Tim Bunce [Sat, 22 Sep 2012 14:41:56 +0000]
Use link nodes to label links.
Tim Bunce [Sat, 22 Sep 2012 14:41:10 +0000]
Fix link-to-link for op_size.
Tim Bunce [Sat, 22 Sep 2012 10:44:36 +0000]
add basic dot format output
Tim Bunce [Sat, 22 Sep 2012 10:15:17 +0000]
Control recursion threshold to enable more accurate perl_size.
Tim Bunce [Thu, 20 Sep 2012 21:02:18 +0000]
rm some old files
Tim Bunce [Thu, 20 Sep 2012 17:09:58 +0000]
Lots of progress.
Reworked attribute pipeline.
Added title concept.
Added logarea (incomplete)
Tim Bunce [Thu, 20 Sep 2012 09:32:53 +0000]
smarten up the tooltip
Tim Bunce [Wed, 19 Sep 2012 16:36:01 +0000]
basic leaves pass-thru
Tim Bunce [Wed, 19 Sep 2012 14:24:26 +0000]
basic attribute pass-thru
Tim Bunce [Wed, 19 Sep 2012 10:30:57 +0000]
A little polish
Tim Bunce [Wed, 19 Sep 2012 09:44:28 +0000]
single-child node merging working
Tim Bunce [Wed, 19 Sep 2012 08:55:47 +0000]
split transform nodes away from fetch
Tim Bunce [Tue, 18 Sep 2012 21:29:20 +0000]
drill down with basic tooltip
Tim Bunce [Mon, 17 Sep 2012 09:38:38 +0000]
on-demand treemap working, though rough
Tim Bunce [Sun, 16 Sep 2012 15:46:17 +0000]
Working(ish) on demand demo using statis data
Tim Bunce [Wed, 12 Sep 2012 20:48:16 +0000]
Working on removing links from the output path
Tim Bunce [Tue, 11 Sep 2012 22:45:05 +0000]
Rename dNPathSetNode->NPathPushNode and rework magic_size nodes.
Tim Bunce [Tue, 11 Sep 2012 22:25:59 +0000]
Generate json and initial experiments with a treemap.
Tim Bunce [Tue, 11 Sep 2012 20:42:47 +0000]
Add memnodes.pl script to process output stream.
Tim Bunce [Mon, 10 Sep 2012 21:05:51 +0000]
Add node stream file format (plain text for now)
Add workaround for a PL_strtab HeVAL == 0xC.
Tim Bunce [Mon, 10 Sep 2012 19:00:49 +0000]
Added perl_size()
Tim Bunce [Mon, 10 Sep 2012 18:59:39 +0000]
Squashed commit of all initial work on the 'name path' mechanism.
Tim Bunce [Wed, 15 Aug 2012 20:23:33 +0000]
Spelling fix to docs from gregor herrmann [CPAN #78766]
Nicholas Clark [Mon, 30 Jul 2012 15:26:56 +0000]
Bump $VERSION to 0.78_50
Nicholas Clark [Thu, 26 Jul 2012 14:39:13 +0000]
This is 0.78 - update META.yml and META.json
Nicholas Clark [Thu, 26 Jul 2012 12:26:04 +0000]
Bump $VERSION to 0.78
Nicholas Clark [Tue, 17 Jul 2012 14:47:37 +0000]
This is 0.77_51 - update META.yml and META.json
bulk 88 [Sat, 2 Jun 2012 00:41:58 +0000]
Correct the code for structured exception handling on MSVC.
Nicholas Clark [Tue, 17 Jul 2012 14:31:29 +0000]
Bump $VERSION to 0.77_51
Nicholas Clark [Sat, 11 Feb 2012 16:10:29 +0000]
This is 0.77_50 - update META.yml, and add META.json
Nicholas Clark [Fri, 10 Feb 2012 20:01:57 +0000]
Skip tests in t/magic.t that use formline on 5.8.1 and 5.8.2
Those versions have a buggy formline that can trigger an assertion failure.
Nicholas Clark [Fri, 10 Feb 2012 19:48:25 +0000]
Refactor t/globs.t to avoid the side effects of a change to strict.pm
As of blead commit
b50b20584a1bbc1a, Implement new 'use 5.xxx' plan,
use strict; will write to %^H. Move the eval that creates the subroutine
into a different scope, a new subroutine generate_glob(), so that its
outside pointer chain doesn't include gv_grew(). This avoids problems with
eval ops changing the side of the %^H copies at runtime.
