source: trunk/essentials/dev-lang/perl/Changes5.001@ 3296

-------------
Version 5.001
-------------

Nearly all the changes for 5.001 were bug fixes of one variety or another,
so here's the bug list, along with the "resolution" for each of them.  If
you wish to correspond about any of them, please include the bug number.

There were a few that can be construed as enhancements:
    NETaa13059: now warns of use of \1 where $1 is necessary.
    NETaa13512: added $SIG{__WARN__} and $SIG{__DIE__} hooks
    NETaa13520: added closures
    NETaa13530: scalar keys now resets hash iterator
    NETaa13641: added Tim's fancy new import whizbangers
    NETaa13710: cryptswitch needed to be more "useable"
    NETaa13716: Carp now allows multiple packages to be skipped out of
    NETaa13716: now counts imported routines as "defined" for redef warnings
    (and, of course, much of the stuff from the perl5-porters)

NETaa12974: README incorrectly said it was a pre-release.
Files patched: README

NETaa13033: goto pushed a bogus scope on the context stack.
From: Steve Vinoski
Files patched: pp_ctl.c
 The goto operator pushed an extra bogus scope onto the context stack.  (This
 often didn't matter, since many things pop extra unrecognized scopes off.)

NETaa13034: tried to get valid pointer from undef.
From: Castor Fu
Also:  Achille Hui, the Day Dreamer 
Also: Eric Arnold
Files patched: pp_sys.c
 Now treats undef specially, and calls SvPV_force on any non-numeric scalar
 value to get a real pointer to somewhere.

NETaa13035: included package info with filehandles.
From: Jack Shirazi - BIU
Files patched: pp_hot.c pp_sys.c
 Now passes a glob to filehandle methods to keep the package info intact.

NETaa13048: didn't give strict vars message on every occurrence.
From: Doug Campbell
Files patched: gv.c
 It now complains about every occurrence.  (The bug resulted from an
 ill-conceived attempt to suppress a duplicate error message in a
 suboptimal fashion.)

NETaa13052: test for numeric sort sub return value fooled by taint magic.
From: Peter Jaspers-Fayer
Files patched: pp_ctl.c sv.h
 The test to see if the sort sub return value was numeric looked at the
 public flags rather than the private flags of the SV, so taint magic
 hid that info from the sort.

NETaa13053: forced a2p to use byacc
From: Andy Dougherty
Files patched: MANIFEST x2p/Makefile.SH x2p/a2p.c
 a2p.c is now pre-byacced and shipped with the kit.

NETaa13055: misnamed constant in previous patch.
From: Conrad Augustin
Files patched: op.c op.h toke.c
 The tokener translates $[ to a constant, but with a special marking in case
 the constant gets assigned to or localized.  Unfortunately, the marking
 was done with a combination of OPf_SPECIAL and OPf_MOD that was easily
 spoofed.  There is now a private OPpCONST_ARYLEN flag for this purpose.

NETaa13055: use of OPf_SPECIAL for $[ lvaluehood was too fragile.
Files patched: op.c op.h toke.c
 (same)

NETaa13056: convert needs to throw away any number info on its list.
From: Jack Shirazi - BIU
Files patched: op.c
 The listiness of the argument list leaked out to the subroutine call because
 of how prepend_elem and append_elem reuse an existing list.  The convert()
 routine just needs to discard any listiness it finds on its argument.

NETaa13058: AUTOLOAD shouldn't assume size of @_ is meaningful.
From: Florent Guillaume
Files patched: ext/DB_File/DB_File.pm ext/Fcntl/Fcntl.pm ext/GDBM_File/GDBM_File.pm ext/Socket/Socket.pm h2xs.SH
 I just deleted the optimization, which is silly anyway since the eventual
 subroutine definition is cached.

NETaa13059: now warns of use of \1 where $1 is necessary.
From: Gustaf Neumann
Files patched: toke.c
 Now says
 
     Can't use \1 to mean $1 in expression at foo line 2
 
 along with an explanation in perldiag.

NETaa13060: no longer warns on attempt to read <> operator's transition state.
From: Chaim Frenkel
Files patched: pp_hot.c
 No longer warns on <> operator's transitional state.

