| Commit message (Collapse) | Author | Age | Files | Lines |
|
|
|
|
|
|
|
|
|
| |
The non-constant folding parts of fold_constants are moved into
separate functions. op_integerize handles converting ops to integer
(and special case of OP_NEGATE), op_std_init handling some standard
functionality (forced scalar context and allocating the TARGET).
Both functions are called where fold_constants is called (but we might
want to make that a bit some selective and use op_std_init in other
places).
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This resolves perl bug #97978.
Many built-in variables, like $], are actually created on the fly
when first accessed. Perl likes to pretend that these variables have
always existed, so it autovivifies the *] glob even in rvalue context
(e.g., defined *{"]"}, close "]").
The list of variables that were autovivified was maintained separ-
ately (in is_gv_magical_sv) from the code that actually creates
them (gv_fetchpvn_flags). ‘Maintained’ is not actually precise: it
*wasn’t* being maintained, and there were new variables that never
got added to is_gv_magical_sv and one deleted variable that was
never removed.
There are only two pieces of code that call is_gv_magical_sv, both in
pp.c: S_rv2gv (called by *{} and also the implicit *{} that functions
like close() provide) and Perl_softrefxv (called by ${}, @{}, %{}).
In both cases, the glob is immediately autovivified if
is_gv_magical_sv returns true.
So this commit eliminates the extra maintenance burden by extirpat-
ing is_gv_magical_sv altogether, and replacing it with a new flag to
gv_fetchpvn_flags, GvADDMG, which will autovivify a glob *if* it’s a
magical one.
It does make defined(*{"frobbly"}) slightly slower, in that it creates
a temporary glob and then frees it when it sees nothing magical has
been done with it. But this case is rare enough it should not matter.
At least I got rid of the bugginess.
|
|
|
|
|
|
|
|
|
|
|
|
| |
For functions that take handles as arguments, this code will need to
call static functions in op.c, like is_handle_constructor.
While we could make is_handle_constructor into a non-static function
and call it from gv.c, that seems backwards, as it would result in a
lot of op-manipulation code in the middle of gv.c.
So this commit creates a new function in op.c, called coresub_op,
which is only called from gv.c, from the &CORE::sub code.
|
|
|
|
|
|
|
|
|
|
| |
Subs in the CORE package with a (_) prototype will use this.
This accepts a CV and a sequence number, so that one can
use it to find the $_ in the caller’s scope. It only uses
the topmost call of a subroutine that is being called recur-
sively, so it’s not really a general-purpose function. But
it suffices for &CORE::abs and friends.
|
|
|
|
|
|
|
|
| |
scalarvoid.
Why: The in place assignment is not just an optimisation but has
significant different behaviour and thus doesn't belong in the
peephole optimiser. Also the optree changes are unified and simpler.
|
|
|
|
|
|
|
|
|
| |
This reverts commit e52d58aa5bea245b66786b4c9029e849a2be69d3.
I don’t quite know how I managed it, but I really screw up
this time! Two completely unrelated commits ended up getting
merged into one, so, to avoid confusion down the road, I’m
reverting it, only to reapply it shortly....
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This patch prevents get-magic from executing twice during autovivifi-
cation when the op doing the autovivification is not directly nested
inside the dereferencing op.
This can happen in cases like this:
${ (), $a } = 1;
Previously (as of 5.13.something), the outer op was marked with the
OPpDEREFed flag, which indicated that get-magic had already been
called by the vivifying op (calling get-magic during vivification is
inevitable):
$ perl5.14.0 -MO=Concise -e '${ $a } = 1'
8 <@> leave[1 ref] vKP/REFC ->(end)
1 <0> enter ->2
2 <;> nextstate(main 2 -e:1) v:{ ->3
7 <2> sassign vKS/2 ->8
3 <$> const[IV 1] s ->4
6 <1> rv2sv sKRM*/DREFed,1 ->7 <-- right here
- <@> scope sK ->6
- <0> ex-nextstate v ->4
5 <1> rv2sv sKM/DREFSV,1 ->6
4 <#> gv[*a] s ->5
-e syntax OK
But in the ${()...} example above, there is a list op in the way that
prevents the flag from being set inside the peephole optimizer. It’s
not even possible to set it correctly in all cases, as in this exam-
ple, which would need it both set and not set depending on which
branch of the ternary operator is executed:
${ $x ? delete $a[0] : $a[0] } = 1
Instead of setting the OPpDEREFed flag, we now make a non-magic copy
of the SV in vivify_ref (the first time get-magic is executed).
