miniperl PP glob/Perl_start_glob() wipes all of %ENV and perlglob.exe proc can't start on VC 2005/WinXP

[p5pRT]

Migrated from rt.perl.org#126041 (status was 'resolved')

Searchable as RT126041$

[p5pRT]

From @bulk88

Created by @bulk88

Pics attached.

With WinXP SP3 and VC 2005\, miniperl's PP glob() can't launch the perlglob.exe process. The process can't be started due to error code 0xc0150004 which is STATUS_SXS_ASSEMBLY_NOT_FOUND\, MS's message for the error is "The referenced assembly is not installed on the system.". Using procmon I see that the OS code is loading a very wrong path\, "C​:\perl521\src\cpan\CPAN\%SystemRoot%\WinSxS\x86_Microsoft.VC80.CRT_1fc8b3b9a1e18e3b_8.0.50727.6195_x-ww_44262b86" instead of "C​:\WINDOWS\WinSxS\x86_Microsoft.VC80.CRT_1fc8b3b9a1e18e3b_8.0.50727.6195_x-ww_44262b86". In VC 2005 and 2008 (but not any older or new VC version)\, the MS CRT (libc) DLL is SXS tracked. SXS is MS's version of DLL versioning. A central OS DB of known DLL versions and their checksums/crypto security which has grandiose visions of being a package manager\, with centralized forwarding (upgrading) in any graph layout a micromanager loves. It changed the term "DLL hell" to "Manifest hell". A manifest is an plain text XML file inside the DLL/EXE that specifies which SXS version/checksums of a DLL's filename a DLL wants to be bound against.

The reason that filepath is messed up is Perl_start_glob is almost wiping all of %ENV which was done in commit http​://perl5.git.perl.org/perl.git/commitdiff/a3342be368647e67883bc6ba3bea07bb06880b80 as a fix for https://rt-archive.perl.org/perl5/Ticket/Display.html?id=93638 . Father C 1 month later caught a bug https://rt-archive.perl.org/perl5/Ticket/Display.html?id=93638#txn-995018 in his own code and fixed it in http​://perl5.git.perl.org/perl.git/commitdiff/93b2dae14541cd2bc39e2d4b1bb5f0cc2c0ef0c7 . 8 months later ticket https://rt-archive.perl.org/perl5/Ticket/Display.html?id=113798 was filed which said a Mingw GCC built perlglob.exe can't be executed since it relies on a GCC DLL that is in PATH but not in C​:/Windows (the Windows system DLLs are special cased ("KnownDLLs" key in registry)). That 2nd ticket got a fix in http​://perl5.git.perl.org/perl.git/commitdiff/25018358f5fe1e7a2595f788f4ad0296fece4086 by FC.

This code that wipes %ENV has been subject to being revised twice in history for causing breakage. This bug ticket is now going to be the THIRD revision of this code. Perhaps the original fix\, instead of localizing and wiping all of %ENV\, should have just localized the "LS_COLORS" env var alone in the %ENV hash. If you dont know why you are deleting $ENV{"ONLY_A_MAGIC_8_BALL_KNOWS_THIS_ENV_VARS_NAME"} maybe you shouldn't be deleting the env var. I can think of a scenario where some C instrumented profiler requires a certain ENV var for the instrumented binary to be loadable into a process. If perlglob.exe was instrumented\, it wouldn't start due to the missing env var since all of them unconditionally were wiped.

There are 2.25 ways to fix this\, add SYSTEMROOT to the list of PATH and HOME being passed through\, or stop localizing and wiping all of %ENV and just localize LS_COLORS\, or\, I dont like this fix "#ifndef WIN32\n/*localize/wipe %ENV */\n#endif" since that would fix it for me\, but doesn't fix the greater design problem of deleting env vars that you dont know why you are deleting them. So which way will this be fixed?

