Wrong aliasing of $_ #17119
Comments
From @krokiThis is a bug report for perl from tomash.brechko@gmail.com, Not tested with v5.30, sorry. With v5.28.2 the code below dies with $ echo 1 | perl -e ' f($_) for "a"; sub f { However the problem is more severe: in a real-world program (that I $/ = "\n"; warn() along with the values from @a also prints values read from $fh Still I believe both cases are instances of the same problem: Flags: Site configuration information for perl 5.28.2: Configured by Red Hat, Inc. at Tue Apr 23 08:23:13 UTC 2019. Summary of my perl5 (revision 5 version 28 subversion 2) configuration: Platform: Locally applied patches: @INC for perl 5.28.2: Environment for perl 5.28.2: PATH=/usr/share/Modules/bin:/usr/local/bin:/usr/local/sbin:/usr/bin:/usr/sbin:/home/tomash/.local/bin:/home/tomash/bin:/home/tomash/NOBACKUP/ANDROID/android-studio/bin:/home/tomash/.local/bin:/home/tomash/bin:/home/tomash/NOBACKUP/ANDROID/android-studio/bin |
From @tonycozOn Sun, 04 Aug 2019 02:23:49 -0700, tomash.brechko@gmail.com wrote:
$_ is global, when you alias $_ it's aliased for all code. When you alias $_ to a read-only value, and try to modify $_ the "Modification of a read-only value attempted at -e line 9." error is expected, not unusual. Adding "local $_;" before the while loop in f() fixes this.
Similarly here the while (<$fh>) in f() is modifying the elements of @a in the caller, again local fixes it. $_ can be convenient, but it's like any other global, it can lead to unexpected action at a distance. For complex code I tend to only use $_ with the aliasing done with operators like grep and map, and rarely with the statement modifier foreach. This isn't a bug. Tony |
|
The RT System itself - Status changed from 'new' to 'open' |
From @krokiOn Sun, 04 Aug 2019 17:06:52 -0700, tonyc wrote:
Ah, I see you point, thanks! Somehow I though that while() should cancel old aliasing, not merely assign through $_, but that was my misunderstanding. Then indeed, postfix for/foreach are quite dangerous. It seems I can't close this myself though... |
From @tonycozOn Sun, 04 Aug 2019 23:05:07 -0700, tomash.brechko@gmail.com wrote:
Closed. Tony |
|
@tonycoz - Status changed from 'open' to 'rejected' |
From Eirik-Berg.Hanssen@allverden.noOn Wed, Aug 7, 2019 at 5:59 AM Tomash Brechko via RT <
If I may, that seems to be the wrong lesson. There are other constructs The common factor that's dangerous is the C<< while (<...>) { ... } >> Eirik |
From @GrinnzOn Wed, Aug 7, 2019 at 11:46 AM Eirik Berg Hanssen <
Harmless to code that calls it, but code it calls can still be dangerous as -Dan |
From Eirik-Berg.Hanssen@allverden.noOn Wed, Aug 7, 2019 at 5:52 PM Dan Book <grinnz@gmail.com> wrote:
Globals are always trouble, hence the "(relatively)". This particular
But sure, I can agree C<< while (my $l = <...>) { ... } >> (or similar) So, fair enough. When next I encounter this, I'll recommend the lexical Eirik |
From @krokiср, 7 авг. 2019 г. в 19:18, Eirik Berg Hanssen via RT <
For good or bad, we don't always control every function we call, so any f($_) for @a; is a potential risk unless you ensured that f() is safe. The following works: foreach ("a", "b") { Likewise, you can safely nest map/grep/whatever, so I still think it's fair I appreciate your advise, but in the end it boils down to "the feature is |
From @GrinnzOn Thu, Aug 8, 2019 at 12:17 AM Tomash Brechko <tomash.brechko@gmail.com>
Just for the sake of driving the point home, "f() for @a" is the potential -Dan |
From @krokiOn Wed, 07 Aug 2019 21:29:03 -0700, grinnz@gmail.com wrote:
I see what you mean, yet I'd like to disagree. $_ is not just like any other global var, it is special and is meant to be used implicitly. In previous posts I provided some code for the sake of example, but in real life it often looks like f(split /.../) for @a; i.e. $_ is assumed but never mentioned. Even when we do mention it, like @a = map { ">$_<" } @b; then it is an rvalue, and nested foreach/map/grep/etc. do localize $_, so no danger here. And if one ever finds him/herself writing $_ = ...; then I agree that it should be replaced with a real variable (and hopefully not global). Now, when I write sub f { I do not violate "The Protocol" of using $_ implicitly, this is what while(<>) is for. However, as other pointed out, in order to allow f() to be nested in other loops I _have_ to "patch" such _implicit_ use of $_ with preceding _explicit_ local $_, which circumvents the implicit nature of $_ and shows that while(<>) violates "Do The Right Thing" motto. Other implicit looping construct do not have this flaw because they alias, or perhaps otherwise: while(<>) wished to alias too, but it doesn't have rvalues to alias to, so it assigns to $_ *without localizing it first as other constructs do*: for ("old") { Thus I judge it is a bug, but as with some other real bugs in their early denial stage it too may be rejected for a while, not argue against that :). |
Migrated from rt.perl.org#134337 (status was 'rejected')
Searchable as RT134337$
The text was updated successfully, but these errors were encountered: