| Commit message (Collapse) | Author | Age | Files | Lines |
| |
|
| |
|
| |
|
| |
|
|
|
|
|
|
|
|
|
|
|
| |
This warning was being generated inappropriately during some internal
operations, such as parsing a program; spotted by Tom Christiansen.
The solution is to move the check for this situation out of the common
code, and into the code where just \p{} and \P{} are handled.
As mentioned in the commit's perldelta, there remains a bug
[perl #114148], where no warning gets generated when it should
|
|
|
|
|
|
| |
Future commits will use this functionality in additional places beyond
the single one currently. It makes sense to abstract it into a
function.
|
|
|
|
|
|
|
| |
Previously, this flag was not looked at in pass 1, but future commits
will require it to be. The regnode is not actually allocated in pass1,
so set a local variable with the value, and set the flag in the regnode
after everything affecting the flag is done.
|
|
|
|
|
|
| |
This commit separates out the building up of code points that match the
posix properties (including \w, \d, \s) from those that match Unicode
properties. This will prove useful in future commits
|
|
|
|
| |
It's a little clearer to use boolean for a true/false valued variable
|
|
|
|
|
|
|
| |
This creates a function to hide some of the internal details of swashes
from the regex engine, which is the only authorized user, enforced
through #ifdefs in embed.fnc. These work closely together, but it's
best to have a clean interface.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
There have been two flavors of ANYOF nodes under /l (locale) (for
bracketed character classes). If a class didn't try to match things
like [:word:], it was smaller by 4 bytes than one that did.
A flag bit was used to indicate which size it was. By making all such
nodes the larger size, whether needed or not, that bit can be freed to
be used for other purposes.
This only affects ANYOF nodes compiled under locale rules. The hope is
to eventually git rid of these nodes anyway, by taking the suggestion of
Yves Orton to compile regular expressions using the current locale, and
automatically recompile the next time they are used after the locale
changes.
This commit is somewhat experimental, done early in the development
cycle to see if there are any breakages. There are other ways to free
up a bit, as explained in the comments. Best would be to split off
nodes that match everything outside Latin1, freeing up the
ANYOF_UNICODE_ALL bit. However, there currently would need to be two
flavors of this, one also for ANYOFV. I'm currently working to
eliminate the need for ANYOFV nodes (which aren't sufficient,
[perl #89774]), so it's easiest to wait for this work to be done before
doing the split, after which we can revert this change in order to gain
back the space, but in the meantime, this will have had the opportunity
to smoke out issues that I would like to know about.
|
| |
|
| |
|
|
|
|
|
| |
I was confused by the earlier documentation. Thanks to Leon Timmermans
for clarifying, and to Vicent Pitt for most of the wording
|
|
|
|
| |
since the number of tests is known in advance.
|
| |
|
|
|
|
|
|
|
|
|
| |
This will reduce the machine code size on Visual C Perl, by removing C stack
clean up opcodes and possible jmp opcodes after croak() and similar
functions. Perl's existing __attribute__noreturn__ macro (and therefore
GCC's __attribute__((noreturn)) ) is fundamentally incompatible with MS's
implementation for noreturn functions. win32.h already has _MSC_VER
aware code blocks, so adding more isn't a problem.
|
| |
|
|
|
|
|
| |
Cygwin fsync() is implemented in terms of the Win32 FlushFileBuffers()
API which requires GENERIC_WRITE access.
|
| |
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
In restore_magic(), which is called after any magic processing, all of
the public OK flags have been shifted into the private OK flags. Thus
the lack of an appropriate public OK flags was used to trigger both get
magic and required conversions. This scheme did not cover ROK, however,
so all properly written code had to make sure mg_get was called the right
number of times anyway. Meanwhile the private OK flags gained a second
purpose of marking converted but non-authoritative values (e.g. the IV
conversion of an NV), and the inadequate flag shift mechanic broke this
in some cases.
This patch removes the shift mechanic for magic flags, thus exposing (and
fixing) some improper usage of magic SVs in which mg_get() was not called
correctly. It also has the side effect of making magic get functions
specifically set their SVs to undef if that is desired, as the new behavior
of empty get functions is to leave the value unchanged. This is a feature,
as now get magic that does not modify its value, e.g. tainting, does not
have to be special cased.
The changes to cpan/ here are only temporary, for development only, to
keep blead working until upstream applies them (or something like them).
