t/op/taint.t fails test 846 on Windows #16921
Comments
From @steve-m-hayCreated by @steve-m-hayBlead is currently failing test 846 in t/op/taint.t on Windows (this I get the same result using VC12, VC14, VC14.1 and MinGW-w64 GCC 7.1.0, all x64. See attached verbose output. Perl Info |
From @steve-m-hayop/taint.t .. Test Summary Report op/taint.t (Wstat: 0 Tests: 1042 Failed: 1) |
From @iabynOn Wed, Apr 03, 2019 at 09:59:15AM -0700, Steve Hay (via RT) wrote:
I can't reproduce it on linux. The test is: { local $ENV{XX} = '\p{IsB}'; # Making it an environment variable taints it fresh_perl_like(<<'EOF', sub IsB { "42" }; This whiffs of stuff that Karl's been tweaking recently. -- |
|
The RT System itself - Status changed from 'new' to 'open' |
From @khwilliamsonOn 4/4/19 3:19 AM, Dave Mitchell wrote:
I can reproduce it on the WIndows box attached to dromedary. Is there |
From @iabynOn Thu, Apr 04, 2019 at 07:51:53AM -0600, Karl Williamson wrote:
Nothing that I'm aware of -- |
From @khwilliamsonOn 4/4/19 9:19 AM, Dave Mitchell wrote:
In particular it is this that isn't getting tainted: local $ENV{XX} = '\p{IsB}'; # Making it an environment variable taints it The other taint tests work that use different ways to taint, such as $prop = "IsA$TAINT"; |
From @khwilliamsonOn Thu, 04 Apr 2019 10:49:37 -0700, public@khwilliamson.com wrote:
I noticed that elsewhere in this .t, tests for ENV being tainted are skipped. Unskipping them causes them to fail. The commit that added the skip dates from the 1990's as part of an omnibus update, so there's no commit message. I suspect the answer to this ticket is to skip this test on the same platforms the other ENV tests are skipped for. |
From @steve-m-hayOn Fri, 12 Apr 2019 11:43:18 -0700, khw wrote:
Now skipped, in commit e8aa9ef. Leaving ticket open for now since it would be nice to understand why the skip is necessary. |
From @khwilliamsonOn 4/19/19 10:18 AM, Steve Hay via RT wrote:
I asked jhi, and he doesn't remember. He suggested Gurusamy Sarathy,
|
From @steve-m-hayOn Sat, 20 Apr 2019 00:05:49 -0700, public@khwilliamson.com wrote:
I found Gurusamy Sarathy, now Sarathy Naicker, on LinkedIn :-) I will ask him to take a quick look at this ticket in the hope that something here may jog his memory. Additional information, which I emailed to Karl (and Jan Dubois) but didn't put on this ticket yet: "I wonder if it might be at least partly related to the Perl Host thing (iperlsys.h), which I think (?) is only used on Windows (win32/perlhost.h). By default, perl on Windows is built with USE_IMP_SYS set to 'define' in the win32/Makefile. That #defines PERL_IMPLICIT_SYS, pulls in a couple of extra source files, including win32/perlhost.h, and results in all OS calls being encapsulated in interfaces like what PerlIO does for IO. For environment handling, the OS calls go through PerlEnv. Maybe that has some problem with taint? If you build with USE_IMP_SYS=define commented out then test 846 now passes, but unfortunately other tests (2, 4 and 5) that are normally skipped still fail when not skipped so this isn't the whole story." So the current situation is that tests 2, 3, 4, 5 and 846 are skipped because they fail in a default build, which has PERL_IMPLICIT_SYS defined. With PERL_IMPLICIT_SYS undefined tests 3 and 846 now pass, but tests 2, 4 and 5 still fail. |
From @steve-m-hayOn Mon, 22 Apr 2019 08:55:24 -0700, shay wrote:
Reply from Sarathy: "the Perl Host option probably uses an implementation that fetches from the environment in a dynamic fashion. The traditional unix notion of a contiguous environment block is not well-supported on windows, and is non-existent in virtualized environments like web servers. Off hand, I can't remember why that would change the semantics for tainting, but perhaps there isn't enough context for the taint flag to propagate when dynamically fetching and creating a temporary SV under those conditions? Might also be worth tracking down how the %ENV magic results in calling the equivalent of getenv() on windows to see at what point the taint propagation fails." |
Migrated from rt.perl.org#133980 (status was 'open')
Searchable as RT133980$
The text was updated successfully, but these errors were encountered: