< 'dummy1','dummy1' >
<_cor6> when 'true' ->
%% Here follows a 'match_fail' primop inside
%% multiple return values:
< primop 'match_fail'({'badmatch','b'}),'dummy2' >
end
in
Z
moving the 'match_fail' primop into a "values" construction.
In the future, we would like to get rid of the v3_life pass (it is
there for historical reasons), so in the mean-time we prefer to not
add more code to it by generalizing the handling of 'match_fail'.
Since the 'match_fail' primop can be simulated by erlang:error/{1,2},
the simplest solution is to translate 'match_fail' to a call to
erlang:error/{1,2} in v3_kernel and remove the handling of
'match_fail' in v3_life and v3_codegen.
It is tempting to get rid of 'match_fail' also in the Core Erlang
format, but there are two issues:
- Removing the support for 'match_fail' completely may break tools
that generate Core Erlang code. We should not do that in a minor
release.
- There is no easy way to generate a 'function_clause' exception
that will remain correct if it will be inlined into another
function. (Calling "erlang:error(function_clause, Args)" is
fine only if it is not inlined into another function.) A good
solution probably involves introducing new instructions, which
is better done in a major release.
Noticed-by: Håkan Matsson
Minimized-test-case-by: Erik Søe Sørensen
|
|
|
|
A bs_start_match2/5 instruction will never fail (take the branch) if
given a match state or a binary. Therefore, the type at the failure
label should be 'term', not match state or binary. Without this
correction, the beam_validator will reject safe programs.
|
|
To facilitate debugging of compiler bugs, teach the compiler the
'no_dead' option. Since the beam_dead pass used to do the necessary
splitting of basic blocks to expose all labels, we must move that
splitting into a separate pass that is always run.
|
|
Mark math:pi/0 as pure, informing the compiler that the value
can be calculated at compile time.
|
|
* bjorn/compiler/options/OTP-9752:
filename documentation: Recommend against using filename:find_src/1,2
Teach filename:find_src/1,2 to handle slim or stripped BEAM files
filename: Eliminate failing call to Mod:module_info(source_file)
filename.erl:filter_options/1: Remove handling of dead options
compiler: Don't include {cwd,_} in module_info(compile)
compiler: Don't include source code options in module_info(compile)
hipe: Teach hipe to handle slim or stripped BEAM files
|
|
* bjorn/major-release-cleanups:
observer tests: Test compatibility with R13, not R12
emulator tests: Test compatibility with R13, not R12
Teach the compiler the 'r14' option
erl_lint: The types introduced in R12B-5 are no longer "newly introduced"
otp_internal: Stop warning for functions removed in R12 or earlier
Conflicts:
lib/stdlib/src/otp_internal.erl
|
|
The {cwd,Dir} option is always included if the module has been
compiled by erlc. Since its presence cannot be relied upon and it
wastes memory, get rid of it.
|
|
As far as I know, the purpose of the compiler options included in
Mod:module_info(compile) has never been documented.
An educated guess is that they are there in case you want to
re-compile the module with the same options, and also to aid in
debugging when you need to know how a module was compiled.
In neither case is there any need to include options given in the
source itself in options included in Mod:module_info(compile).
Including those options will only waste memory.
Therefore, only include in the BEAM file the options that were
given to compile:file/2.
|
|
|
|
Starting in ff432e262e65243cbc983fcb002527f8fae8c78b, sys_pre_expand
passes external funs through to the downstream passes. It used to
translate external funs to a call to erlang:make_fun/3. Therefore, we
will now need to handle external funs in sys_expand_pmod.
Noticed-by: Stavros Aronis
|
|
* bjorn/line-numbers/OTP-9468:
beam_type: Improve FP optimizations in the presence of line numbers
|
|
* bjorn/fun-improvements/OTP-9667:
sys_pre_expand: Remove incorrect comment
compiler: Eliminate use of deprecated erlang:hash/2
beam_asm: Fix broken NewIndex in fun entries
beam_asm: Strenghten the calculation of Uniq for funs
|
|
sys_pre_expand does not keep track of used or new variables (and have
not done since about the time Core Erlang was introduced).
|
|
Now that beam_asm computes the Index and Uniq values for funs, there
is no need to compute those values in the sys_pre_expand and v3_kernel
modules, thus eliminating the calls to the deprecated erlang:hash/2
function.
It would be tempting to stop generating the name for the fun in
sys_pre_expand so that we did not have to add the Info field to a tuple.
But:
* The debugger depends on the name being there. (Simple solution:
Let the debugger generate the name itself.)
