| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| |
|
|
| |
Add `my` to synopsis.
|
| |
|
|
|
|
|
| |
This reverts commit d62feba66bf43f35d092bb026694f927e9f94d38.
As explained in its commit message. It adds some comments to point out
that the commit exists, for the curious.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Several of the POSIXA classes are a single range on ASCII platforms, and
[:digit:] is a single range on both ASCII and EBCDIC. This regnode was
designed to replace the POSIXA regnode for such classes to get a bit of
performance by not needing to do an array lookup. Instead it encodes
some bits in the flags field that with shifting and masking get the
right values for the single range's bounds for any such node.
However, performance tests conducted by Sergey Aleynikov showed this was
actually slower than what it intended to replace. Rather than
completely drop this work, I'm adding it to blead, and immediately
reverting it, so that should parts of it ever become useful, it would be
available.
A few tests fail; those are skipped for the purposes of this commit so
that it doesn't interfere with bisecting.
The code also isn't completely commented.
One could add a regnode for each posix class it was decided should have
the expected performance boost. But regnodes are a finite resource, and
the boost is probably not large enough to justify doing so.
|
| | |
|
| |
|
|
| |
Bump $CARP::VERSION and $Carp::Heavy::VERSION
|
| |
|
|
|
|
|
|
|
|
| |
- Data::Dumper::Dump() MUST be called as class method
- indent example for code markup
Also added myself to AUTHORS as requested:
Ran './Porting/checkAUTHORS.pl --update --from=v5.30.0' and
fixed the entry by hand because of broken Unicode chars.
|
| | |
|
| |
|
|
|
|
|
|
| |
Carp's CARP_NOT variable is meant to have package names. caller in list
context returns the calling file and line in addition to the package
name.
Enforce scalar context on the call to caller to fix this.
|
| |
|
|
|
|
|
|
| |
There are a number of files excluded using gitignore rules that are
included in the repository. This can lead to confusion if something
other than git tries to read the ignore files.
Add rules to the gitignore files so that these files won't be ignored.
|
| | |
|
| |
|
|
|
|
|
|
|
|
|
|
|
| |
rt52610_crash.t was introduced by commit
4764858cb80e76fdba33cc1b3be8fcdef26df754, and is predicated on
the mythical effectiveness of that commit's code change in avoiding
stack-not-refcounted crashes, an effectiveness also expressed in that
commit's very inaccurate commit message. In fact the code change will
avoid *some* crashes, but cannot guarantee to avoid crashing in any
particular situation of the kind that it targets. It is therefore not
possible to have a test for it avoiding a stack-not-refcounted crash,
with any expectation that the test would reliably pass or even reliably
avoid crashing. rt52610_crash.t must therefore be deleted.
|
| |
|
|
|
|
|
|
|
|
|
| |
Using ${^LAST_FH}, available in 5.18, in faster than using eval/die
to get the last file handle.
Also, the eval/die method is not going to produce anything if $. is 0
so skip that entire block if $. is false (by removing the defined check).
(Not significant enough for perldelta. carp("zonk") gets faster by
only 5% or so.)
|
| |
|
|
| |
It uses IPC::Open3, which uses fork(), which doesn't exist.
|
| | |
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Commit 915a6810d added a UNIVERSAL::isa check to format_arg (used by
longmess, which generates stack traces) to see whether an argument is
blessed before trying CARP_TRACE, to speed things up when the argu-
ment is not blessed.
Because this would cause infinite recursion when the UNIVERSAL::isa
module is loaded, a check was put in place to avoid this problem. But
the check was a run-time check, and so the speed-up was minimal.
If we move the check to compile time (and save the original
&UNIVERSAL::isa in case the module gets loaded later), then the speed-
up is signifant. That is what this patch does.