Thanks to Rik and Father C for review input.
|
| |
|
| |
|
|
|
|
|
|
|
| |
This was an early attempt to fix leaking of ops after syntax errors,
disabled because it was deemed to fragile. The new slab allocator
(8be227a) has solved this problem another way, so latefree(d) no
longer serves any purpose.
|
|
|
|
|
|
|
|
| |
It is not necessary for the op containing the sv containing the name
of the package to last until the end of the block. Perl_package can
free the op immediately, as the only information it needs from it it
copies to the sv in PL_curstname. The version number can be treated
the same way.
|
|
|
|
|
| |
It calls ck_fun, which applies list context to all the arguments,
since grepstart and mapstart have an L in regen/opcodes.
|
|
|
|
|
| |
The branch in question is only reached when it is called from ck_grep,
which itself calls LINKLIST on exactly the same op.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
ck_grep is a tangled mess and I am trying to simplify it, but at each
step it is not at all obvious what is happening.
grep/map with a block starts out like this:
grepstart
pushmark
null
scope/leave
block contents...
arg1
arg2
...
The if (o->op_flags & OPf_STACKED) at the top of ck_grep runs for
blocks. The firstkid var refers to the null above, and gets its
op_next set (this code was in ck_sort until recently [ck_grep used to
call ck_sort], and this particular op_next assignment was specific to
grep and map).
Later, the thing is fed through ck_fun, whose OA_CVREF case transforms
the tree into this:
grepstart
pushmark
null ← new op
null ← original null
scope/leave
block contents...
arg1
arg2
...
Then cUNOPx(cLISTOP->op_first->op_sibling)->op_first gets its op_next
set. By that point, that expression refers to the original null op,
so setting its op_next pointer to begin with seems pointless.
But it is not actually pointless. Removing that statement breaks
things. It turns out, as usual, that everything is more complicated
than it seems. Setting kid->op_next = (OP*)gwop in the block-specific
if statement after setting the first null op’s op_next pointer results
in op_next pointers like this (the newly-created grepwhile doesn’t
link to the grepstart yet at this point):
[grepwhile]
grepstart
pushmark
null -> block contents
scope/leave -> grepwhile
block contents...
arg1
arg2
...
So ck_fun’s OA_CVREF handling, when it calls LINKLIST, sees that the
original null op already has its op_next set and does nothing. With-
out it set, it copies it from the first kid, resulting in this (when
LINKLIST copies the op_next pointer out, it makes the kid it was cop-
ied from point to its sibling or its parent):
[grepwhile]
grepstart
pushmark
null
null -> grepwhile
scope/leave -> null (prev line)
block contents...
arg1
arg2
...
That nonsensical arrangement of op_next pointers basically prevents
the block from running, because the grepwhile’s other pointer, copied
from the first null’s kid, ends up pointing to the grepwhile itself.
If we also remove the kid->op_next = (OP*)gwop assignment from the
if(OPf_STACKED) block, then we end up with this after ck_fun:
[grepwhile]
grepstart
pushmark
null
null -> first op in block
scope/leave -> null (prev line)
block contents...
arg1
arg2
...
Then the op_next poiner from the first null’s kid is copied to
grepwhile’s op_other pointer, and everything works.
This also means we can remove the now-redundant LINKLIST call from the
if(OPfSTACKED) block, since ck_fun’s OA_CVREF handling also does that.
So now every vestige of the original call to ck_sort is gone.
This also means we no longer have to repeat NewOp(1101, gwop...).
|
|
|
|
|
| |
It calls ck_fun first, which does the same check, so it is
unnecessary.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Back in perl 5.000, ck_grep would call ck_sort (which was still the
case until 354dd559d99 just recently) and the latter would traverse
its way through the op_next pointers to find an op that tried to
escape the block, setting its op_next pointer to null. Then ck_grep
would traverse op_next pointers itself to find the place where ck_sort
stopped. That caused problems for grep which were fixed in 748a93069b
(perl 5.001). It was fixed by setting op_next to 0 on the first op,
so that the loop in ck_grep would not do anything. But that loop was
left there. This commit removes it.
There are also a couple of things I missed when disentangling
ck_grep and ck_sort in commit /354dd559d9. I accidentally put
if (o->op_flags & OPf_STACKED) inside if (o->op_flags & OPf_STACKED).
And the OPf_SPECIAL flag was only being set for sort’s use, so ck_grep
doesn’t need to copy that.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This code currently chases op_next pointers in the sort block to make
sure there is none that can point outside. To avoid getting stuck in
loops, it has to check for those explicitly.
Instead of nulling out any op_next pointer that points to the leave
op, we can simply turn the leave into a null op (which we already do)
and have the leave op’s op_next pointer point to nothing.