Nicholas Clark [Mon, 16 May 2011 12:40:06 +0000]
Bump $VERSION to 0.77_50
Nicholas Clark [Mon, 16 May 2011 12:39:00 +0000]
This is 0.77 - update META.yml
Nicholas Clark [Thu, 12 May 2011 16:34:34 +0000]
Bump $VERSION to 0.77
Nicholas Clark [Thu, 12 May 2011 16:33:04 +0000]
This is 0.76_50 - update META.yml
Nicholas Clark [Thu, 12 May 2011 16:26:28 +0000]
Count the size of the "effective names" of a hash.
This code is somewhat cheeky, making direct structure accesses.
Nicholas Clark [Thu, 12 May 2011 14:04:17 +0000]
Count HvNAME(), the HV "aux" struct, and the mro_meta struct.
Nicholas Clark [Thu, 12 May 2011 11:28:38 +0000]
GvNAME() is shared from 5.10 onwards.
Nicholas Clark [Thu, 12 May 2011 10:57:13 +0000]
Use GvFILE_HEK(), if present, in preference to GvFILE()
Nicholas Clark [Thu, 12 May 2011 10:39:10 +0000]
Handle shared hash key scalars correctly.
Nicholas Clark [Thu, 12 May 2011 09:25:19 +0000]
Split out HEK size calculation into hek_size().
Add the shared HE overhead to the total size.
Nicholas Clark [Thu, 12 May 2011 08:54:21 +0000]
Add a comment to force emacs to use C mode.
Nicholas Clark [Thu, 12 May 2011 08:43:02 +0000]
Add a single (void **) cast to make the C code also valid as C++
Nicholas Clark [Wed, 11 May 2011 09:02:19 +0000]
Bump $VERSION to 0.76_50
Nicholas Clark [Wed, 11 May 2011 08:38:53 +0000]
This is 0.76 - update META.yml
Nicholas Clark [Wed, 11 May 2011 08:38:00 +0000]
Bump $VERSION to 0.76
Nicholas Clark [Mon, 9 May 2011 20:00:36 +0000]
This is 0.75_52 - update META.yml
Nicholas Clark [Mon, 9 May 2011 07:18:18 +0000]
Don't count PL_sv_placeholder in the size returned.
PL_sv_placeholder is a singleton, process wide, so shouldn't count as part of
the size.
Nicholas Clark [Sun, 8 May 2011 20:16:43 +0000]
sv_size() can return void once more.
All the code that relied on it returning the result of its call to check_new()
has been removed.
Nicholas Clark [Sun, 8 May 2011 17:48:33 +0000]
Reinstate 5.005_xx "support".
i.e. it compiles and passes tests.
Nicholas Clark [Sun, 8 May 2011 17:41:13 +0000]
Replace c*OPx macros with their expansions.
The expansions are 2 characters longer than the macros. The macros are not
present in 5.005.
Nicholas Clark [Sun, 8 May 2011 08:16:57 +0000]
Reinstate 5.6.x "support".
i.e. it compiles and passes tests.
Nicholas Clark [Sun, 8 May 2011 07:38:02 +0000]
Don't recurse into op_size() on op->pmnext.
It doesn't point to an OP owned by the current OP - it is used to form a linked
list that reset uses to find regexps to reset. Moreover, prior to 5.8.0 it can
end up pointing to freed memory, which results in much jollity.
[Bug
20010301.005, a.k.a. RT #5935, fixed in cb55de95c99e4650]
Nicholas Clark [Sat, 7 May 2011 18:59:24 +0000]
Reinstate weaken(), inadvertently removed by
8c394e1251fdfe38.
The test isn't testing as much as intended without it.
Nicholas Clark [Sat, 7 May 2011 10:23:01 +0000]
Magic vtables aren't freed when magic is freed, so don't count them.
They are static structures. Anything that assumes otherwise is buggy.
Nicholas Clark [Fri, 6 May 2011 19:29:18 +0000]
Correctly handle SvOOK scalars. 5.12 and later don't use SvIVX().
Nicholas Clark [Fri, 6 May 2011 19:16:19 +0000]
ppport.h defined NV if necessary, so no need to duplicate that.
Nicholas Clark [Fri, 6 May 2011 19:07:10 +0000]
Add -DPURIFY sizes to body_sizes[]
Use two macros to simplify the logic - MAYBE_OFFSET() where the normal version
subtracts a STRUCT_OFFSET(), and the purify version subtracts nothing, and
MAYBE_PURIFY() where there is not a simple relationship between the two sizes.
Nicholas Clark [Thu, 5 May 2011 16:04:36 +0000]
Use a table for SV body sizes. These incorporate the space saving post 5.8.x
For now, the sizes under -DPURIFY aren't correct.
Nicholas Clark [Thu, 5 May 2011 11:45:10 +0000]
In sv_size(), refactor the PV-derivatives to share as much code as possible.