NETaa13140: warning said $ when @ would be more appropriate.
From: David J. MacKenzie
Files patched: op.c pod/perldiag.pod
 Now says
 
     (Did you mean $ or @ instead of %?)
 
 and added more explanation to perldiag.

NETaa13149: was reading freed memory to make incorrect error message.
Files patched: pp_ctl.c
 It was reading freed memory to make an error message that would be
 incorrect in any event because it had the inner filename rather than
 the outer.

NETaa13149: confess was sometimes less informative than croak
From: Jack Shirazi
Files patched: lib/Carp.pm
 (same)

NETaa13150: stderr needs to be STDERR in package
From: Jack Shirazi
Files patched: lib/File/CheckTree.pm
 Also fixed pl2pm to translate the filehandles to uppercase.

NETaa13150: uppercases stdin, stdout and stderr
Files patched: pl2pm
 (same)

NETaa13154: array assignment didn't notice package magic.
From: Brian Reichert
Files patched: pp_hot.c
 The list assignment operator looked for only set magic, but set magic is
 only on the elements of a magical hash, not on the hash as a whole.  I made
 the operator look for any magic at all on the target array or hash.

NETaa13155: &DB::DB left trash on the stack.
From: Thomas Koenig
Files patched: lib/perl5db.pl pp_ctl.c
 The call by pp_dbstate() to &DB::DB left trash on the stack.  It now
 calls DB in list context, and DB returns ().

NETaa13156: lexical variables didn't show up in debugger evals.
From: Joergen Haegg
Files patched: op.c
 The code that searched back up the context stack for the lexical scope
 outside the eval only partially took into consideration that there
 might be extra debugger subroutine frames that shouldn't be used, and
 ended up comparing the wrong statement sequence number to the range of
 valid sequence numbers for the scope of the lexical variable.  (There
 was also a bug fixed in passing that caused the scope of lexical to go
 clear to the end of the subroutine even if it was within an inner block.)

NETaa13157: any request for autoloaded DESTROY should create a null one.
From: Tom Christiansen
Files patched: lib/AutoLoader.pm
 If DESTROY.al is not located, it now creates sub DESTROY {} automatically.

NETaa13158: now preserves $@ around destructors while leaving eval.
From: Tim Bunce
Files patched: pp_ctl.c
 Applied supplied patch, except the whole second hunk can be replaced with
 
     sv_insert(errsv, 0, 0, message, strlen(message));

NETaa13160: clarified behavior of split without arguments
From: Harry Edmon
Files patched: pod/perlfunc.pod
 Clarified the behavior of split without arguments.

NETaa13162: eval {} lost list/scalar context
From: Dov Grobgeld
Files patched: op.c
 LEAVETRY didn't propagate number to ENTERTRY.

NETaa13163: clarified documentation of foreach using my variable
From: Tom Christiansen
Files patched: pod/perlsyn.pod
 Explained that foreach using a lexical is still localized.

NETaa13164: the dot detector for the end of formats was over-rambunctious.
From: John Stoffel
Files patched: toke.c
 The dot detector for the end of formats was over-rambunctious.  It would
 pick up any dot that didn't have a space in front of it.

NETaa13165: do {} while 1 never linked outer block into next chain.
From: Gisle Aas
Files patched: op.c
 When the conditional of do {} while 1; was optimized away, it confused the
 postfix order construction so that the block that ordinarily sits around the
 whole loop was never executed.  So when the loop tried to unstack between
 iterations, it got the wrong context, and blew away the lexical variables
 of the outer scope.  Fixed it by introducing a NULL opcode that will be
 optimized away later.

NETaa13167: coercion was looking at public bits rather than private bits.
From: Randal L. Schwartz
Also: Thomas Riechmann
Also: Shane Castle
Files patched: sv.c
 There were some bad ifdefs around the various varieties of set*id().  In
 addition, tainting was interacting badly with assignment to $> because
 sv_2iv() was examining SvPOK rather than SvPOKp, and so couldn't coerce
 a string uid to an integer one.

