This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
Relocatable @INC and parallel builds pretty much work now.
[perl5.git] / pod / perltodo.pod
CommitLineData
7711098a
GS
1=head1 NAME
2
3perltodo - Perl TO-DO List
4
5=head1 DESCRIPTION
e50bb9a1 6
0bdfc961
NC
7This is a list of wishes for Perl. The tasks we think are smaller or easier
8are listed first. Anyone is welcome to work on any of these, but it's a good
9idea to first contact I<perl5-porters@perl.org> to avoid duplication of
10effort. By all means contact a pumpking privately first if you prefer.
e50bb9a1 11
0bdfc961
NC
12Whilst patches to make the list shorter are most welcome, ideas to add to
13the list are also encouraged. Check the perl5-porters archives for past
14ideas, and any discussion about them. One set of archives may be found at:
e50bb9a1 15
0bdfc961 16 http://www.xray.mpe.mpg.de/mailing-lists/perl5-porters/
938c8732 17
617eabfa
NC
18What can we offer you in return? Fame, fortune, and everlasting glory? Maybe
19not, but if your patch is incorporated, then we'll add your name to the
20F<AUTHORS> file, which ships in the official distribution. How many other
21programming languages offer you 1 line of immortality?
938c8732 22
4e577f8b 23=head1 The roadmap to 5.10
938c8732 24
4e577f8b
NC
25The roadmap to 5.10 envisages feature based releases, as various items in this
26TODO are completed.
27
4e577f8b
NC
28=head2 Needed for a 5.9.4 release
29
30=over
31
32=item *
78ef48ad
RGS
33
34Review assertions. Review syntax to combine assertions. Assertions could take
35advantage of the lexical pragmas work. L</What hooks would assertions need?>
4e577f8b 36
860f190d
RGS
37=item *
38
39C<encoding::warnings> should be turned into a lexical pragma.
0d720714 40C<encoding> should, too (probably).
860f190d 41
4e577f8b
NC
42=back
43
44=head2 Needed for a 5.9.5 release
45
46=over
47
48=item *
49Implement L</_ prototype character>
50
51=item *
52Implement L</state variables>
53
54=back
55
56=head2 Needed for a 5.9.6 release
57
58Stabilisation. If all goes well, this will be the equivalent of a 5.10-beta.
e50bb9a1 59
0bdfc961 60=head1 Tasks that only need Perl knowledge
e50bb9a1 61
0bdfc961 62=head2 common test code for timed bail out
e50bb9a1 63
0bdfc961
NC
64Write portable self destruct code for tests to stop them burning CPU in
65infinite loops. This needs to avoid using alarm, as some of the tests are
66testing alarm/sleep or timers.
e50bb9a1 67
0bdfc961 68=head2 POD -> HTML conversion in the core still sucks
e50bb9a1 69
938c8732 70Which is crazy given just how simple POD purports to be, and how simple HTML
adebf063
NC
71can be. It's not actually I<as> simple as it sounds, particularly with the
72flexibility POD allows for C<=item>, but it would be good to improve the
73visual appeal of the HTML generated, and to avoid it having any validation
74errors. See also L</make HTML install work>, as the layout of installation tree
75is needed to improve the cross-linking.
938c8732 76
dc0fb092
SP
77The addition of C<Pod::Simple> and its related modules may make this task
78easier to complete.
79
aa237293
NC
80=head2 Parallel testing
81
82The core regression test suite is getting ever more comprehensive, which has
83the side effect that it takes longer to run. This isn't so good. Investigate
84whether it would be feasible to give the harness script the B<option> of
85running sets of tests in parallel. This would be useful for tests in
86F<t/op/*.t> and F<t/uni/*.t> and maybe some sets of tests in F<lib/>.
87
88Questions to answer
89
90=over 4
91
92=item 1
93
94How does screen layout work when you're running more than one test?
95
96=item 2
97
98How does the caller of test specify how many tests to run in parallel?
99
100=item 3
101
102How do setup/teardown tests identify themselves?
103
104=back
105
106Pugs already does parallel testing - can their approach be re-used?
107
0bdfc961 108=head2 Make Schwern poorer
e50bb9a1 109
0bdfc961
NC
110We should have for everything. When all the core's modules are tested,
111Schwern has promised to donate to $500 to TPF. We may need volunteers to
112hold him upside down and shake vigorously in order to actually extract the
113cash.