Greping the code shows some other possibly questionable ENV var manipulations which can be causing similar problems to this where it wipes critical env vars or all of %ENV.

http​://perl5.git.perl.org/perl.git/blob/2efb8b4b644d5f3c28974a8f577081b4142decd2​:/t/loc_tools.pl#l222 http​://perl5.git.perl.org/perl.git/blob/2efb8b4b644d5f3c28974a8f577081b4142decd2​:/t/test.pl#l765 http​://perl5.git.perl.org/perl.git/blob/2efb8b4b644d5f3c28974a8f577081b4142decd2​:/taint.c#l83 http​://perl5.git.perl.org/perl.git/blob/2efb8b4b644d5f3c28974a8f577081b4142decd2​:/cpan/CPAN/lib/CPAN/Distribution.pm#l2165 http​://perl5.git.perl.org/perl.git/blob/2efb8b4b644d5f3c28974a8f577081b4142decd2​:/cpan/ExtUtils-MakeMaker/t/lib/MakeMaker/Test/Utils.pm#l22 http​://perl5.git.perl.org/perl.git/blob/2efb8b4b644d5f3c28974a8f577081b4142decd2​:/dist/PathTools/Cwd.pm#l220 http​://perl5.git.perl.org/perl.git/blob/2efb8b4b644d5f3c28974a8f577081b4142decd2​:/dist/PathTools/t/tmpdir.t#l25 http​://perl5.git.perl.org/perl.git/blob/2efb8b4b644d5f3c28974a8f577081b4142decd2​:/ext/File-Glob/t/basic.t#l40

I never encountered this bug before since I almost never use VC 2005 on WinXP to build perl. But I often use VC 2008 on Server 2003 R2. Both VC 2005 and VC 2008 uses SXS CRTs\, but I think MS revised the SXS (part of DLL) loader so env var SYSTEMROOT is not used on Server 2003 and up (maybe its 2003 R2 and up since I have R2)\, since that would evade the "central management" and "security" principles of SXS. Server 2003 is not the Server version of Windows XP\, development wise. Windows XP is 5.1. Server 2003 and "Windows XP for 64 bits" are Windows 5.2. 5.2 is a later development branch of Windows than 5.1. For example\, I know in Windows XP\, CreateProcess checks the App Compat DB by memory mapping C​:\WINDOWS\AppPatch\sysmain.sdb into the perl process. On Server 2003 CreateProcess checks the App Compat DB by doing an inter-process RPC call to a daemon. Hence Server 2003 is evolutionary ahead of Windows XP.

Not very interesting callstack of broken perlglob.exe process which has halted to show the fatal error popup\, C main() was never reached before the process fatally errored and halted. The fatal error popup box is not being drawn by perlglob.exe but by csrss.exe\, the "Win32 [emulation] layer daemon".

  ntdll.dll!_KiFastSystemCallRet@​0()
  ntdll.dll!_ZwRaiseHardError@​24() + 0xc
  ntdll.dll!_LdrpInitializationFailure@​4() + 0x2d
  ntdll.dll!__LdrpInitialize@​12() + 0x25fe0
  ntdll.dll!_KiUserApcDispatcher@​20() + 0x7