NETaa13167: had some ifdefs wrong on set*id.
Files patched: mg.c pp_hot.c
 (same)

NETaa13168: relaxed test for comparison of new and old fds
From: Casper H.S. Dik
Files patched: t/lib/posix.t
 I relaxed the comparison to just check that the new fd is greater.

NETaa13169: autoincrement can corrupt scalar value state.
From: Gisle Aas
Also: Tom Christiansen
Files patched: sv.c
 It assumed a PV didn't need to be upgraded to become an NV.

NETaa13169: previous patch could leak a string pointer.
Files patched: sv.c
 (same)

NETaa13170: symbols missing from global.sym
From: Tim Bunce
Files patched: global.sym
 Applied suggested patch.

NETaa13171: \\ in <<'END' shouldn't reduce to \.
From: Randal L. Schwartz
Files patched: toke.c
 <<'END' needed to bypass ordinary single-quote processing.

NETaa13172: 'use integer' turned off magical autoincrement.
From: Erich Rickheit KSC
Files patched: pp.c pp_hot.c
 The integer versions of the increment and decrement operators were trying too
 hard to be efficient.

NETaa13172: deleted duplicate increment and decrement code
Files patched: opcode.h opcode.pl pp.c
 (same)

NETaa13173: install should make shared libraries executable.
From: Brian Grossman
Also: Dave Nadler
Also: Eero Pajarre
Files patched: installperl
 Now gives permission 555 to any file ending with extension specified by $dlext.

NETaa13176: ck_rvconst didn't free the const it used up.
From: Nick Duffek
Files patched: op.c
 I checked in many random memory leaks under this bug number, since it
 was an eval that brought many of them out.

NETaa13176: didn't delete XRV for temp ref of destructor.
Files patched: sv.c
 (same)

NETaa13176: didn't delete op_pmshort in matching operators.
Files patched: op.c
 (same)

NETaa13176: eval leaked the name of the eval.
Files patched: scope.c
 (same)

NETaa13176: gp_free didn't free the format.
Files patched: gv.c
 (same)

NETaa13176: minor leaks in loop exits and constant subscript optimization.
Files patched: op.c
 (same)

NETaa13176: plugged some duplicate struct allocation memory leaks.
Files patched: perl.c
 (same)

NETaa13176: sv_clear of an FM didn't clear anything.
Files patched: sv.c
 (same)

NETaa13176: tr/// didn't mortalize its return value.
Files patched: pp.c
 (same)

NETaa13177: SCOPE optimization hid line number info
From: David J. MacKenzie
Also: Hallvard B Furuseth
Files patched: op.c
 Every pass on the syntax tree has to keep track of the current statement.
 Unfortunately, the single-statement block was optimized into a single
 statement between the time the variable was parsed and the time the
 void code scan was done, so that pass didn't see the OP_NEXTSTATE
 operator, because it has been optimized to an OP_NULL.
 
 Fortunately, null operands remember what they were, so it was pretty easy
 to make it set the correct line number anyway.

NETaa13178: some linux doesn't handle nm well
From: Alan Modra
Files patched: hints/linux.sh
 Applied supplied patch.

NETaa13180: localized slice now pre-extends array
From: Larry Schuler
Files patched: pp.c
 A localized slice now pre-extends its array to avoid reallocation during
 the scope of the local.

NETaa13181: m//g didn't keep track of whether previous match matched null.
From: "philippe.verdret"
Files patched: mg.h pp_hot.c
 A pattern isn't allowed to match a null string in the same place twice in
 a row.  m//g wasn't keeping track of whether the previous match matched
 the null string.

NETaa13182: now includes whitespace as a regexp metacharacter.
From: Larry Wall
Files patched: toke.c
 scan_const() now counts " \t\n\r\f\v" as metacharacters when scanning a pattern.

NETaa13183: sv_setsv shouldn't try to clone an object.
From: Peter Gordon
Files patched: sv.c
 The sv_mortalcopy() done by the return in STORE called sv_setsv(),
 which cloned the object.  sv_setsv() shouldn't be in the business of
 cloning objects.