Before this patch, the following one-liner runs on my machine in 6
seconds on average:
$ ./perl -MCarp -Ilib -e 'sub f { my $c = shift; if ($c == 100) { Carp::longmess() } else { f($c+1,{}) } } f(0,{}) for 1..500'
If I disable the isa check (just to see how much it was speeding
things up), it averages 6.5 seconds, not much of a difference.
If I move the $UNIVERSAL::isa::VERSION safety check to compile time
instead of run time, I can reduce the time to 4.9 seconds.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Carp needs to avoid loading modules while reporting errors, because
it may be invoked via $SIG{__DIE__} after a syntax error, when BEGIN
blocks are forbidden.
Before this commit (as of v5.27.8-360-gc99363a) it was doing just that
for reference arguments within stack traces.
That means we either need to load overload.pm at start-up so that
overload::StrVal is already available, or avoid overload::StrVal
altogether.
It turns out that various versions of overload::StrVal have
their own problems that prevent Carp from using them (out-
lined in the comments added to Carp.pm and also described at
<https://rt.perl.org/Ticket/Display.html?id=132902#txn-1535564>).
So we now follow two approaches: If overloading.pm is available, use
that; otherwise, use a hideous workaround inspired by ancient imple-
entations of overload::StrVal and Scalar::Util::blessed, while avoid-
ing the bugs in those old versions.
|
| |
|
|
|
|
|
|
| |
If a module has its own ‘can’ (or even UNIVERSAL::can) implementation,
it may impede Carp’s use of ->can to detect overloading. Instead,
use UNIVERSAL::can directly, or, in the presence of an override,
use overload::mycan. Don’t use overload::Overloaded, since old
versions of overload call ->can.
|
| | |
|
| | |
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
In 4764858cb80e76fdba33cc1b3be8fcdef26df754 I added code that
eval's the argument extraction from DB::args() (code written by
both Pali and myself independently).
Unfortunately the commit message for that commit was somewhat
misleading, making it sound like the patch was a complete fix for
the underlying problem of stack-not-refcounted bugs, when in fact
it is actually a workaround, and a not-universally popular one
either due to its incompleteness. For more details of why it is
not popular see Zeframs commentary in
https://rt.perl.org/Public/Bug/Display.html?id=131046
Despite the concerns expressed by Zefram when considered from the
POV of a large enterprise using Carp to trap exceptions from
application code, having Carp throw its own exceptions about
stack-not-refcounted bugs and thus hiding the applications error
data is a significant issue.
So in 4764858cb80e76fdba33cc1b3be8fcdef26df754 we use eval to
handle those cases where Perl /can/ detect a stack-not-refcounted
error, and thus preserve as much of the applications error data
as possible.
This commit adds a comment to Carp which spells all of this out.
|
| |
|
|
|
|
|
|
|
|
|
|
| |
We use $obj->can("((") to see if a package has enabled overloads,
however in some cases can() is overridden with a custom implementation,
which may be unaware of these methods and then call Carp itself. This
then results in an infinite loop. An example is Class::Std v0.013.
While technically this is a bug in whatever overrides UNIVERSAL::can(),
Carp is so prolific in its use that we might as well treat it as a Carp
bug also and avoid the problem outright.
See also: https://github.com/chorny/Class-Std/pull/2
|
| |
|
|
|
|
|
|
| |
This reverts commit f61fbe7bb36915d6a598df82ede2511d261e41d3.
After further research it appears that this does NOT break .pmc's
The docs on .pmc's leave much to be desired.
|
| |
|
|
|
|
|
|
| |
This reverts commit 7bb6e12450c58b7094ccdfa7025fa495d3996bcf.
