Concat with uninitialized vars

[p5pRT]

Migrated from rt.perl.org#127877 (status was 'resolved')

Searchable as RT127877$

[p5pRT]

From @VadimPushtaev

We discovered that `my` and `our` variables behave differently in terms of using uninitialized vars in concatenation.

Try this​:

perl -e 'use warnings q{uninitialized}; my $x; $x .= "xxx"'
perl -e 'use warnings q{uninitialized}; my $x; $x = $x . "xxx"'
perl -e 'use warnings q{uninitialized}; my $x; $x .= "xxx" . undef'
perl -e 'use warnings q{uninitialized}; my $x; $x = $x . "xxx" . undef'
perl -e 'use warnings q{uninitialized}; our $x; $x .= "xxx"'
perl -e 'use warnings q{uninitialized}; our $x; $x = $x . "xxx"'
perl -e 'use warnings q{uninitialized}; our $x; $x .= "xxx" . undef'
perl -e 'use warnings q{uninitialized}; our $x; $x = $x . "xxx" . undef'

Get this​:

1) perl -e 'use warnings q{uninitialized}; my $x; $x .= "xxx"'

2) perl -e 'use warnings q{uninitialized}; my $x; $x = $x . "xxx"'
# None here. WHY!?

3) perl -e 'use warnings q{uninitialized}; my $x; $x .= "xxx" . undef'
Use of uninitialized value in concatenation (.) or string at -e line 1.

4) perl -e 'use warnings q{uninitialized}; my $x; $x = $x . "xxx" . undef'
Use of uninitialized value $x in concatenation (.) or string at -e line 1.
Use of uninitialized value in concatenation (.) or string at -e line 1.

5) perl -e 'use warnings q{uninitialized}; our $x; $x .= "xxx"'

6) perl -e 'use warnings q{uninitialized}; our $x; $x = $x . "xxx"'
Use of uninitialized value $x in concatenation (.) or string at -e line 1.

7) perl -e 'use warnings q{uninitialized}; our $x; $x .= "xxx" . undef'
Use of uninitialized value in concatenation (.) or string at -e line 1.

8) perl -e 'use warnings q{uninitialized}; our $x; $x = $x . "xxx" . undef'
Use of uninitialized value $x in concatenation (.) or string at -e line 1.
Use of uninitialized value in concatenation (.) or string at -e line 1.

There is no consistency in this behavior​:
* (2) and (6) behave differently though the only difference is `my` and `our`.
* If (2) doesn't throw the warning, why does (4) throw TWO of them?

Btw, there is no such problem with the addition (`+`).

P.S. I also don't quite understand why `.=` doesn't throw any warnings, but this behavior is understandable and consistent, so no problem here.

P.P.S. The problem was originally discovered by ivan.yazykov@​gmail.com

[p5pRT]

From @demerphq

On 11 April 2016 at 12​:56, Vadim Pushtaev <perlbug-followup@​perl.org> wrote​:

# New Ticket Created by Vadim Pushtaev
# Please include the string​: [perl #127877]
# in the subject line of all future correspondence about this issue.
# <URL​: https://rt-archive.perl.org/perl5/Ticket/Display.html?id=127877 >

We discovered that `my` and `our` variables behave differently in terms of using uninitialized vars in concatenation.

Try this​:

perl -e 'use warnings q{uninitialized}; my $x; $x .= "xxx"'
perl -e 'use warnings q{uninitialized}; my $x; $x = $x . "xxx"'
perl -e 'use warnings q{uninitialized}; my $x; $x .= "xxx" . undef'
perl -e 'use warnings q{uninitialized}; my $x; $x = $x . "xxx" . undef'
perl -e 'use warnings q{uninitialized}; our $x; $x .= "xxx"'
perl -e 'use warnings q{uninitialized}; our $x; $x = $x . "xxx"'
perl -e 'use warnings q{uninitialized}; our $x; $x .= "xxx" . undef'
perl -e 'use warnings q{uninitialized}; our $x; $x = $x . "xxx" . undef'

Get this​:

1) perl -e 'use warnings q{uninitialized}; my $x; $x .= "xxx"'

This is expected.