NETaa13184: bogus warning on quoted signal handler name removed.
From: Dan Carson
Files patched: toke.c
 Now doesn't complain unless the first non-whitespace character after the =
 is an alphabetic character.

NETaa13186: now croaks on chop($')
From: Casper H.S. Dik
Files patched: doop.c
 Now croaks on chop($') and such.

NETaa13187: "${foo::bar}" now counts as mere delimitation, not as a bareword.
From: Jay Rogers
Files patched: toke.c
 "${foo::bar}" now counts as mere delimitation, not as a bareword inside a
 reference block.

NETaa13188: for backward compatibility, looks for "perl -" before "perl".
From: Russell Mosemann
Files patched: toke.c
 Now allows non-whitespace characters on the #! line between the "perl"
 and the "-".

NETaa13188: now allows non-whitespace after #!...perl before switches.
Files patched: toke.c
 (same)

NETaa13189: derivative files need to be removed before recreation
From: Simon Leinen
Also: Dick Middleton
Also: David J. MacKenzie
Files patched: embed_h.sh x2p/Makefile.SH
 Fixed various little nits as suggested in several messages.

NETaa13190: certain assignments can spoof pod directive recognizer
From: Ilya Zakharevich
Files patched: toke.c
 The lexer now only recognizes pod directives where a statement is expected.

NETaa13194: now returns undef when there is no curpm.
From: [email protected]
Files patched: mg.c
 Since there was no regexp prior to the "use", it was returning whatever the
 last successful match was within the "use", because there was no current
 regexp, so it treated it as a normal variable.  It now returns undef.

NETaa13195: semop had one S too many.
From: Joachim Huober
Files patched: opcode.pl
 The entry in opcode.pl had one too many S's.

NETaa13196: always assumes it's a Perl script if -c is used.
From: Dan Carson
Files patched: toke.c
 It now will assume it's a Perl script if the -c switch is used.

NETaa13197: changed implicit -> message to be more understandable.
From: Bruce Barnett
Files patched: op.c pod/perldiag.pod
 I changed the error message to be more understandable.  It now says
 
     Can't use subscript on sort...
 

NETaa13201: added OPpCONST_ENTERED flag to properly enter filehandle symbols.
From: E. Jay Berkenbilt
Also: Tom Christiansen
Files patched: op.c op.h toke.c
 The grammatical reduction of a print statement didn't properly count
 the filehandle as a symbol reference because it couldn't distinguish
 between a symbol entered earlier in the program and a symbol entered
 for the first time down in the lexer.

NETaa13203: README shouldn't mention uperl.o any more.
From: Anno Siegel
Files patched: README

NETaa13204: .= shouldn't warn on uninitialized target.
From: Pete Peterson
Files patched: pp_hot.c
 No longer warns on uninitialized target of .= operator.

NETaa13206: handy macros in XSUB.h
From: Tim Bunce
Files patched: XSUB.h
 Added suggested macros.

NETaa13228: commonality checker didn't treat lexicals as variables.
From: [email protected]
Files patched: op.c opcode.pl
 The list assignment operator tries to avoid unnecessary copies by doing the
 assignment directly if there are no common variables on either side of the
 equals.  Unfortunately, the code that decided that only recognized references
 to dynamic variables, not lexical variables.

NETaa13229: fixed sign stuff for complement, integer coercion.
From: Larry Wall
Files patched: perl.h pp.c sv.c
 Fixed ~0 and integer coercions.

NETaa13230: no longer tries to reuse scratchpad temps if tainting in effect.
From: Luca Fini
Files patched: op.c
 I haven't reproduced it, but I believe the problem is the reuse of scratchpad
 temporaries between statements.  I've made it not try to reuse them if
 tainting is in effect.

NETaa13231: *foo = *bar now prevents typo warnings on "foo"
From: Robin Barker
Files patched: sv.c
 Aliasing of the form *foo = *bar is now protected from the typo warnings.
 Previously only the *foo = \$bar form was.

NETaa13235: require BAREWORD now introduces package name immediately.
From: Larry Wall
Files patched: toke.c
 require BAREWORD now introduces package name immediately.  This lets the
 method intuit code work right even though the require hasn't actually run
 yet.