3958b146 114
0bdfc961 115See F<t/lib/1_compile.t> for the 3 remaining modules that need tests.
e50bb9a1 116
0bdfc961 117=head2 Improve the coverage of the core tests
e50bb9a1 118
0bdfc961
NC
119Use Devel::Cover to ascertain the core's test coverage, then add tests that
120are currently missing.
30222c0f 121
0bdfc961 122=head2 test B
e50bb9a1 123
0bdfc961 124A full test suite for the B module would be nice.
e50bb9a1 125
0bdfc961 126=head2 A decent benchmark
e50bb9a1 127
617eabfa 128C<perlbench> seems impervious to any recent changes made to the perl core. It
0bdfc961
NC
129would be useful to have a reasonable general benchmarking suite that roughly
130represented what current perl programs do, and measurably reported whether
131tweaks to the core improve, degrade or don't really affect performance, to
132guide people attempting to optimise the guts of perl. Gisle would welcome
133new tests for perlbench.
6168cf99 134
0bdfc961 135=head2 fix tainting bugs
6168cf99 136
0bdfc961
NC
137Fix the bugs revealed by running the test suite with the C<-t> switch (via
138C<make test.taintwarn>).
e50bb9a1 139
0bdfc961 140=head2 Dual life everything
e50bb9a1 141
0bdfc961
NC
142As part of the "dists" plan, anything that doesn't belong in the smallest perl
143distribution needs to be dual lifed. Anything else can be too. Figure out what
144changes would be needed to package that module and its tests up for CPAN, and
145do so. Test it with older perl releases, and fix the problems you find.
e50bb9a1 146
0bdfc961 147=head2 Improving C<threads::shared>
722d2a37 148
0bdfc961
NC
149Investigate whether C<threads::shared> could share aggregates properly with
150only Perl level changes to shared.pm
722d2a37 151
0bdfc961 152=head2 POSIX memory footprint
e50bb9a1 153
0bdfc961
NC
154Ilya observed that use POSIX; eats memory like there's no tomorrow, and at
155various times worked to cut it down. There is probably still fat to cut out -
156for example POSIX passes Exporter some very memory hungry data structures.
e50bb9a1 157
eed36644
NC
158=head2 embed.pl/makedef.pl
159
160There is a script F<embed.pl> that generates several header files to prefix
161all of Perl's symbols in a consistent way, to provide some semblance of
162namespace support in C<C>. Functions are declared in F<embed.fnc>, variables
163in F<interpvar.h> and F<thrdvar.h>. Quite a few of the functions and variables
164are conditionally declared there, using C<#ifdef>. However, F<embed.pl>
165doesn't understand the C macros, so the rules about which symbols are present
166when is duplicated in F<makedef.pl>. Writing things twice is bad, m'kay.
167It would be good to teach C<embed.pl> to understand the conditional
168compilation, and hence remove the duplication, and the mistakes it has caused.
e50bb9a1 169
e50bb9a1 170
e50bb9a1 171
e50bb9a1 172
adebf063 173
adebf063 174
0bdfc961 175=head1 Tasks that need a little sysadmin-type knowledge
e50bb9a1 176
0bdfc961
NC
177Or if you prefer, tasks that you would learn from, and broaden your skills
178base...
e50bb9a1 179
cd793d32 180=head2 make HTML install work
e50bb9a1 181
adebf063
NC
182There is an C<installhtml> target in the Makefile. It's marked as
183"experimental". It would be good to get this tested, make it work reliably, and
184remove the "experimental" tag. This would include
185
186=over 4
187
188=item 1
189
190Checking that cross linking between various parts of the documentation works.
191In particular that links work between the modules (files with POD in F<lib/>)
192and the core documentation (files in F<pod/>)
193
194=item 2
195
617eabfa
NC
196Work out how to split C<perlfunc> into chunks, preferably one per function
197group, preferably with general case code that could be used elsewhere.
