Revision history for indirect 0.26 2011-10-23 14:25 UTC + Add : "no indirect 'global'" enables the pragma for the whole program, except for lexical scopes that "use indirect" explicitely. + Chg : Passing both the 'fatal' and 'hook' options to unimport() is now considered an error, and will result in an exception. unimport() used to consider only the first passed option of those two, and silently ignored the other. + Tst : Test failures of t/41-threads-teardown.t and t/50-external.t on Cygwin should have been addressed. + Tst : Threads tests will not fail anymore if resources constraints prevent the system from creating all the required threads. 0.25 2011-08-24 15:40 UTC + Fix : RT #69291 is now also fixed for perl 5.8. The pragma will no longer vivify the "indirect" entry in the hints hash %^H on perl 5.8. + Tst : Attempt to make t/50-external.t pass on Cygwin. 0.24 2011-07-17 23:15 UTC + Fix : [RT #64521] : "no indirect" leaking into eval. This is currently only fixed for perl 5.10 (perl 5.12 and higher were never affected). It was caused by a very stupid mistake of mine that was introduced in indirect version 0.23. Thanks Michael G Schwern for reporting. + Fix : [RT #69291] : indirect.pm breaks %^H. This was caused by the same mistake as for the previous bug, and as such it is also only fixed for perl 5.10 (and never affected perl 5.12). Thanks Andrew Main for reporting. + Doc : C++ compilers are officially NOT supported. 0.23 2010-10-03 00:15 UTC + Fix : Some indirect constructs could be incorrectly reported when several modules were used in the same scope. This caused t/30-scope.t to fail randomly. + Tst : Threads tests are now only run on perl 5.13.4 and higher. They could segfault randomly because of what seems to be an internal bug of Perl, which has been addressed in 5.13.4. There is also an environment variable that allows you to forcefully run those tests, but it should be set only for author testing and not for end users. 0.22 2010-08-16 16:00 UTC + Add : Indirect constructs are now reported for code interpolated in quote-like environments, like "${\( ... )}", "@{[ ... ]}", s/pattern/ ... /e, qr/(?{ ... })/ or qr/(??{ ... })/. + Add : You can now make the pragma lethal by passing anything matching /^:?fatal$/i to import(), including "FATAL" and ":Fatal". + Fix : [RT #60378] : segmentation fault on indirect_ck_method. This caused constructs like "@{[ $obj->$meth ]}" to segfault when $meth was a lexical. Thanks Tokuhiro Matsuno for reporting. 0.21 2010-05-31 23:10 UTC + Chg : perl 5.8.1 is now required (instead of 5.8.0). + Fix : [RT #57699] : indirect fail with 64-bit int on 5.13.1. It was actually a problem with thread destructors segfaulting because they weren't called at the right time anymore. Thanks Andrew Main for reporting. + Tst : A few more regression tests about the scope leak bug. 0.20 2010-04-18 21:25 UTC + Fix : [RT #50570] : "indirect" leaking into LWP. Thanks Andrew Main for reporting. More generally, the require propagation workaround on 5.8-5.10.0 has been overhauled, and other scope leaks should be fixed. + Fix : Test failures with 5.12 on Windows where Strawberry Perl crashes because the SystemRoot environment variable is missing. + Fix : Work around Kwalitee test misfailures. 0.19 2009-08-28 18:40 UTC + Add : The new constant I_FORKSAFE can be tested to know whether the module will behave nicely when fork()ing. It's currently always true except on Windows where you need perl 5.10.1 for it to be true. + Fix : I_THREADSAFE and I_FORKSAFE ought to be true when PERL_INDIRECT_PM_DISABLE is set. + Fix : The pragma could still leak if you passed to the "hook" option a reference to the same (named) subroutine from two different require scopes. The fix also provides a better solution for RT #47902. + Fix : Segfaults when indirect is loaded for the first time from inside a thread. + Fix : Leaks of memory associated with the root interpreter. + Opt : Less memory will be used for non-threaded perls version 5.10.0 and below, and for threaded perls from version 5.10.1. 0.18 2009-08-23 16:15 UTC + Add : When set, the PERL_INDIRECT_PM_DISABLE environment variable disables the pragma globally. 