NETaa13289: didn't calculate correctly using arybase.
From: Jared Rhine
Files patched: pp.c pp_hot.c
 The runtime code didn't use curcop->cop_arybase correctly.

NETaa13301: store now throws exception on error
From: Barry Friedman
Files patched: ext/GDBM_File/GDBM_File.xs ext/NDBM_File/NDBM_File.xs ext/ODBM_File/ODBM_File.xs ext/SDBM_File/SDBM_File.xs
 Changed warn to croak in ext/*DBM_File/*.xs.

NETaa13302: ctime now takes Time_t rather than Time_t*.
From: Rodger Anderson
Files patched: ext/POSIX/POSIX.xs
 Now declares a Time_t and takes the address of that in CODE.

NETaa13302: shorter way to do this patch
Files patched: ext/POSIX/POSIX.xs
 (same)

NETaa13304: could feed too large $@ back into croak, whereupon it croaked.
From: Larry Wall
Files patched: perl.c
 callist() could feed $@ back into croak with more than a bare %s.  (croak()
 handles long strings with a bare %s okay.)

NETaa13305: compiler misoptimized RHS to outside of s/a/print/e
From: Brian S. Cashman <[email protected]>
Files patched: op.c
 The syntax tree was being misconstructed because the compiler felt that
 the RHS was invariant, so it did it outside the s///.

NETaa13314: assigning mortal to lexical leaks
From: Larry Wall
Files patched: sv.c
 In stealing strings, sv_setsv was checking SvPOK to see if it should free
 the destination string.  It should have been checking SvPVX.

NETaa13316: wait4pid now recalled when errno == EINTR
From: Robert J. Pankratz
Files patched: pp_sys.c util.c
 system() and the close() of a piped open now recall wait4pid if it returned
 prematurely with errno == EINTR.

NETaa13329: needed to localize taint magic
From: Brian Katzung
Files patched: sv.c doio.c mg.c pp_hot.c pp_sys.c scope.c taint.c
 Taint magic is now localized better, though I had to resort to a kludge
 to allow a value to be both tainted and untainted simultaneously during
 the assignment of
 
     local $foo = $_[0];
 
 when $_[0] is a reference to the variable $foo already.

NETaa13341: clarified interaction of AnyDBM_File::ISA and "use"
From: Ian Phillipps
Files patched: pod/modpods/AnyDBMFile.pod
 The doc was misleading.

NETaa13342: grep and map with block would enter block but never leave it.
From: Ian Phillipps
Files patched: op.c
 The compiler use some sort-checking code to handle the arguments of
 grep and map.  Unfortunately, this wiped out the block exit opcode while
 leaving the block entry opcode.  This doesn't matter to sort, but did
 matter to grep and map.  It now leave the block entry intact.
 
 The reason it worked without the my is because the block entry and exit
 were optimized away to an OP_SCOPE, which it doesn't matter if it's there
 or not.

NETaa13343: goto needed to longjmp when in a signal handler.
From: Robert Partington
Files patched: pp_ctl.c
 goto needed to longjmp() when in a signal handler to get back into the
 right run() context.
 

NETaa13344: strict vars shouldn't apply to globs or filehandles.
From: Andrew Wilcox
Files patched: gv.c
 Filehandles and globs will be excepted from "strict vars", so that you can
 do the standard Perl 4 trick of
 
     use strict;
     sub foo {
         local(*IN);
         open(IN,"file");
     }
 

NETaa13345: assert.pl didn't use package DB
From: Hans Mulder
Files patched: lib/assert.pl
 Now it does.

NETaa13348: av_undef didn't free scalar representing $#foo.
From: David Filo
Files patched: av.c
 av_undef didn't free scalar representing $#foo.

NETaa13349: sort sub accumulated save stack entries
From: David Filo
Files patched: pp_ctl.c
 COMMON only gets set if assigning to @_, which is reasonable.  Most of the
 problem was a memory leak.

NETaa13351: didn't treat indirect filehandles as references.
From: Andy Dougherty
Files patched: op.c
 Now produces