As a side effect, PV"BM"s are now handled correctly on 5.10.0 and later.
Nicholas Clark [Thu, 5 May 2011 09:50:44 +0000]
In sv_size(), coalesce all calls to magic_size().
Nicholas Clark [Thu, 5 May 2011 09:15:51 +0000]
Bump $VERSION to 0.75_52
Nicholas Clark [Thu, 5 May 2011 09:14:23 +0000]
This is 0.75_51 - update META.yml
Nicholas Clark [Thu, 5 May 2011 09:00:07 +0000]
For other-than-gcc, generate calls to check_new(...)
Only use a static array of vtables on gcc. Some other compilers may be happy
with it, but VC certainly isn't. It's only trivial to determine gcc or not.
Nicholas Clark [Thu, 5 May 2011 08:24:30 +0000]
Bump $VERSION to 0.75_51
Nicholas Clark [Thu, 5 May 2011 08:21:20 +0000]
This is 0.75_50 - update META.yml
Nicholas Clark [Wed, 4 May 2011 20:41:45 +0000]
eval $ExtUtils::MakeMaker::VERSION to avoid warnings about _ not being numeric.
Frustratingly ExtUtils::MakeMaker->VERSION() returns the literal value of
$ExtUtils::MakeMaker::VERSION so it's just as "not numeric" as reading the
scalar directly.
Nicholas Clark [Wed, 4 May 2011 20:41:00 +0000]
Re-indent Makefile.PL consistently.
Nicholas Clark [Wed, 4 May 2011 20:37:24 +0000]
Simplify magic_size() by knowing that check_new() returns false for NULL.
Nicholas Clark [Wed, 4 May 2011 20:24:58 +0000]
Add the size of mg_len if mg_ptr is non-NULL.
With special cases for UTF-8 caching magic, which abuses mg_len, and for
mg_len == HEf_SVKEY
Nicholas Clark [Wed, 4 May 2011 19:18:39 +0000]
Follow mg_obj, which points to an SV.
Nicholas Clark [Wed, 4 May 2011 19:14:43 +0000]
Add tests for sizing magic.
Nicholas Clark [Wed, 4 May 2011 18:41:54 +0000]
The core's magic vtables are global constants, so aren't part of the size.
Parse perl.h to find the names of the vtables' names, and generate code to add
them all in new_state() as "seen".
Nicholas Clark [Wed, 4 May 2011 16:10:22 +0000]
Bump $VERSION to 0.75_50
Nicholas Clark [Wed, 4 May 2011 15:23:39 +0000]
This is 0.75 - update META.yml
Nicholas Clark [Wed, 4 May 2011 15:06:21 +0000]
Bump $VERSION to 0.75
Nicholas Clark [Wed, 4 May 2011 15:00:30 +0000]
This is 0.74_53 - update META.yml
Nicholas Clark [Sun, 1 May 2011 21:24:07 +0000]
Skip the aelemfast size comparison test for perls which predate the code change.
It was merged to maint-5.8 between 5.8.3 and 5.8.4, so 5.8.4 and later have it.
For 5.8.3 and earlier, the assumptions the test makes about sizes (based on op
structure) are bogus.
Nicholas Clark [Sun, 1 May 2011 21:02:59 +0000]
Skip GvFILE on ithreaded 5.8.8 and earlier, as it can point to freed memory.
Probably better to slightly under report sizes, than report something random or
SEGV.
Nicholas Clark [Sun, 1 May 2011 20:06:51 +0000]
Set the pointer alignment (in bits) via a C macro generated by the Makefile.PL
The previous approach of using sizeof(void *) is flat out wrong on a 64 bit
system. The right answer is 3. It gives 4. Which isn't a problem once the
pointer rotation code in check_new() isn't buggy, but isn't as efficient as
desired.
Nicholas Clark [Sun, 1 May 2011 19:56:48 +0000]
Fix typo error in bit-vector tracking mechanism, causing bogus "seen before"s.
On 64 bit platforms which allocate on 8 byte alignment (rather than 16), a low
pointer bit could get lost, resulting in new pointers being considered already
"seen". Joy, hilarity, failing tests and different results running under gdb.
"Found" by BinGOs' smoker, fixed on spectre.mongueurs.net. The help was
appreciated.
Nicholas Clark [Sun, 1 May 2011 19:51:58 +0000]
Bump $VERSION to 0.74_53
Nicholas Clark [Mon, 25 Apr 2011 08:50:28 +0000]
This is 0.74_52 - update META.yml
Nicholas Clark [Sun, 24 Apr 2011 21:07:24 +0000]
SvOK() && SvROK() is a tautology - SvROK() true has always implied SvOK() true.