2) perl -e 'use warnings q{uninitialized}; my $x; $x = $x . "xxx"'
# None here. WHY!?

Yes, agreed. I find this not warning to be odd, and a bug. I wonder if
it is a regression.

3) perl -e 'use warnings q{uninitialized}; my $x; $x .= "xxx" . undef'
Use of uninitialized value in concatenation (.) or string at -e line 1.

This is expected.

4) perl -e 'use warnings q{uninitialized}; my $x; $x = $x . "xxx" . undef'
Use of uninitialized value $x in concatenation (.) or string at -e line 1.
Use of uninitialized value in concatenation (.) or string at -e line 1.

This is expected.

5) perl -e 'use warnings q{uninitialized}; our $x; $x .= "xxx"'

This is expected.

6) perl -e 'use warnings q{uninitialized}; our $x; $x = $x . "xxx"'
Use of uninitialized value $x in concatenation (.) or string at -e line 1.

This is expected. And is what I expect #2 to do.

7) perl -e 'use warnings q{uninitialized}; our $x; $x .= "xxx" . undef'
Use of uninitialized value in concatenation (.) or string at -e line 1.

This is expected.

8) perl -e 'use warnings q{uninitialized}; our $x; $x = $x . "xxx" . undef'
Use of uninitialized value $x in concatenation (.) or string at -e line 1.
Use of uninitialized value in concatenation (.) or string at -e line 1.

Also expected.

There is no consistency in this behavior​:
* (2) and (6) behave differently though the only difference is `my` and `our`.

Yes, I agree this is a bug.

* If (2) doesn't throw the warning, why does (4) throw TWO of them?

2 is buggy. What it does should not be used to determine what 4 does.

We get two warnings because there are two undefined values involved.

my $x; $x = $x . "xxx" . undef

The first is $x starting off undefined, and the second is the undef call.

Btw, there is no such problem with the addition (`+`).

Interesting.

P.S. I also don't quite understand why `.=` doesn't throw any warnings, but this behavior is understandable and consistent, so no problem here.

.= is a mutator. Most mutators DONT warn on uninitialized value on the
lhs but a few do (IIRC it DWIM pretty well). This is documented
somewhere in the perldocs.

That is why ++ does not warn, and why +=1 doesnt warn, etc.

So the only bug here is that

perl -e 'use warnings q{uninitialized}; my $x; $x = $x . "xxx"'

does not warn when it should.

Yves

--
perl -Mre=debug -e "/just|another|perl|hacker/"

[p5pRT]

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

[p5pRT]

From @VadimPushtaev

On Mon Apr 11 05​:01​:17 2016, demerphq wrote​:

On 11 April 2016 at 12​:56, Vadim Pushtaev <perlbug-followup@​perl.org>
wrote​:

# New Ticket Created by Vadim Pushtaev
# Please include the string​: [perl #127877]
# in the subject line of all future correspondence about this issue.
# <URL​: https://rt-archive.perl.org/perl5/Ticket/Display.html?id=127877 >

We discovered that `my` and `our` variables behave differently in
terms of using uninitialized vars in concatenation.

Try this​:

perl -e 'use warnings q{uninitialized}; my $x; $x .= "xxx"'
perl -e 'use warnings q{uninitialized}; my $x; $x = $x . "xxx"'
perl -e 'use warnings q{uninitialized}; my $x; $x .= "xxx" . undef'
perl -e 'use warnings q{uninitialized}; my $x; $x = $x . "xxx" .
undef'
perl -e 'use warnings q{uninitialized}; our $x; $x .= "xxx"'
perl -e 'use warnings q{uninitialized}; our $x; $x = $x . "xxx"'
perl -e 'use warnings q{uninitialized}; our $x; $x .= "xxx" . undef'
perl -e 'use warnings q{uninitialized}; our $x; $x = $x . "xxx" .
undef'

Get this​:

1) perl -e 'use warnings q{uninitialized}; my $x; $x .= "xxx"'

This is expected.