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
unpack is the only op that takes an implicit $_ for its second argu-
ment. (For others it’s the first.)
Instead of special-casing unpack with its own ck_ routine, we can sim-
ply modify the logic in ck_fun to apply OA_DEFGV to the first optional
argument, not just the first argument.
Currently OA_DEFGV is not set in PL_opargs[OP_UNPACK], which means the
automatically-generated prototype is ($;$), instead of ($_).
This commit sets the flag on the op, changes it to use ck_fun
directly, and updates ck_fun and the prototype-generation code accord-
ingly. I couldn’t put this in multiple commits, as the changes are
interdependent.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Commit b8c38f0a2a65 refactored pp_prototype by moving much of its
code to a new function in op.c, called core_prototype. This served
two purposes: (1) to allow the code to be simplified, which required
the use of static functions in op.c, and (2) to allow the &CORE::subs
feature to share the same code.
But some code was moved to core_prototype which, in hindsight, did not
need to be moved, such as the ‘Can’t find an opnumber’ message.
This commit moves that code back to pp_prototype, resulting in a sim-
pler (and possibly faster, at least for &CORE::subs) core_prototype.
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This refactoring requires the caller to provide the keyword
number to core_prototype. Consequently, it speeds up the code in
gv.c:gv_fetchpvn_flags by allowing it to avoid an extra call to
keyword().
This takes the place of the len parameter, which is no longer used.
It used to be used only as an argument to keyword(). Since the code
that uses strEQ is only reached if the keyword has already been veri-
fied by keyword(), the name simply cannot have embedded nulls, so len
is not necessary.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This commit allows this to work:
BEGIN { *entangle = \&CORE::tie };
entangle $foo, $package;
And the entangle call gets inlined as a tie op, the resulting op tree
being indistinguishable.
These subs are not yet callable via &foo syntax or through a refer-
ence. That will come later, except for some functions, like sort(),
which will probably never support it.
Almost all overridable functions are supported. These few are not:
- infix operators
- not and getprotobynumber (can’t get the precedence right yet;
prototype problem)
- dump
Subsequent commits (hopefully!) will deal with those.
How this works:
gv_fetchpvn_flags is extended with hooks to create subs inside the
CORE package. Those subs are XSUBs (whose C function dies with an
error, for now at least) with a call checker that blows away the
entersub op and replaces it with whatever op the sub represents.
This is slightly inefficient right now, as gv_fetchpvn_flags calls
keyword(), only to have core_prototype call it again. That will
be fixed in a future refactoring.
|
|
|
|
|
|
|
| |
Since it has to calculate it, it might as well provide it, so callers
do not have to go through that while(i < MAXO) loop yet again.
(The &CORE::foo feature will use this.)
|
|
|
|
|
|
|
|
|
|
| |
[perl #95998]
The bareword checking is moved from the peephole optimizer to finalize_optree.
newRANGE needs additional bareword checking because the constants may
be optimized away by 'gen_constant_list'.
The OPpCONST_STRICT flag is removed after giving an error about a
bareword to prevent giving multiple errors about the same bareword.
|
| |
|
|
|
|
|
|
|
| |
After mod was renamed to op_lvalue, this stub was added temporarily
to provide a smoother transition for the compilers. The compiler
maintainer is happy with its extirpation at this stage. See
ticket #78908.
|
|
|
|
|
|
|
|
|
|
|
|
| |
checking/finalization now being done by the peephole optimizer.
This function takes the optree after it is finished building. It
takes over some of the checking and final conversions which are currently being
done by the peephole optimizer.
Add the moment this is an unnecessary extra step after the peephole optimizer, but with
a separate code generation step, the current peephole optimizer can't exists and
this function will take over all its essential compile time functions.
|
|
|
|
|
|
|
|
| |
core_prototype now calls scalar_mod_type in the OA_SCALARREF case.