198Challenges here are correctly identifying the groups of functions that go
199together, and making the right named external cross-links point to the right
200page. Things to be aware of are C<-X>, groups such as C<getpwnam> to
201C<endservent>, two or more C<=items> giving the different parameter lists, such
202as
adebf063
NC
203
204 =item substr EXPR,OFFSET,LENGTH,REPLACEMENT
205
206 =item substr EXPR,OFFSET,LENGTH
207
208 =item substr EXPR,OFFSET
209
210and different parameter lists having different meanings. (eg C<select>)
211
212=back
3a89a73c 213
0bdfc961
NC
214=head2 compressed man pages
215
216Be able to install them. This would probably need a configure test to see how
217the system does compressed man pages (same directory/different directory?
218same filename/different filename), as well as tweaking the F<installman> script
219to compress as necessary.
220
30222c0f
NC
221=head2 Add a code coverage target to the Makefile
222
223Make it easy for anyone to run Devel::Cover on the core's tests. The steps
224to do this manually are roughly
225
226=over 4
227
228=item *
229
230do a normal C<Configure>, but include Devel::Cover as a module to install
231(see F<INSTALL> for how to do this)
232
233=item *
234
235 make perl
236
237=item *
238
239 cd t; HARNESS_PERL_SWITCHES=-MDevel::Cover ./perl -I../lib harness
240
241=item *
242
243Process the resulting Devel::Cover database
244
245=back
246
247This just give you the coverage of the F<.pm>s. To also get the C level
248coverage you need to
249
250=over 4
251
252=item *
253
254Additionally tell C<Configure> to use the appropriate C compiler flags for
255C<gcov>
256
257=item *
258
259 make perl.gcov
260
261(instead of C<make perl>)
262
263=item *
264
265After running the tests run C<gcov> to generate all the F<.gcov> files.
266(Including down in the subdirectories of F<ext/>
267
268=item *
269
270(From the top level perl directory) run C<gcov2perl> on all the C<.gcov> files
271to get their stats into the cover_db directory.
272
273=item *
274
275Then process the Devel::Cover database
276
277=back
278
279It would be good to add a single switch to C<Configure> to specify that you
280wanted to perform perl level coverage, and another to specify C level
281coverage, and have C<Configure> and the F<Makefile> do all the right things
282automatically.
283
0bdfc961
NC
284=head2 Make Config.pm cope with differences between build and installed perl
285
286Quite often vendors ship a perl binary compiled with their (pay-for)
287compilers. People install a free compiler, such as gcc. To work out how to
288build extensions, Perl interrogates C<%Config>, so in this situation
289C<%Config> describes compilers that aren't there, and extension building
290fails. This forces people into choosing between re-compiling perl themselves
291using the compiler they have, or only using modules that the vendor ships.
292
293It would be good to find a way teach C<Config.pm> about the installation setup,
294possibly involving probing at install time or later, so that the C<%Config> in
295a binary distribution better describes the installed machine, when the
296installed machine differs from the build machine in some significant way.
297
728f4ecd
NC
298=head2 linker specification files
299
300Some platforms mandate that you provide a list of a shared library's external
301symbols to the linker, so the core already has the infrastructure in place to
302do this for generating shared perl libraries. My understanding is that the
303GNU toolchain can accept an optional linker specification file, and restrict
304visibility just to symbols declared in that file. It would be good to extend
305F<makedef.pl> to support this format, and to provide a means within
306C<Configure> to enable it. This would allow Unix users to test that the
307export list is correct, and to build a perl that does not pollute the global
308namespace with private symbols.
309
8523e164 310
0bdfc961
NC
311
312
313=head1 Tasks that need a little C knowledge
314
315These tasks would need a little C knowledge, but don't need any specific
316background or experience with XS, or how the Perl interpreter works
317
318=head2 Make it clear from -v if this is the exact official release
89007cb3 319
617eabfa
NC
320Currently perl from C<p4>/C<rsync> ships with a F<patchlevel.h> file that
321usually defines one local patch, of the form "MAINT12345" or "RC1". The output
322of perl -v doesn't report that a perl isn't an official release, and this
89007cb3 323information can get lost in bugs reports. Because of this, the minor version
fa11829f 324isn't bumped up until RC time, to minimise the possibility of versions of perl
89007cb3
NC
325escaping that believe themselves to be newer than they actually are.
326
327It would be useful to find an elegant way to have the "this is an interim
328maintenance release" or "this is a release candidate" in the terse -v output,
329and have it so that it's easy for the pumpking to remove this just as the
330release tarball is rolled up. This way the version pulled out of rsync would
331always say "I'm a development release" and it would be safe to bump the
332reported minor version as soon as a release ships, which would aid perl
333developers.