2) perl -e 'use warnings q{uninitialized}; my $x; $x = $x . "xxx"'
# None here. WHY!?

Yes, agreed. I find this not warning to be odd, and a bug. I wonder if
it is a regression.

3) perl -e 'use warnings q{uninitialized}; my $x; $x .= "xxx" .
undef'
Use of uninitialized value in concatenation (.) or string at -e line
1.

This is expected.

4) perl -e 'use warnings q{uninitialized}; my $x; $x = $x . "xxx" .
undef'
Use of uninitialized value $x in concatenation (.) or string at -e
line 1.
Use of uninitialized value in concatenation (.) or string at -e line
1.

This is expected.

5) perl -e 'use warnings q{uninitialized}; our $x; $x .= "xxx"'

This is expected.

6) perl -e 'use warnings q{uninitialized}; our $x; $x = $x . "xxx"'
Use of uninitialized value $x in concatenation (.) or string at -e
line 1.

This is expected. And is what I expect #2 to do.

7) perl -e 'use warnings q{uninitialized}; our $x; $x .= "xxx" .
undef'
Use of uninitialized value in concatenation (.) or string at -e line
1.

This is expected.

8) perl -e 'use warnings q{uninitialized}; our $x; $x = $x . "xxx" .
undef'
Use of uninitialized value $x in concatenation (.) or string at -e
line 1.
Use of uninitialized value in concatenation (.) or string at -e line
1.

Also expected.

There is no consistency in this behavior​:
* (2) and (6) behave differently though the only difference is `my`
and `our`.

Yes, I agree this is a bug.

* If (2) doesn't throw the warning, why does (4) throw TWO of them?

2 is buggy. What it does should not be used to determine what 4 does.

We get two warnings because there are two undefined values involved.

Yes, that's exactly what I meant.

my $x; $x = $x . "xxx" . undef

The first is $x starting off undefined, and the second is the undef
call.

Btw, there is no such problem with the addition (`+`).

Interesting.

P.S. I also don't quite understand why `.=` doesn't throw any
warnings, but this behavior is understandable and consistent, so no
problem here.

.= is a mutator. Most mutators DONT warn on uninitialized value on the
lhs but a few do (IIRC it DWIM pretty well). This is documented
somewhere in the perldocs.

I know it's unrelated, but I still can't find it in perdocs.

That is why ++ does not warn, and why +=1 doesnt warn, etc.

So the only bug here is that

perl -e 'use warnings q{uninitialized}; my $x; $x = $x . "xxx"'

does not warn when it should.

Yves

[p5pRT]

From @demerphq

On 11 April 2016 at 14​:17, Vadim Pushtaev via RT
<perlbug-followup@​perl.org> wrote​:

On Mon Apr 11 05​:01​:17 2016, demerphq wrote​:

.= is a mutator. Most mutators DONT warn on uninitialized value on the
lhs but a few do (IIRC it DWIM pretty well). This is documented
somewhere in the perldocs.

I know it's unrelated, but I still can't find it in perdocs.

It is in perlsyn, section "Declarations"​:

Operators such as C<++>, C<-->, C<+=>,
C<-=>, and C<.=>, that operate on undefined variables such as​:

  undef $a;
  $a++;

are also always exempt from such warnings.

--
perl -Mre=debug -e "/just|another|perl|hacker/"

[p5pRT]

From @arc

demerphq <demerphq@​gmail.com> wrote​:

So the only bug here is that

perl -e 'use warnings q{uninitialized}; my $x; $x = $x . "xxx"'

does not warn when it should.

Fix attached; I plan to apply it early in the 5.25 cycle.