That produces the same result in the end: Execution of the sort block
is prevented from running rampant and escaping out of it, but exits
the runloop instead, returning control to pp_sort.
|
| |
|
| |
|
|
|
|
|
| |
Removed some out-of-date modules and add Math::GMPq, Math::GMPz and
Math:GMPf.
|
| |
|
|
|
|
|
|
|
|
|
| |
Originally, sort and grep/map used the same code for handling the
block’s ops.
But that didn’t quite work, so exceptions were added over time. Now
the thing is a complex sprawling mess. By folding the grep-specific
code into ck_grep, things are greatly simplified.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
When the peephole optimiser encounters a __SUB__, it looks to see
whether the current sub is clonable. If it is not, it inlines the
__SUB__ as a const op.
This works most of the time. A forward declaration will cause the sub
definition to reuse the existing stub. When that happens, the sub
visible during compilation in PL_compcv is not the sub that the
op tree will finally be attached to. But the peephole optimiser
is called after that, with PL_compcv set to the other CV (what
was a stub).
ck_sort was calling the peephole optimiser on the sort block ahead of
time. So this caused __SUB__ to point to the wrong subroutine.
By removing the CALL_PEEP call from ck_sort and adding logic to the
peephole optimiser itself to traverse the sort block (it is not in the
usual op_next chain), this bug is eliminated.
I modified the DEFER macro to work as a single statement. You don’t
want to know how much time I spent debugging the bus errors that were
occurring because if(foo) DEFER; didn’t do what I though.
It turns out that grepstart and mapstart, which also use ck_sort,
had their blocks go through the peephole optimiser twice, because
grepwhile already has special-casing in the peephole optimiser.
This also has the side-effect of making map and grep slightly more
efficient, in that they no longer execute a scope op (which is just
pp_null). By temporarily disconnecting the subtree before running the
optimiser, ck_sort was hiding a possible optimisation (skipping the
scope op).
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
It seems that under certain circumstances the optimiser handles the
MEOL operator (what $ turns into under /m), improperly including
things that follow. This results in compilation like this:
Compiling REx "( [^z] $ [^z]+ )"
Final program:
1: OPEN1 (3)
3: ANYOF[\x00-y{-\xff][{unicode}0100-INFINITY] (14)
14: MEOL (15)
15: PLUS (27)
16: ANYOF[\x00-y{-\xff][{unicode}0100-INFINITY] (0)
27: CLOSE1 (29)
29: END (0)
anchored ""$ at 2 stclass ANYOF[\x00-y{-\xff][{unicode}0100-INFINITY]
Where the '""$ at 2' is sign of the bug. The problem is that the optimiser
does not "commit" the string when it encounters an MEOL, which means that
text that follows it is included. This has probably always been wrong as
$ is a multichar pattern (it matches before an \n or including an \n). This
failure to commit then interacts with the implementation for PLUS leading to
an incorrect offset. By adding a SCAN_COMMIT() as part of the optimisers
handling of EOL constructs this problem is avoided. Note that most uses of
$ were ok due to other reasons.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This ‘Can't use an undefined value as a subroutine reference’ message
used to occur for &{+undef}, until commit 15ff848f77796 in 1997.
Then it started happening only for magic variables.
Non-magical variables would treat it the same as "", possibly dying
with ‘Undefined subroutine main:: called’, and possibly calling the
subroutine with that name.
Magical variables (&{$tied_undef}) started behaving the same way
(treating undef as "") with commit 7c75014e, in 2010.
This commit restores that message.
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This takes the pessimistic approach of skipping it for any first argu-
ment that is not a plain non-magical PV, just in case there is a 'p'
or 'P' in the stringified form.
Otherwise it scans the PV for 'p' or 'P' and skips the folding if either
is present.
Then it falls through to the usual op-filtering logic.
I nearly made ‘pack;’ crash, so I added a test to bproto.t.
|
| |
|
|
|
|
|
| |
The test is trying to test the behaviour being warned about,
so the code is right but the warning can be suppressed.
|
| |
|
|
|
|
|
| |
--lipods warns, and I don't think it is needed in this test,
so I have removed it
|
|
|
|
|
| |
This warns when void* and IV have different sizes
There is already a macro to do this conversion cleanly: PTR2IV
|
| |
|
|
|
|
|
| |
It was not attaching magic to the array, preventing subsequent changes
to the array from updating isa caches.
|
|
|
|
|
|
|
|
|
|
| |
I broke this in 5.14 with commit 6624142a.
In trying to make *ISA = *Other::ISA work, I added logic to make
@Other::ISA’s existing magic now point to *ISA’s stash. I skipped
that logic if *Other::ISA did not contain an array. But in so
doing, I inadvertently skipped the call to mro_isa_changed_in at the
same time.
|