334
0bdfc961
NC
335This task is really about thinking of an elegant way to arrange the C source
336such that it's trivial for the Pumpking to flag "this is an official release"
337when making a tarball, yet leave the default source saying "I'm not the
338official release".
339
0f788cd2
NC
340=head2 Ordering of "global" variables.
341
342F<thrdvar.h> and F<intrpvarh> define the "global" variables that need to be
343per-thread under ithreads, where the variables are actually elements in a
344structure. As C dictates, the variables must be laid out in order of
345declaration. There is a comment
346C</* Important ones in the first cache line (if alignment is done right) */>
347which implies that at some point in the past the ordering was carefully chosen
348(at least in part). However, it's clear that the ordering is less than perfect,
349as currently there are things such as 7 C<bool>s in a row, then something
350typically requiring 4 byte alignment, and then an odd C<bool> later on.
351(C<bool>s are typically defined as C<char>s). So it would be good for someone
352to review the ordering of the variables, to see how much alignment padding can
353be removed.
354
d7939546
NC
355It's also worth checking that all variables are actually used. Perl 5.8.0
356shipped with C<PL_nrs> still defined in F<thrdvar.h>, despite it being unused
357since a change over a year earlier. Had this been spotted before release, it
358could have been removed, but now it has to remain in the 5.8.x releases to
359keep the structure the same size, to retain binary compatibility.
360
c1ab7b38
NC
361It's probably worth checking if all need to be the types they are. For example
362
363 PERLVAR(Ierror_count, I32) /* how many errors so far, max 10 */
364
365might work as well if stored in a signed (or unsigned) 8 bit value, if the
366comment is accurate. C<PL_multi_open> and C<PL_multi_close> can probably
367become C<char>s. Finding variables to downsize coupled with rearrangement
368could shrink the interpreter structure; a size saving which is multiplied by
369the number of threads running.
370
62403a3c
NC
371=head2 am I hot or not?
372
373The idea of F<pp_hot.c> is that it contains the I<hot> ops, the ops that are
374most commonly used. The idea is that by grouping them, their object code will
375be adjacent in the executable, so they have a greater chance of already being
376in the CPU cache (or swapped in) due to being near another op already in use.
377
378Except that it's not clear if these really are the most commonly used ops. So
379anyone feeling like exercising their skill with coverage and profiling tools
380might want to determine what ops I<really> are the most commonly used. And in
381turn suggest evictions and promotions to achieve a better F<pp_hot.c>.
382
98fed0ad
NC
383=head2 Shrink struct context
384
385In F<cop.h>, we have
386
387 struct context {
388 U32 cx_type; /* what kind of context this is */
389 union {
390 struct block cx_blk;
391 struct subst cx_subst;
392 } cx_u;
393 };
394
395There are less than 256 values for C<cx_type>, and the constituent parts
396C<struct block> and C<struct subst> both contain some C<U8> and C<U16> fields,
397so it should be possible to move them to the first word, and share space with
398a C<U8> C<cx_type>, saving 1 word.
399
400=head2 Allocate OPs from arenas
401
402Currently all new OP structures are individually malloc()ed and free()d.
403All C<malloc> implementations have space overheads, and are now as fast as
404custom allocates so it would both use less memory and less CPU to allocate
405the various OP structures from arenas. The SV arena code can probably be
406re-used for this.
407
0bdfc961
NC
408
409
410
0bdfc961 411=head1 Tasks that need a knowledge of XS
e50bb9a1 412
0bdfc961
NC
413These tasks would need C knowledge, and roughly the level of knowledge of
414the perl API that comes from writing modules that use XS to interface to
415C.
416
f23930d5 417=head2 shrink C<PVBM>s
4a750395 418
35b64ab6 419By removing unused elements and careful re-ordering, the structures for C<AV>s,
f23930d5
NC
420C<HV>s, C<CV>s and C<GV>s have recently been shrunk considerably. C<PVIO>s
421probably aren't worth it, as typical programs don't use more than 8, and
422(at least) C<Filter::Util::Call> uses C<SvPVX>/C<SvCUR>/C<SvLEN> on a C<PVIO>,
423so it would mean code changes to modules on CPAN. C<PVBM>s might have some
424savings to win.