This patch breaks the behavior of .pmc's. :-(
Until we decide if breaking this rarely used feature is ok I am reverting.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Carp’s particular use of string eval is unnecessary in this case, and
slower than the alternative. Carp’s reason for using string eval is
to avoid the effect of bareword require vivifying a package. But that
only applies to bareword require, and not other forms of require:
$ perl -le 'print $::{"overload::"}||"nothing"; require overload'
*main::overload::
$ perl -le 'print $::{"overload::"}||"nothing"; require "overload.pm"'
nothing
Since string eval has to set up a parser and a new scope, it is much
slower that require, and quite unnecessary here.
|
| |
|
|
|
|
| |
The comments originally added by commit
c99363aa273278adcad39f32026629b700f9bbc3 for [perl #132828] were unclear
and a bit over-excited.
|
| |
|
|
|
|
|
|
| |
Commit 4764858cb80e76fdba33cc1b3be8fcdef26df754 unwisely
changed Carp's version number to an underscored one. Commit
c99363aa273278adcad39f32026629b700f9bbc3 incremented the underscored
part and copied it to Carp::Heavy. With blead being upstream for Carp,
there is no reason for it to have an underscored version number.
|
| |
|
|
|
|
|
|
|
|
|
|
|
| |
the internals don't need overload.pm to be loaded to enable overloads
which means that Carp needs to defend against overload without checking
if overload.pm is loaded either.
One odd thing about this change is that if I remove the "eval" that
wraps the "require" then we fail tests in dist/Carp/t/vivify_stash.t
which defies expectation as the require is never actually executed
from that code.
This patch doesn't have tests yet as it can segfault perl.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Trying to read values from array @DB::args can lead to perl fatal error
"Bizarre copy of ARRAY in scalar assignment". But missing, incomplete or
possible incorrect value in @DB::args is not a fatal error for Carp.
Carp is primary used for reporting warnings and errors from other
modules, so it should not crash perl when trying to print error message.
This patch safely iterates all elements of @DB::args array via eval { }
block and replace already freed scalars for Carp usage by string
"** argument not available anymore **".
This prevent crashing perl and allows to use Carp module. It it not a
proper fix but rather workaround for Carp module. At least it allows to
safely use Carp.
Patch amended by Yves Orton
|
| |
|
|
|
|
|
|
| |
The test added to Carp by commit 915a6810d3e3198d759f025f85d1fd6f3171dd27
for UNIVERSAL::isa being loaded had the side effect of vivifying the
UNIVERSAL::isa stash. Take more care about checking for UNIVERSAL::isa
to avoid vivifying it, as for the other checks for things in optional
modules. Fixes [perl #132788].
|
| |
|
|
|
|
| |
Previous Carp update from 915a6810d3 was merged
after 5.27.6 release... so we need to bump
the release version.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
RT #132274
This is a very minimal patch after RT discussion.
When using the CPAN version of UNIVERSAL::isa we cannot
use UNIVERSAL::isa on Carp without taking the risk of
going into one infinite loop.
As UNIVERSAL::isa on CPAN is the only one to advertise
a VERSION, we can use this value to disable the UNIVERSAL
check.
Note version bump is not required as it already occurred
since v5.27.5 release.
Signed-off-by: Nicolas R <atoomic@cpan.org>
|
| |
|
|
|
|
| |
This reverts commit 7a831b721c469aeccfe1110a2d177dd115d5998d. It was
buggy and mostly pointless, and following criticism on p5p it should
never have been committed.
|
| |
|
|
|
|
|
| |
Avoid downgrading the string when not required.
Author: J. Nick Koston <nick@cpanel.net>
References: CPANEL-15140
|
| | |
|
| | |
|
| |
|
|
|
|
|
|
|
|
|
|
| |
This broke some time between 1.29 (perl 5.18) and 1.3301 (perl 5.20):
$ perl5.20.1 -e 'package Foo { use constant ISA => 42; Bar::f() } package Bar { use Carp; sub f { carp "tun syn" } }'
Not a GLOB reference at /usr/local/lib/perl5/5.20.1/Carp.pm line 560.
and still persisted in bleadperl (Carp 1.43) until this commit.