Perl Info

``` Flags: category=core severity=high Site configuration information for perl 5.23.2: Configured by Owner at Wed Jul 22 12:17:22 2015. Summary of my perl5 (revision 5 version 23 subversion 2) configuration: Derived from: f8ccc6c719cd8ef25c59a4a21fe1bab9f6a38c5b Ancestor: 30dc90f1923c08aae65b60f713ac12490a716e10 Platform: osname=MSWin32, osvers=5.1, archname=MSWin32-x86-multi-thread uname='' config_args='undef' hint=recommended, useposix=true, d_sigaction=undef useithreads=define, usemultiplicity=define use64bitint=undef, use64bitall=undef, uselongdouble=undef usemymalloc=n, bincompat5005=undef Compiler: cc='cl', ccflags ='-nologo -GF -W3 -O1 -MD -Zi -DNDEBUG -GL -DWIN32 -D_CONSOLE -DNO_STRICT -DPERL_TEXTMODE_SCRIPTS -DPERL_IMPLICIT_CONTEXT -DPERL_IMPLICIT_SYS -D_USE_32BIT_TIME_T', optimize='-O1 -MD -Zi -DNDEBUG -GL', cppflags='-DWIN32' ccversion='13.10.6030', gccversion='', gccosandvers='' intsize=4, longsize=4, ptrsize=4, doublesize=8, byteorder=1234, doublekind=3 d_longlong=undef, longlongsize=8, d_longdbl=define, longdblsize=8, longdblkind=0 ivtype='long', ivsize=4, nvtype='double', nvsize=8, Off_t='__int64', lseeksize=8 alignbytes=8, prototype=define Linker and Libraries: ld='link', ldflags ='-nologo -nodefaultlib -debug -opt:ref,icf -ltcg -libpath:"c:\perl521\lib\CORE" -machine:x86' libpth=\lib libs=oldnames.lib kernel32.lib user32.lib gdi32.lib winspool.lib comdlg32.lib advapi32.lib shell32.lib ole32.lib oleaut32.lib netapi32.lib uuid.lib ws2_32.lib mpr.lib winmm.lib version.lib odbc32.lib odbccp32.lib comctl32.lib msvcrt.lib perllibs=oldnames.lib kernel32.lib user32.lib gdi32.lib winspool.lib comdlg32.lib advapi32.lib shell32.lib ole32.lib oleaut32.lib netapi32.lib uuid.lib ws2_32.lib mpr.lib winmm.lib version.lib odbc32.lib odbccp32.lib comctl32.lib msvcrt.lib libc=msvcrt.lib, so=dll, useshrplib=true, libperl=perl523.lib gnulibc_version='' Dynamic Linking: dlsrc=dl_win32.xs, dlext=dll, d_dlsymun=undef, ccdlflags=' ' cccdlflags=' ', lddlflags='-dll -nologo -nodefaultlib -debug -opt:ref,icf -ltcg -libpath:"c:\perl521\lib\CORE" -machine:x86' Locally applied patches: uncommitted-changes a3b5f9fb00dafcb11d1766baaec76abc90f9fd06 f8ccc6c719cd8ef25c59a4a21fe1bab9f6a38c5b @INC for perl 5.23.2: C:/perl521/site/lib C:/perl521/lib . Environment for perl 5.23.2: HOME (unset) LANG (unset) LANGUAGE (unset) LD_LIBRARY_PATH (unset) LOGDIR (unset) PATH=C:\perl521\bin;C:\Program Files\Microsoft Visual Studio .NET 2003\Common7\IDE;C:\Program Files\Microsoft Visual Studio .NET 2003\VC7\BIN;C:\Program Files\Microsoft Visual Studio .NET 2003\Common7\Tools;C:\Program Files\Microsoft Visual Studio .NET 2003\Common7\Tools\bin\prerelease;C:\WINDOWS\system32;C:\WINDOWS;C:\WINDOWS\system32\wbem; PERL_BADLANG (unset) SHELL (unset) ```

[p5pRT]

From @bulk88

syscall_log.PNG

[p5pRT]

From @bulk88

env_vars_of_paused_mini_proc-normal.PNG

[p5pRT]

From @bulk88

env_vars_of_paused_perlglob.exe-proc.PNG

[p5pRT]

From @bulk88

fatal_error_from_perlglob.exe.PNG

[p5pRT]

From @bulk88

On Fri Sep 11 01​:13​:01 2015\, bulk88 wrote​:

This is a bug report for perl from bulk88@​hotmail.com\, generated with the help of perlbug 1.40 running under perl 5.23.2.

----------------------------------------------------------------- [Please describe your issue here]

Pics attached.

With WinXP SP3 and VC 2005\, miniperl's PP glob() can't launch the perlglob.exe process. The process can't be started due to error code 0xc0150004 which is STATUS_SXS_ASSEMBLY_NOT_FOUND.

When I comment out the %ENV localize code in doio.c\, and smoke that perl build\, I get a test failure because it tried to open a path "C​:\perl521\src\t\%SystemRoot%\WinSxS\x86_Microsoft.VC80.CRT_1fc8b3b9a1e18e3b_8.0.50727.6195_x-ww_44262b86" when loading Win32​:: and its XS DLL.

---

C​:\perl521\src\t>perl harness -v op/magic.t op/magic.t .. 1..190 ok 1 - *SIG appears to be defined at the outset ok 2 - *^OPEN appears to be defined at the outset ok 3 - *^TAINT appears to be defined at the outset ok 4 - *^UNICODE appears to be defined at the outset ok 5 - *^UTF8LOCALE appears to be defined at the outset ok 6 - *^WARNING_BITS appears to be defined at the outset ok 7 - *1 appears to be defined at the outset ok 8 - *2 appears to be defined at the outset ok 9 - *3 appears to be defined at the outset ok 10 - *4 appears to be defined at the outset ok 11 - *5 appears to be defined at the outset ok 12 - *6 appears to be defined at the outset ok 13 - *7 appears to be defined at the outset ok 14 - *8 appears to be defined at the outset ok 15 - *9 appears to be defined at the outset ok 16 - *42 appears to be defined at the outset ok 17 - *& appears to be defined at the outset ok 18 - *` appears to be defined at the outset ok 19 - *' appears to be defined at the outset ok 20 - *​: appears to be defined at the outset ok 21 - *? appears to be defined at the outset ok 22 - *! appears to be defined at the outset ok 23 - *_ appears to be defined at the outset ok 24 - *- appears to be defined at the outset ok 25 - *[ appears to be defined at the outset ok 26 - *^ appears to be defined at the outset ok 27 - *~ appears to be defined at the outset ok 28 - *= appears to be defined at the outset ok 