For core functions, the only thing distinguishing the \$ and
\[$@%*] cases during parsing is the call to scalar_mod_type in
op_lvalue_flags. So calling this same function here just works.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This commit moves the code for generating core prototypes into a sepa-
rate function, core_prototype, in op.c. This serves two porpoises:
• It allows the lock and tie exceptional cases to be incorporated into
the main prototype=generation code, which requires the use of a
static function in op.c.
• It allows other parts of the core (e.g., the upcoming \&CORE::foo
feature) to use the same code.
The docs for it are in a section boringly entitled ‘Functions in
op.c’, for lack of a better name. This, I believe, is the only op.c
function that is in perlintern currently, so it’s hard to see what to
name a section that will, at least for now, contain nothing else.
|
|
|
|
|
| |
PERL_IMPLICIT_SYS only builds on Win32.
Correct embed.fnc to reflect the reality.
|
|
|
|
|
|
|
|
|
| |
9555a685dbd794b0 replaced it with a macro and added the full-name version in
mathoms.c to retain compatibility with any program whose source code uses
the full name. However, as do_exec was never in the API, no program would be
using it. (It's also unconditionally explicitly not exported on various
platforms including Win32. Google Codesearch and grep.cpan.me find no users
of it outside the core.)
|
|
|
|
|
|
|
|
| |
This makes them consistent with other functions that put the basic
datum type first (like hv_*, sv_*, cophh_*).
Since fetch_cop_label is marked as experimental (M), this change
should be OK.
|
|
|
|
|
|
| |
cc76b5cc1552a605 added all 3 functions to the API, but declared prototypes
with const U32 flags, whilst the definitions had that parameter non-const.
Some compilers issue warnings about this inconsistency.
|
| |
|
|
|
|
|
|
|
| |
Move several pad functions into the core API. Document the pad
functions more consistently for perlapi. Fix the interface issues
around delimitation of lexical variable names, providing _pvn, _pvs,
_pv, and _sv forms of pad_add_name and pad_findmy.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Before this commit, this code would fail:
$foo = "foo";
sub foo :lvalue{ return index "foo","o" }
sub bar :lvalue { foo }
$x = bar;
(It would fail for ‘return $]’ as well. Whether it’s a PADTMP or a
read-only scalar makes no difference.)
foo would think it was being called in true lvalue context, because
the entersub op that called it (in bar) was marked that way, bar being
an lvalue sub as well.
The PUSHSUB macro in cop.h needed to be modified to account for
dynamic, or indetermine, context (i.e., indeterminable at compile
time). This happens when an entersub op is an argument to return or
the last statement in a subroutine. In those cases it has to propa-
gate the context from the caller.
So what we now do is this: Both lvalue and in-args flags are turned on
for an entersub op when op_lvalue is called with OP_LEAVESUBLV as the
type. Then PUSHSUB copies into the context stack only those flags
that are set both on the current entersub op and in the context stack
for the previous sub call.
|
|
|
|
|
| |
These functions are internal only with names beginning with underscore.
I hadn't realized that their definitions could be restricted.
|
| |
|
|
|
|
|
| |
This is in preparation for them to be called from another file. Note
that they are still protected by an #ifdef in embed.fnc.
|
|
|
|
|
|
| |
This is in preparation for it to be called from another file. If
for performance reasons it needs to be made inline again, it could
then be moved into a header.
|
|
|
|
|
|
| |
The names now begin with an underscore to emphasize that they are
for internal use only. This is in preparation for making them
accessible beyond regcomp.c.
|
|
|
|
| |
This adds inversion, cloning, and set subtraction
|
|
|
|
|
| |
This element is restricted to either 0 or 1. The comments detail
how its use enables an inversion list to be efficiently inverted.
|
|
|
|
|
| |
Future changes will make the length no longer the same as SvCUR,
so create an element to hold the correct length
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This hasn't been used since 626725768b7b17463e9ec7b92e2da37105036252
Author: Nicholas Clark <nick@ccl4.org>
Date: Thu May 26 22:29:40 2011 -0600
regcomp.c: Fix memory leak regression
here was a remaining memory leak in the new inversion lists data
structure under threading. This solves it by changing the
implementation to use a SVpPV instead of doing our own memory
management. Then the already existing code for handling SVs
returns the memory when done.
|
|
|
|
|
|
| |
The invlist_destroy function was misleading, as it has changed to
just decrement the reference count, which may or may not lead to
immediate destruction
|
|
|
|
|
|
|
| |
These are static functions so no external effect. Revise the calling
sequence of two functions so that they can know enough to free
memory if appropriate of the other parameters. This hides from the
callers the need for tracking when to free memory.
|
|
|
|
|
|
| |
The inversion list is an opaque object, currently implemented as an SV.