4a750395 425
0bdfc961
NC
426=head2 Implicit Latin 1 => Unicode translation
427
428Conversions from byte strings to UTF-8 currently map high bit characters
429to Unicode without translation (or, depending on how you look at it, by
430implicitly assuming that the byte strings are in Latin-1). As perl assumes
431the C locale by default, upgrading a string to UTF-8 may change the
432meaning of its contents regarding character classes, case mapping, etc.
433This should probably emit a warning (at least).
434
435This task is incremental - even a little bit of work on it will help.
e50bb9a1 436
cd793d32 437=head2 autovivification
e50bb9a1 438
cd793d32 439Make all autovivification consistent w.r.t LVALUE/RVALUE and strict/no strict;
e50bb9a1 440
0bdfc961 441This task is incremental - even a little bit of work on it will help.
e50bb9a1 442
0bdfc961 443=head2 Unicode in Filenames
e50bb9a1 444
0bdfc961
NC
445chdir, chmod, chown, chroot, exec, glob, link, lstat, mkdir, open,
446opendir, qx, readdir, readlink, rename, rmdir, stat, symlink, sysopen,
447system, truncate, unlink, utime, -X. All these could potentially accept
448Unicode filenames either as input or output (and in the case of system
449and qx Unicode in general, as input or output to/from the shell).
450Whether a filesystem - an operating system pair understands Unicode in
451filenames varies.
e50bb9a1 452
0bdfc961
NC
453Known combinations that have some level of understanding include
454Microsoft NTFS, Apple HFS+ (In Mac OS 9 and X) and Apple UFS (in Mac
455OS X), NFS v4 is rumored to be Unicode, and of course Plan 9. How to
456create Unicode filenames, what forms of Unicode are accepted and used
457(UCS-2, UTF-16, UTF-8), what (if any) is the normalization form used,
458and so on, varies. Finding the right level of interfacing to Perl
459requires some thought. Remember that an OS does not implicate a
460filesystem.
e50bb9a1 461
0bdfc961
NC
462(The Windows -C command flag "wide API support" has been at least
463temporarily retired in 5.8.1, and the -C has been repurposed, see
464L<perlrun>.)
969e704b 465
0bdfc961 466=head2 Unicode in %ENV
969e704b 467
0bdfc961 468Currently the %ENV entries are always byte strings.
e50bb9a1 469
0bdfc961 470=head2 use less 'memory'
e50bb9a1 471
0bdfc961
NC
472Investigate trade offs to switch out perl's choices on memory usage.
473Particularly perl should be able to give memory back.
e50bb9a1 474
0bdfc961 475This task is incremental - even a little bit of work on it will help.
0abe3f7c 476
0bdfc961 477=head2 Re-implement C<:unique> in a way that is actually thread-safe
0abe3f7c 478
0bdfc961
NC
479The old implementation made bad assumptions on several levels. A good 90%
480solution might be just to make C<:unique> work to share the string buffer
481of SvPVs. That way large constant strings can be shared between ithreads,
482such as the configuration information in F<Config>.
0abe3f7c 483
0bdfc961 484=head2 Make tainting consistent
0abe3f7c 485
0bdfc961
NC
486Tainting would be easier to use if it didn't take documented shortcuts and
487allow taint to "leak" everywhere within an expression.
0abe3f7c 488
0bdfc961 489=head2 readpipe(LIST)
0abe3f7c 490
0bdfc961
NC
491system() accepts a LIST syntax (and a PROGRAM LIST syntax) to avoid
492running a shell. readpipe() (the function behind qx//) could be similarly
493extended.
0abe3f7c 494
e50bb9a1 495
e50bb9a1 496
e50bb9a1 497
f86a8bc5 498
0bdfc961 499=head1 Tasks that need a knowledge of the interpreter
3298bd4d 500
0bdfc961
NC
501These tasks would need C knowledge, and knowledge of how the interpreter works,
502or a willingness to learn.
3298bd4d 503
0bdfc961
NC
504=head2 lexical pragmas
505
78ef48ad
RGS
506Document the new support for lexical pragmas in 5.9.3 and how %^H works.
507Maybe C<re>, C<encoding>, maybe other pragmas could be made lexical.