29 - *% appears to be defined at the outset ok 30 - *. appears to be defined at the outset ok 31 - *( appears to be defined at the outset ok 32 - *) appears to be defined at the outset ok 33 - *\< appears to be defined at the outset ok 34 - *> appears to be defined at the outset ok 35 - *\ appears to be defined at the outset ok 36 - */ appears to be defined at the outset ok 37 - *$ appears to be defined at the outset ok 38 - *| appears to be defined at the outset ok 39 - *+ appears to be defined at the outset ok 40 - *; appears to be defined at the outset ok 41 - *] appears to be defined at the outset ok 42 - *^A appears to be defined at the outset ok 43 - *^C appears to be defined at the outset ok 44 - *^D appears to be defined at the outset ok 45 - *^E appears to be defined at the outset ok 46 - *^F appears to be defined at the outset ok 47 - *^H appears to be defined at the outset ok 48 - *^I appears to be defined at the outset ok 49 - *^L appears to be defined at the outset ok 50 - *^N appears to be defined at the outset ok 51 - *^O appears to be defined at the outset ok 52 - *^P appears to be defined at the outset ok 53 - *^S appears to be defined at the outset ok 54 - *^T appears to be defined at the outset ok 55 - *^V appears to be defined at the outset ok 56 - *^W appears to be defined at the outset ok 57 - *^UTF8CACHE appears to be defined at the outset ok 58 - *​::12345 appears to be defined at the outset ok 59 - *main​::98732 appears to be defined at the outset ok 60 - *^LAST_FH appears to be defined at the outset ok 61 ok 62 - Unlinked file can't be opened ok 63 # skip SIGINT not safe on this platform ok 64 # skip SIGINT not safe on this platform ok 65 # skip SIGINT not safe on this platform ok 66 # skip SIGINT not safe on this platform ok 67 # skip SIGINT not safe on this platform ok 68 ok 69 ok 70 - deleting $​::{ENV} ok 71 ok 72 ok 73 ok 74 ok 75 - referencing @​` before $` etc. still saws off ampersands ok 76 - referencing @​& before $& etc. still saws off ampersands ok 77 - referencing @​' before $' etc. still saws off ampersands ok 78 ok 79 ok 80 ok 81 ok 82 ok 83 ok 84 ok 85 - no %@​ ok 86 ok 87 - $$ can be modified ok 88 # skip no fork ok 89 ok 90 ok 91 ok 92 ok 93 ok 94 ok 95 ok 96 ok 97 ok 98 ok 99 - Assigning $^I does not clobber $^O ok 100 - test $^P = 0 ok 101 - test $^P = 1 ok 102 # skip We don't have prctl() here\, or we're on Android ok 103 # skip We don't have prctl() here\, or we're on Android ok 104 ok 105 ok 106 ok 107 ok 108 - defined *{"!"} does not stop %! from working ok 109 - Referencing %{"powie​::!"} doesn't load Errno ok 110 - Referencing %{"powie​::!"} after mentioning ${"powie​::!"} doesn't load E rrno ok 111 - Referencing %{"powie​::+"} doesn't load Tie​::Hash​::NamedCapture ok 112 - Referencing %{"powie​::+"} after mentioning ${"powie​::+"} doesn't load T ie​::Hash​::NamedCapture ok 113 ok 114 ok 115 ok 116 ok 117 ok 118 ok 119 ok 120 ok 121 ok 122 ok 123 - Push a constant on a magic array ok 124 - Assign a constant to a magic hash ok 125 - Assign a shared key to a magic hash ok 126 - Can assign to __WARN__ ok 127 - Can delete from __WARN__ ok 128 - __WARN__ is now gone ok 129 - __WARN__ remains gone ok 130 - Can assign to INT ok 131 - Can delete from INT ok 132 - INT is now gone ok 133 - INT remains gone ok 134 - Can assign to HUNGRY ok 135 - Can delete from HUNGRY ok 136 - HUNGRY is now gone ok 137 - HUNGRY remains gone ok 138 - __DIE__ is not present ok 139 - delete of __DIE__ returns undef ok 140 - _BOGUS_HOOK is not present ok 141 - delete of _BOGUS_HOOK returns undef ok 142 - KILL is not present ok 143 - delete of KILL returns undef ok 144 - THIRSTY is not present ok 145 - delete of THIRSTY returns undef ok 146 - [perl \#72850] Core dump in bleadperl from perl -e '$! = 9999; $a = $!; ' ok 147 - defined *{"+"} does not stop %+ from working ok 148 - defined *{"-"} does not stop %- from working ok 149 - %- still loads Tie​::Hash​::NamedCapture when stash and UNIVERSAL​::AUTOLO AD exist ok 150 - $[ still loads arybase when stash and UNIVERSAL​::AUTOLOAD exist ok 151 - %! still loads Errno when stash and UNIVERSAL​::AUTOLOAD exist ok 152 - ${^LAST_FH} after tell ok 153 - ${^LAST_FH} after another tell ok 154 - ${^LAST_FH} referencing lexical coercible glob ok 155 - ${^LAST_FH} is undef when PL_last_in_gv is NULL ok 156 - [perl \#4760] print $| = ~$| ok 157 - [perl \#115206] no crash when vivifying $| while *{+select}{IO} is unde f ok 158 - $^H modifies ${^OPEN} consistently ok 159 - deleting $​::{"\cH"} ok 160 - ${^MPE} starts undefined ok 161 - ${^MPE} can be incremented ok 162 - ${^MPEN} starts undefined ok 163 - ${^MPEN} can be incremented ok 164 - Setting ${^E_NCODING} croaks ok 165 - can read ${^E_NCODING} without blowing up ok 166 - ${^E_NCODING} is undef ok 167 Can't load '../lib/auto/Win32/Win32.dll' for module Win32​: load_file​:The referen ced assembly is not installed on your system at ../lib/DynaLoader.pm line 194.   at ../lib/DynaLoader.pm line 100.   DynaLoader​::confess("Can't load '../lib/auto/Win32/Win32.dll' for module Win32​: lo"...) called at ../lib/DynaLoader.pm line 194   DynaLoader​::bootstrap("Win32") called at ../lib/Win32.pm line 650   require Win32.pm called at op/magic.t line 72   main​::env_is("__NoNeSuCh"\, "foo"\, "setting \$0 does not break %ENV") cal led at op/magic.t line 711 Compilation failed in require at op/magic.t line 72. # Looks like you planned 190 tests but ran 167. Dubious\, test returned 1 (wstat 256\, 0x100) Failed 23/190 subtests   (less 8 skipped subtests​: 159 okay)