Even if it ends up being an HV in the future, Nicholas is of the opinion
that it should be presented to the world as an SV*.
|
|
|
|
| |
This is just a refactoring. There should be no functional changes.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Extract most of the body of pp_ncmp() (numeric compare) into a separate
function, do_ncmp(), then make the following ops use it:
pp_ncmp
pp_lt
pp_le
pp_eq
pp_ne
pp_ge
pp_gt
This removes a lot of similar or duplicated code, most of which is
dedicated to handling the various combinations of IV verses UV verses NV
verses NaN.
The various ops first check for, and directly process, the simple and common
case of both args being SvIOK_notUV(), and pass the processing on to
do_ncmp() otherwise. Benchmarking seems to indicate (but with a lot of
noise) that the SvIOK_notUV case is slightly faster than before, and the
do_ncmp() branch slightly slower.
|
|
|
|
|
|
|
|
|
|
| |
Add flags param to op_lvalue, so that the caller can ask it not to
croak when encountering an unmodifiable op (upcoming).
This is in preparation for making the \$ prototype accept any lvalue.
There is no mathom, as the changes that this will support
are by no means suitable for maint.
|
|
|
|
|
|
|
|
| |
There was a remaining memory leak in the new inversion lists data
structure under threading. This solves it by changing the
implementation to use a SVpPV instead of doing our own memory
management. Then the already existing code for handling SVs
returns the memory when done.
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Move body of hfreeentries()' central loop into a new function,
hfree_next_entry(); leaving hfreeentries() as a simple loop that calls
hfree_next_entry() until there are no entries left.
This will in future allow sv_clear() to free a hash iteratively rather
than recursively.
Similarly, turn hv_free_ent() into a thin wrapper around a new function,
hv_free_ent_ret(), which doesn't free HeVAL(), but rather just returns the
SV instead.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Formats are compiled down to a sequence of U32 opcodes in doparseform().
Previously the block of opcodes was stored in the buffer of SvPVX() after
the raw string by extending the buffer, and calculating the first U32 aligned
address after SvCUR(). A flag bit on the scalar was set to signal this hackery,
tested with SvCOMPILED()
The flag bit used happened to be the same as one of the two used by to signal
Boyer-Moore compiled scalars. The assumption was that no scalar can be used for
both. Unfortunately, this isn't quite true.
Given that the scalar is alway upgraded to PVMG to add PERL_MAGIC_fm magic,
to clear the cached compiled version, there's no extra memory cost in using
mg_ptr in the MAGIC struct to point directly to the block of U32 opcodes. The
test for "is there a compiled version" can switch to mg_find(..., PERL_MAGIC_fm)
returning a pointer, and the use of a flag bit abolished.
Retain SvCOMPILED() and SvCOMPILED_{on,off}() as compatibility for XS code on
CPAN - the first is always 0, the other two now no-ops.
|
|
|
|
|
|
|
|
|
|
|
| |
a1b950687051c32e added an error condition in S_doparseform() but used DIE(...)
to report it. DIE is defined as C<return Perl_die>, which acts as a hint to the
compiler about the control flow [as Perl_die() never returns], but also forces
the return type to be OP *. Whilst this is appropriate for pp functions, it's
not for S_doparseform() - a1b950687051c32e had to change the return type to OP*
and return NULL, just to appease DIE(). Hence use Perl_die() instead, remove
return statements, and remove the didn't-return-NULL (dead) code from
pp_formline.
|
|
|
|
|
|
|
|
|
|
| |
And also to_uni_fold().
The flag allows retrieving either simple or full folds.
The interface is subject to change, so these are marked experimental
and their names begin with underscore. The old versions are turned
into macros calling the new versions with the correct extra parameter.
|
|
|
|
|
|
|
| |
Code within the function doesn't assume that the parameter is non-null,
and in fact the specials are retrieved by swash_init(). Having the
parameter null just means that no specials will be retrieved in the
current call.
|
|
|
|
| |
See ticket #80626.
|
| |
|