The code that goes poking through the symbol table needs to take into
account that not all stash elements are globs.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Commit 975fe8546427b5f6259103912b13925be148becd introduced partial EBCDIC
support to Carp, but simultaneously introduced some bugs into the module
and the tests. Multiple issues are addressed in this commit:
* The main check for whether a character needs a non-literal
representation when dumping a string or regexp argument, which used
to be a regexp character range [ -~], was expanded to an explicit
character set not using range syntax, but in the expansion the "&"
was omitted. This caused unwanted \x representation of any "&" in an
argument in a stack trace. Add the "&" back in and fix the sorting
of the character set.
* The substitute version of this check for Perls on which Carp can't
safely apply a regexp to an upgraded string, but new enough to have
utf8::native_to_unicode(), was applying that function to some fixed
codepoint values that were already Unicode codepoints. Remove those
calls, and compare the fixed codepoints directly to codepoints correctly
converted through that function.
* That version of the check, by referring to utf8::native_to_unicode()
directly in source that is always compiled, caused the utf8:: stash to
be vivified on Perl 5.6, causing havoc (and failed tests). Hide that
version of the check behind a (compile-time) string eval.
* Another version of the printability check, for EBCDIC on Perl 5.6,
treated as printable any codepoint above 0xff. Change that to correctly
treat all such codepoints as not safely printable.
* Some tests in t/arg_regexp.t which were originally about non-ASCII
characters specified in a regexp by using \x regexp syntax got changed
to use the non-ASCII characters literally at the regexp syntax level
(by interpolating them from a constructed string). Restore these to
using \x syntax, with the appropriate variability of the hex digits.
* Add a couple of "fixme" comments about parts of the EBCDIC support
that are incomplete.
* Some tests involving non-ASCII characters were later made to skip on
any Perl prior to 5.17.1. In practice they work fine on earlier Perls,
and they're fairly important. Suspect that the problem that led to
the skipping being added was dependent on the tests having been broken
as described above, so remove the skipping logic.
* Incidentally, correct a comment about the purpose of t/arg_string.t
and add a similar one to t/arg_regexp.t.
* Incidentally, add Changes entries for versions 1.41 and 1.42, which
were omitted when those changes were made.
|
| |
|
|
|
| |
I added alternates to a regex for matching a f/p number, but forgot
to put parentheses around them. So it was being ridiculously over-liberal
|
| |
|
|
|
|
|
|
|
|
| |
[perl #129954] dist/Carp/t/arg_string.t: Test fails
This test script checks that args are displayed sensibly in longmess()
output, but floating-point numbers can be displayed in various formats
depending on platform, so make the regex more forgiving.
Also add a comment to the top of the script explaining its purpose.
|
| |
|
|
| |
The 3e100 seems to have no special meaning, except being in floating point.
|
| |
|
|
| |
Nowhere did it actually describe what cluck does.
|
| |
|
|
|
|
|
|
|
| |
Commit 89988fbd2f7d8a44526a3cd9ab671b3102898bc9 added the relevant phrase for
line-oriented filehandles; this extends that to also cover chunk-oriented
ones.
The patch was supplied by Chris R Donnelly <chris.donnelly@vauto.com> on the
rt.cpan.org ticket; I have extended it to include a test.
|
| |
|
|
|
| |
This doesn't guarantee the tests pass in that release; just that it will
actually get to testing.
|
| |
|
|
|
|
|
| |
It lacked a link to the VCS repository so I had to ask rjbs about it on
the #p5p IRC channel. And one should always fix a problem twice:
* http://www.joelonsoftware.com/articles/customerservice.html .
|
| | |
|
| |
|
|
|
|
|
| |
See: https://rt.cpan.org/Public/Bug/Display.html?id=107225 . Also
discovered as a bug in perl -d by me (= Shlomi Fish) and reported
after the original report. longmess() returns "1" when called in scalar
context if passed a reference.
|
| | |
|
| | |
|
| | |
|
| | |
|