0562c0e3 508
d10fc472 509=head2 Attach/detach debugger from running program
1626a787 510
cd793d32
NC
511The old perltodo notes "With C<gdb>, you can attach the debugger to a running
512program if you pass the process ID. It would be good to do this with the Perl
0bdfc961
NC
513debugger on a running Perl program, although I'm not sure how it would be
514done." ssh and screen do this with named pipes in /tmp. Maybe we can too.
1626a787 515
0bdfc961
NC
516=head2 LVALUE functions for lists
517
518The old perltodo notes that lvalue functions don't work for list or hash
519slices. This would be good to fix.
520
521=head2 LVALUE functions in the debugger
522
523The old perltodo notes that lvalue functions don't work in the debugger. This
524would be good to fix.
525
526=head2 _ prototype character
527
528Study the possibility of adding a new prototype character, C<_>, meaning
529"this argument defaults to $_".
530
4e577f8b
NC
531=head2 state variables
532
533C<my $foo if 0;> is deprecated, and should be replaced with
534C<state $x = "initial value\n";> the syntax from Perl 6.
535
0bdfc961
NC
536=head2 regexp optimiser optional
537
538The regexp optimiser is not optional. It should configurable to be, to allow
539its performance to be measured, and its bugs to be easily demonstrated.
540
541=head2 UNITCHECK
542
543Introduce a new special block, UNITCHECK, which is run at the end of a
544compilation unit (module, file, eval(STRING) block). This will correspond to
545the Perl 6 CHECK. Perl 5's CHECK cannot be changed or removed because the
546O.pm/B.pm backend framework depends on it.
547
548=head2 optional optimizer
549
550Make the peephole optimizer optional. Currently it performs two tasks as
551it walks the optree - genuine peephole optimisations, and necessary fixups of
552ops. It would be good to find an efficient way to switch out the
553optimisations whilst keeping the fixups.
554
555=head2 You WANT *how* many
556
557Currently contexts are void, scalar and list. split has a special mechanism in
558place to pass in the number of return values wanted. It would be useful to
559have a general mechanism for this, backwards compatible and little speed hit.
560This would allow proposals such as short circuiting sort to be implemented
561as a module on CPAN.
562
563=head2 lexical aliases
564
565Allow lexical aliases (maybe via the syntax C<my \$alias = \$foo>.
566
567=head2 entersub XS vs Perl
568
569At the moment pp_entersub is huge, and has code to deal with entering both
570perl and XS subroutines. Subroutine implementations rarely change between
571perl and XS at run time, so investigate using 2 ops to enter subs (one for
572XS, one for perl) and swap between if a sub is redefined.
2810d901
NC
573
574=head2 Self ties
575
576self ties are currently illegal because they caused too many segfaults. Maybe
577the causes of these could be tracked down and self-ties on all types re-
578instated.
0bdfc961
NC
579
580=head2 Optimize away @_
581
582The old perltodo notes "Look at the "reification" code in C<av.c>".
583
0bdfc961
NC
584=head2 What hooks would assertions need?
585
586Assertions are in the core, and work. However, assertions needed to be added
587as a core patch, rather than an XS module in ext, or a CPAN module, because
588the core has no hooks in the necessary places. It would be useful to
589investigate what hooks would need to be added to make it possible to provide
590the full assertion support from a CPAN module, so that we aren't constraining
591the imagination of future CPAN authors.
592
593
594
595
596
0bdfc961
NC
597=head1 Big projects
598
599Tasks that will get your name mentioned in the description of the "Highlights
600of 5.10"
601
602=head2 make ithreads more robust
603
4e577f8b 604Generally make ithreads more robust. See also L</iCOW>
0bdfc961
NC
605
606This task is incremental - even a little bit of work on it will help, and
607will be greatly appreciated.
608
6c047da7
YST
609One bit would be to write the missing code in sv.c:Perl_dirp_dup.
610
0bdfc961
NC
611=head2 iCOW
612
613Sarathy and Arthur have a proposal for an improved Copy On Write which
614specifically will be able to COW new ithreads. If this can be implemented
615it would be a good thing.
616
617=head2 (?{...}) closures in regexps
618
619Fix (or rewrite) the implementation of the C</(?{...})/> closures.
620
621=head2 A re-entrant regexp engine
622
623This will allow the use of a regex from inside (?{ }), (??{ }) and
624(?(?{ })|) constructs.