Test Summary Report

---

op/magic.t (Wstat​: 256 Tests​: 167 Failed​: 0)   Non-zero exit status​: 1   Parse errors​: Bad plan. You planned 190 tests but ran 167. Files=1\, Tests=167\, 2 wallclock secs ( 0.03 usr + 0.00 sys = 0.03 CPU) Result​: FAIL

---

-- bulk88 ~ bulk88 at hotmail.com

[p5pRT]

From @cpansprout

On Fri Sep 11 21​:34​:06 2015\, bulk88 wrote​:

When I comment out the %ENV localize code in doio.c\, and smoke that perl build\, I get a test failure because it tried to open a path "C​:\perl521\src\t\%SystemRoot%\WinSxS\x86_Microsoft.VC80.CRT_1fc8b3b9a1e18e3b_8.0.50727.6195_x- ww_44262b86" when loading Win32​:: and its XS DLL.

What accounts for that?

--

Father Chrysostomos

[p5pRT]

The RT System itself - Status changed from 'new' to 'open'

[p5pRT]

From @tonycoz

On Fri Sep 11 01​:13​:01 2015\, bulk88 wrote​:

There are 2.25 ways to fix this\, add SYSTEMROOT to the list of PATH and HOME being passed through\, or stop localizing and wiping all of %ENV and just localize LS_COLORS\, or\, I dont like this fix "#ifndef WIN32\n/*localize/wipe %ENV */\n#endif" since that would fix it for me\, but doesn't fix the greater design problem of deleting env vars that you dont know why you are deleting them. So which way will this be fixed?

Another option would be to bring the mechanics of csh_glob from File-Glob into core\, similar to what was suggested in

https://rt-archive.perl.org/perl5/Ticket/Display.html?id=93638#txn-951162

To me that seems the safest mechanism\, it would eliminate the need for shelling out at all.

Greping the code shows some other possibly questionable ENV var manipulations which can be causing similar problems to this where it wipes critical env vars or all of %ENV.

http​://perl5.git.perl.org/perl.git/blob/2efb8b4b644d5f3c28974a8f577081b4142decd2​:/t/loc_tools.pl#l222

This seems safe enough to me. Except for PATH\, only "questionable" variables that change the behaviour of the shell are removed\, PATH isn't an issue since it seems like no executable is run on Win32.

http​://perl5.git.perl.org/perl.git/blob/2efb8b4b644d5f3c28974a8f577081b4142decd2​:/t/test.pl#l765

Again\, only the questionable names are removed\, and PATH is modified.

http​://perl5.git.perl.org/perl.git/blob/2efb8b4b644d5f3c28974a8f577081b4142decd2​:/taint.c#l83

The same questionable names\, and PERL5SHELL\, which is sensitive under taint.

http​://perl5.git.perl.org/perl.git/blob/2efb8b4b644d5f3c28974a8f577081b4142decd2​:/cpan/CPAN/lib/CPAN/Distribution.pm#l2165

This simply ensures none of %ENV's keys are undef.

http​://perl5.git.perl.org/perl.git/blob/2efb8b4b644d5f3c28974a8f577081b4142decd2​:/cpan/ExtUtils- MakeMaker/t/lib/MakeMaker/Test/Utils.pm#l22

Doesn't modify any system variables.

http​://perl5.git.perl.org/perl.git/blob/2efb8b4b644d5f3c28974a8f577081b4142decd2​:/dist/PathTools/Cwd.pm#l220

This may not be cleaning up the environment enough.

http​://perl5.git.perl.org/perl.git/blob/2efb8b4b644d5f3c28974a8f577081b4142decd2​:/dist/PathTools/t/tmpdir.t#l25

That one might be a bit risky\, but File​::Spec​::Win32​::tmpdir doesn't call any external processes\, so it should be safe.

http​://perl5.git.perl.org/perl.git/blob/2efb8b4b644d5f3c28974a8f577081b4142decd2​:/ext/File- Glob/t/basic.t#l40

That seems safe.

Tony

[p5pRT]

From @bulk88

This is still an issue today with VC 2005 on WinXP.

I think the specific reason systemroot env var must be kept to start a VC 2005 compiled binary (automatically means that binary is SXS aware)\, the (WinXP but not newer) DLL loader when doing SXS passes string "%SystemRoot%\\WinSxS\\" to RtlExpandEnvironmentStrings_U() inside RtlpAssemblyStorageMapResolutionDefaultCallback. SystemRoot has to be defined or else the SXS code can't find the SXS directory where all the SXS tracked DLLs live.

-- bulk88 ~ bulk88 at hotmail.com

[p5pRT]

From @cpansprout

On Mon Sep 21 23​:49​:05 2015\, tonyc wrote​:

On Fri Sep 11 01​:13​:01 2015\, bulk88 wrote​:

There are 2.25 ways to fix this\, add SYSTEMROOT to the list of PATH and HOME being passed through\, or stop localizing and wiping all of %ENV and just localize LS_COLORS\,

Does the attached patch work for you?

or\, I dont like this fix "#ifndef WIN32\n/*localize/wipe %ENV */\n#endif" since that would fix it for me\, but doesn't fix the greater design problem of deleting env vars that you dont know why you are deleting them. So which way will this be fixed?

Another option would be to bring the mechanics of csh_glob from File- Glob into core\, similar to what was suggested in

https://rt-archive.perl.org/perl5/Ticket/Display.html?id=93638#txn-951162

To me that seems the safest mechanism\, it would eliminate the need for shelling out at all.

But it would be a lot more work. (And ‘work’ is a four-letter word.)

--

Father Chrysostomos

[p5pRT]

From @cpansprout

Inline Patch