* When a fun has been inlined into another function, the fun name in
'id' annotation will be used to notice the inlining and change the
final clause of the top-level case from generating a 'function_clause'
exception to a case_clause exception. (Possible workaround: Have the
inliner set an inlined attribute on functions that have been inlined, or
have the inliner rewrite 'function_clause' exceptions itself.)
|
|
The calculation of the NewIndex field in fun entries is broken: the
sys_pre_expand and v3_kernel modules keep separate index counters
starting at zero; thus there is no guarantee that each fun within a
module will have its own unique NewIndex.
We don't really need the NewIndex any more (see below), but since
we do need the NewUniq field, we should fix NewIndex for cleanliness
sake. The simplest way is to assign NewIndex as late as possible,
namely in beam_asm, and to set it to the same value as Index.
Historical Note: Why NewIndex Was Introduced
There was once an idea that the debugger should be able to interpret
only a single function in a module (for speed). To make sure that
interpreted funs could be called from BEAM code and vice versa,
the fun identification must be visible in the abstract code.
Therefore a NewIndex field was introduced in each fun in the abstract
code.
However, it turned out that interpreting single functions does not
play well with aggressive code optimization. For example, in this
code:
f() ->
X = 1,
fun() -> X+2 end.
the variable X will seem to be free in the fun, but an aggressive
optimizer will replace X with 1 in the fun; thus the fun will no
longer have any free variables. Therefore, the debugger will always
interpret entire modules.
|
|
Funs are identified by a triple, <Module,Uniq,Index>, where Module is
the module name, Uniq is a 27 bit hash value of some intermediate
representation of the code for the fun, and index is a small integer.
When a fun is loaded, the triple for the fun will be compared to
previously loaded funs. If all elements in the triple in the newly
loaded fun are the same, the newly loaded fun will replace the previous
fun. The idea is that if Uniq are the same, the code for the fun is also
the same.
The problem is that Uniq is only based on the intermediate representation
of the fun itself. If the fun calls local functions in the same module,
Uniq may remain the same even if the behavior of the fun has been changed.
See
http://erlang.org/pipermail/erlang-bugs/2007-June/000368.htlm
for an example.
As a long-term plan to fix this problem, the NewIndex and NewUniq
fields was added to each fun in the R8 release (where NewUniq is the
MD5 of the BEAM code for the module). Unfortunately, it turns
out that the compiler does not assign unique value to NewIndex (if it
isn't tested, it doesn't work), so we cannot use the
<Module,NewUniq,NewIndex> triple as identification.
It would be possible to use <Module,NewUniq,Index>, but that seems
ugly. Therefore, fix the problem by making Uniq more unique by
taking 27 bits from the MD5 for the BEAM code. That only requires
a change to the compiler.
Also update a test case for cover, which now fails because of the
stronger Uniq calculation. (The comment in test case about why the
Pid2 process survived is not correct.)
|
|
Currently, the external fun syntax "fun M:F/A" only supports
literals. That is, "fun lists:reverse/1" is allowed but not
"fun M:F/A".
In many real-life situations, some or all of M, F, A are
not known until run-time, and one is forced to either use
the undocumented erlang:make_fun/3 BIF or to use a
"tuple fun" (which is deprecated).
EEP-23 suggests that the parser (erl_parse) should immediately
transform "fun M:F/A" to "erlang:make_fun(M, F, A)". We have
not followed that approach in this implementation, because we
want the abstract code to mirror the source code as closely
as possible, and we also consider erlang:make_fun/3 to
be an implementation detail that we might want to remove in
the future.
Instead, we will change the abstract format for "fun M:F/A" (in a way
that is not backwards compatible), and while we are at it, we will
move the translation from "fun M:F/A" to "erlang:make_fun(M, F, A)"
from sys_pre_expand down to the v3_core pass. We will also update
the debugger and xref to use the new format.
We did consider making the abstract format backward compatible if
no variables were used in the fun, but decided against it. Keeping
it backward compatible would mean that there would be different
abstract formats for the no-variable and variable case, and tools
would have to handle both formats, probably forever.
Reference: http://www.erlang.org/eeps/eep-0023.html
|
|
Allow line/1 instructions to be part of a sequence of floating point
instructions to avoid outputting fclearerror / fcheckerror around
every floating point instruction.
|
|
'behaviour_info(callbacks)' is a special function that is defined in a module
which describes a behaviour and returns a list of its callbacks.
This function is now automatically generated using the '-callback' specs. An
error is returned by lint if user defines both '-callback' attributes and the
behaviour_info/1 function. If no type info is needed for a callback use a
generic spec for it.
|
|
Behaviours may define specs for their callbacks using the familiar spec syntax,
replacing the '-spec' keyword with '-callback'. Simple lint checks are performed
to ensure that no callbacks are defined twice and all types referred are
declared.
These attributes can be then used by tools to provide documentation to the
behaviour or find discrepancies in the callback definitions in the callback
module.
|
|
* dev:
Update copyright years
|
|
|
|
|