0.17 2009-07-16 12:10 UTC + Fix : [RT #47902] : "no indirect" leaking again. This actually turned out to be a bug in how the hook coderefs were stored in the hints hash. Thanks Andrew Main for reporting once again. + Fix : t/80-regressions.t failing on Windows. + Tst : Yet more cleanups. 0.16 2009-07-14 16:50 UTC + Add : Indirect calls on blocks are now reported. For those, '{' is passed to the hook as the object description. + Add : The new indirect::msg() function publicizes the default warning/exception message. + Fix : [RT #47866] : Segfault with UTF-8 regexps. Thanks Andrew Main for reporting. + Tst : Cleanups. 0.15 2009-07-08 22:55 UTC + Fix : Invalid constructs with the same method and package name were not reported. + Fix : The error line number used to point to the end of the expression instead of its beginning. 0.14 2009-06-04 21:55 UTC + Fix : Prevent bogus invalid syntaxes caused by reallocated memory chunks. Thanks Andrew Main for reporting with a reproducible test case. 0.13 2009-05-24 18:50 UTC + Add : The hook now receives the file name and the line where the error happened in respectively $_[2] and $_[3]. + Fix : Pass mortalized copies of the method name and object to the hook. This seems to fix some rare crashes. + Fix : Work around a bug in perl 5.10.0 and lower. Thanks Andrew Main for teaching me about this issue. + Fix : Report the correct file in error messages (a regression from the previous version). 0.12 2009-05-03 14:30 UTC + Add : You can specify the action to execute for each indirect construct encountered with the new "hook => $coderef" unimport() option. + Chg : A ptable is now used internally for the op => position mapping. + Fix : The pragma is now built with thread safety features enabled whenever possible (a notable exception is perl 5.8.x on Win32, as something seems wrong with its context handling). The new indirect::I_THREADSAFE() constant reflects this. + Fix : A negation precedence nit in indirect_ck_entersub(). + Tst : "use/no indirect" while parsing an indirect construct. + Tst : Thread safety. 0.11 2009-02-08 18:35 UTC + Fix : Potential collisions by hashing pointers with a wrong format. + Upd : Resources in META.yml. 0.10 2009-01-17 12:40 UTC Re-release 0.09_01 as stable. 0.09_01 2008-12-08 17:55 UTC + Fix : Invalid memory read with "${\(new Foo)}" constructs. The corresponding test is turned back on. + Tst : Refinements in t/30-scope.t 0.09 2008-12-05 20:35 UTC + Add : Support for perl 5.8. + Tst : Skip a test in t/10-good.t that randomly segfaults for (I guess) systems stricter than linux in the way they manage their memory. 0.08 2008-10-22 14:45 UTC + Fix : A rare edge case for package whose names are prefix of 'main'. + Tst : Test $$ as variable and state variables. 0.07_03 2008-10-17 20:10 UTC + Add : Support and tests for variables with spaces after the sigil. + Upd : META.yml spec updated to 1.4. 0.07_02 2008-10-15 21:10 UTC + Add : Support and tests for package variables. + Tst : Coverage improved by removing dead code. 0.07_01 2008-10-15 16:00 UTC + Fix : [RT #40055] : Not handling RV2SV => GV(SV) correctly, which could cause 'no indirect; print' segfaults. Thanks Goro Fuji for reporting. 0.06 2008-10-11 16:45 UTC + Doc : Nits. + Tst : Test "no indirect 'anything'", "foo Class->bar", and indirect uses of exec() and system(). 0.05 2008-10-02 14:40 UTC + Chg : Now the old check function is always called before storing an op into the map. + Fix : Misc code and docs refinements. 0.04 2008-08-30 19:00 UTC + Fix : Clean up the op->src hash when we're done with an entersub. + Tst : No longer fork for testing. IPC::Cmd isn't required anymore. 0.03 2008-08-12 15:25 UTC This release is kindly supported by Copenhagen Hotel Centrum WiFi. + Fix : Tests used not to pass PERL5OPTS to their kids. This lead to failures under CPAN. I think. + Tst : Refinements. 0.02 2008-08-11 15:55 UTC + Fix : Some building failures with old gcc versions that didn't seem to like the ((hint == 2) ? croak : warn)(msg) construct. I think. + Rem : Unused cruft from a previous implementation. + Tst : Fail more gracefully when we can't capture buffers or when the child returned an error. 0.01 2008-08-10 20:40 UTC First version, released on an unsuspecting world.