```diff diff --git a/doio.c b/doio.c index 856b19a..2dc7082 100644 --- a/doio.c +++ b/doio.c @@ -2608,14 +2608,11 @@ Perl_vms_start_glob #endif /* !CSH */ #endif /* !DOSISH */ { - GV * const envgv = gv_fetchpvs("ENV", 0, SVt_PVHV); - SV ** const home = hv_fetchs(GvHV(envgv), "HOME", 0); - SV ** const path = hv_fetchs(GvHV(envgv), "PATH", 0); - if (home && *home) SvGETMAGIC(*home); - if (path && *path) SvGETMAGIC(*path); - save_hash(gv_fetchpvs("ENV", 0, SVt_PVHV)); - if (home && *home) SvSETMAGIC(*home); - if (path && *path) SvSETMAGIC(*path); + SV ** const svp = hv_fetchs(GvHVn(PL_envgv), "LS_COLORS", 0); + if (svp && *svp) + save_helem_flags(GvHV(PL_envgv), + newSVpvs_flags("LS_COLORS", SVs_TEMP), svp, + SAVEf_SETMAGIC); } (void)do_open6(PL_last_in_gv, SvPVX_const(tmpcmd), SvCUR(tmpcmd), NULL, NULL, 0); ```

[p5pRT]