--
Aaron Crane ** http​://aaroncrane.co.uk/

[p5pRT]

From @arc

0001-perl-127877-Emit-undef-warning-on-sassign-concat.patch

From 22fd7a732f3df8b11bbf167562a79756e3a34df0 Mon Sep 17 00:00:00 2001
From: Aaron Crane <arc@cpan.org>
Date: Sun, 8 May 2016 18:13:30 +0100
Subject: [PATCH] [perl #127877] Emit undef warning on sassign+concat

Code like this:

    my $x;
    $x .= 'a';

is specifically exempted from "use of uninitialized value" warnings,
according to the "Declarations" section of perlsyn, to allow the idiom of
building up a value piecemeal. The same is true of the += and -= operators.
However, breaking the combined assignment up into the underlying operator
and a simple assignment, as in this code:

    my $x;
    $x = $x . 'a';

*should* produce a warning.

That warning was correctly being emitted for addition and subtraction, but
concatenation was behaving as the ".=" case, because "$x = $x . EXPR" is
optimized to the equivalent of "$x .= EXPR".

So we now explicitly detect this case, and emit the desired warning.
---
 pp_hot.c               |  7 +++++--
 t/lib/warnings/9uninit | 21 +++++++++++++++++++++
 2 files changed, 26 insertions(+), 2 deletions(-)

diff --git a/pp_hot.c b/pp_hot.c
index d6cb1aa..223169b 100644
--- a/pp_hot.c
+++ b/pp_hot.c
@@ -284,8 +284,11 @@ PP(pp_concat)
     }
     else { /* $l .= $r   and   left == TARG */
 	if (!SvOK(left)) {
-	    if (left == right && ckWARN(WARN_UNINITIALIZED)) /* $l .= $l */
-		report_uninit(right);
+            if ((left == right                          /* $l .= $l */
+                 || (PL_op->op_private & OPpTARGET_MY)) /* $l = $l . $r */
+                && ckWARN(WARN_UNINITIALIZED)
+                )
+                report_uninit(left);
 	    sv_setpvs(left, "");
 	}
         else {
diff --git a/t/lib/warnings/9uninit b/t/lib/warnings/9uninit
index ef9b4f6..42915d9 100644
--- a/t/lib/warnings/9uninit
+++ b/t/lib/warnings/9uninit
@@ -2138,3 +2138,24 @@ Use of uninitialized value $i in array element at - line 12.
 Use of uninitialized value $k in hash element at - line 12.
 Use of uninitialized value $i in array element at - line 13.
 Use of uninitialized value $k in hash element at - line 13.
+########
+# perl #127877
+use warnings 'uninitialized';
+my ($p, $q, $r, $s, $t, $u, $v, $w, $x, $y);
+$p = $p . "a";
+$q .= "a";
+$r = $r + 17;
+$s += 17;
+$t = $t - 17;
+$u -= 17;
+use integer;
+$v = $v + 17;
+$w += 17;
+$x = $x - 17;
+$y -= 17;
+EXPECT
+Use of uninitialized value $p in concatenation (.) or string at - line 4.
+Use of uninitialized value $r in addition (+) at - line 6.
+Use of uninitialized value $t in subtraction (-) at - line 8.
+Use of uninitialized value $v in integer addition (+) at - line 11.
+Use of uninitialized value $x in integer subtraction (-) at - line 13.
-- 
2.7.4

[p5pRT]

From @arc

Applied as 51f69a2; resolving ticket.

--
Aaron Crane ** http​://aaroncrane.co.uk/

[p5pRT]

@arc - Status changed from 'open' to 'resolved'

[p5pRT]

From @VadimPushtaev

On Sun May 08 10​:26​:07 2016, arc wrote​:

demerphq <demerphq@​gmail.com> wrote​:

So the only bug here is that

perl -e 'use warnings q{uninitialized}; my $x; $x = $x . "xxx"'

does not warn when it should.

Fix attached; I plan to apply it early in the 5.25 cycle.

Thanks! We actually suspected that some optimization trick involed :).