So no need to check SvOK() first on 5.10 or earlier.
Nicholas Clark [Sun, 24 Apr 2011 20:59:00 +0000]
total_size() and size() can be XS ALIASes, now all the logic is in sv_size()
Nicholas Clark [Sun, 24 Apr 2011 20:51:52 +0000]
Move iteration over array values from total_size() to sv_size()
For now this requires some bodgery over when array values should be recursed
into. Arrays presented to total_size() should be properly recursed into.
However, historically the inner arrays of padlists were never traversed, which
means there is no code to deal with the special case rules for them. So to avoid
the perfect being the enemy of the good, defer writing that code, by skipping
the inner arrays of padlists for now.
Nicholas Clark [Sun, 24 Apr 2011 20:39:18 +0000]
Handle sizes for PVGVs in sv_size() only.
Previously the logic was duplicated in total_size().
Nicholas Clark [Sun, 24 Apr 2011 15:32:40 +0000]
Move iteration over hash values from total_size() to sv_size()
For now this requires some bodgery over when hash values should be recursed
into. Hashes presented to total_size() should be properly recursed into, else
the size of symbol tables isn't correctly reported. However, the back-link
from a CV to its stash shouldn't be traversed, as shouldn't GVs referenced from
OPs, else we run the risk of passing through %::, and adding the size of all
the symbol tables (and subroutines) to the size of a subroutine reference we
were directly passed. Whilst this might be useful in some cases, it isn't useful
as a default, and it isn't even vaguely consistent with the sizes total_size()
historically reported.
Nicholas Clark [Sun, 24 Apr 2011 19:45:25 +0000]
Fix potential SEGVs for PVBMs on 5.10.0 and later.
5.10.0 stores PVBMs as PVGVs. Hence not all PVGVs are typeglobs, and XS code
should check isGV_with_GP() is true before performing typeglob lookups.
The PVGV code currently ignores SvCUR()/SvLEN(), so for now the size of PVBMs
is under-reported on 5.10.0 and later.
Nicholas Clark [Sun, 24 Apr 2011 17:43:32 +0000]
The optree holds pointers to GVs under multiplicity. It's ithreads that differs.
Hence the tests that are conditional on the different behaviours should be
{unthreaded,multiplicity} vs {ithreads}, not
{unthreaded} vs {multiplicity,ithreads}
Nicholas Clark [Sat, 23 Apr 2011 20:55:48 +0000]
In op_size, add break statement missing for OPc_PVOP.
This has been missing since op_size was added in 0.60. Without it, PVOPs would
run on into the OPc_LOOP case, which would likely SEGV as it would consider
the char *op_pv as OP *op_first, and non-existent op_last, op_redoop, op_nextop
and op_lastop.
Nicholas Clark [Sat, 23 Apr 2011 20:07:10 +0000]
Fix potential SEGVs for OP_AELEMFAST on a lexical.
OP_AELEMFAST is wonderfully special, in that it can be used to replace both
OP_GV and OP_PADAV. The latter is only a baseop, whereas OP_AELEMFAST is a
flagged as a padop/svop. baseops are smaller, and (therefore) don't have memory
allocated where op_sv "should" be.
Nicholas Clark [Sat, 23 Apr 2011 17:48:36 +0000]
Add tests for subroutines and subroutine references.
Nicholas Clark [Sun, 24 Apr 2011 21:02:44 +0000]
Bump $VERSION to 0.74_52
Nicholas Clark [Fri, 22 Apr 2011 16:43:36 +0000]
This is 0.74_51 - update META.yml
Nicholas Clark [Fri, 22 Apr 2011 16:30:03 +0000]
Fix classic off-by-one error - the answer is strlen() + 1, not just strlen()
This has been in the code since year dot, and no-one has spotted that the
returned sizes have been under-reporting.
Nicholas Clark [Thu, 21 Apr 2011 11:19:08 +0000]
Move the check_new() test to the start of sv_size()
For now, this requires sv_size() to return new-or-not, but it prepares the way
to move all structure recursion into sv_size(). Currently it's in two places -
sv_size() and total_size().
This resolves two long-standing bugs in typeglobs - total_size() was double
counting entries in typeglobs, and sv_size() was double-counting the PVGV size
if GvEGV() looped back.
Nicholas Clark [Fri, 22 Apr 2011 11:26:33 +0000]
Add many tests for typeglobs.
These reveal several bugs.
Nicholas Clark [Mon, 18 Apr 2011 10:21:53 +0000]
Don't count PL_sv_{undef,no,yes} in the size returned
PL_sv_{undef,no,yes} are singletons, interpreter wide, so shouldn't count as
part of the size.