From @bulk88

On Sun Jul 17 16​:19​:57 2016\, sprout wrote​:

On Mon Sep 21 23​:49​:05 2015\, tonyc wrote​:

On Fri Sep 11 01​:13​:01 2015\, bulk88 wrote​:

There are 2.25 ways to fix this\, add SYSTEMROOT to the list of PATH and HOME being passed through\, or stop localizing and wiping all of %ENV and just localize LS_COLORS\,

Does the attached patch work for you?

Yes.

-- bulk88 ~ bulk88 at hotmail.com

[p5pRT]

From @cpansprout

On Wed Aug 03 20​:43​:57 2016\, bulk88 wrote​:

On Sun Jul 17 16​:19​:57 2016\, sprout wrote​:

On Mon Sep 21 23​:49​:05 2015\, tonyc wrote​:

On Fri Sep 11 01​:13​:01 2015\, bulk88 wrote​:

There are 2.25 ways to fix this\, add SYSTEMROOT to the list of PATH and HOME being passed through\, or stop localizing and wiping all of %ENV and just localize LS_COLORS\,

Does the attached patch work for you?

Yes.

Thank you for confirming. I have applied it as acffc8a.

--

Father Chrysostomos

[p5pRT]

@cpansprout - Status changed from 'open' to 'pending release'

[p5pRT]

From @khwilliamson

Thank you for filing this report. You have helped make Perl better.

With the release today of Perl 5.26.0\, this and 210 other issues have been resolved.

Perl 5.26.0 may be downloaded via​: https://metacpan.org/release/XSAWYERX/perl-5.26.0

If you find that the problem persists\, feel free to reopen this ticket.

[p5pRT]

@khwilliamson - Status changed from 'pending release' to 'resolved'