106 lines
3.1 KiB
Diff
106 lines
3.1 KiB
Diff
|
From 695d6585affc8f13711f013329fb4810ab89d833 Mon Sep 17 00:00:00 2001
|
|||
|
From: Father Chrysostomos <sprout@cpan.org>
|
|||
|
Date: Tue, 14 Nov 2017 18:55:55 -0800
|
|||
|
Subject: [PATCH] [perl #132442] Fix stack with do {my sub l; 1}
|
|||
|
MIME-Version: 1.0
|
|||
|
Content-Type: text/plain; charset=UTF-8
|
|||
|
Content-Transfer-Encoding: 8bit
|
|||
|
|
|||
|
A block in perl usually compiles to a leave op with an enter inside
|
|||
|
it, followed by the statements:
|
|||
|
|
|||
|
leave
|
|||
|
enter
|
|||
|
nextstate
|
|||
|
... expr ...
|
|||
|
nextstate
|
|||
|
... expr ...
|
|||
|
|
|||
|
If a block contains only one statement, and that statement is suffic-
|
|||
|
iently innocuous, then the enter/leave pair to create the scope at run
|
|||
|
time get skipped, and instead we have a simple scope op which is not
|
|||
|
even executed:
|
|||
|
|
|||
|
scope
|
|||
|
ex-nextstate
|
|||
|
... expr ...
|
|||
|
|
|||
|
The nextstate in this case also gets nulled.
|
|||
|
|
|||
|
In the case of do { my sub l; 1 } we were getting a variation of the
|
|||
|
latter, that looked like this:
|
|||
|
|
|||
|
scope
|
|||
|
introcv
|
|||
|
clonecv
|
|||
|
nextstate
|
|||
|
... expr ...
|
|||
|
|
|||
|
The problem here is that nextstate resets the stack, even though a new
|
|||
|
scope has not been pushed, so we end up with all existing stack items
|
|||
|
from the *outer* scope getting clobbered.
|
|||
|
|
|||
|
One can have fun with this and erase everything pushed on to the stack
|
|||
|
so far in a given statement:
|
|||
|
|
|||
|
$ ./perl -le 'print join "-", 1..10, do {my sub l; ","}, 11..20'
|
|||
|
11,12,13,14,15,16,17,18,19,20
|
|||
|
|
|||
|
Here I replaced the first argument to join() from within the do{}
|
|||
|
block, after having cleared the stack.
|
|||
|
|
|||
|
Why was the op tree was getting muddled up like this? The ‘my sub’
|
|||
|
declaration does not immediately add any ops to the op tree; those ops
|
|||
|
get added when the current scope finishing compiling, since those ops
|
|||
|
must be inserted at the beginning of the block.
|
|||
|
|
|||
|
I have not fully looked into the order that things happen, and why the
|
|||
|
nextstate op does not get nulled; but it did not matter, because of
|
|||
|
the simple fix: Treat lexical sub declarations as ‘not innocuous’ by
|
|||
|
setting the HINT_BLOCK_SCOPE flag when a lexical sub is declared.
|
|||
|
Thus, we end up with an enter/leave pair, which creates a
|
|||
|
proper scope.
|
|||
|
|
|||
|
Signed-off-by: Petr Písař <ppisar@redhat.com>
|
|||
|
---
|
|||
|
op.c | 2 ++
|
|||
|
t/op/lexsub.t | 5 ++++-
|
|||
|
2 files changed, 6 insertions(+), 1 deletion(-)
|
|||
|
|
|||
|
diff --git a/op.c b/op.c
|
|||
|
index 8fa5aad876..c617ad2a00 100644
|
|||
|
--- a/op.c
|
|||
|
+++ b/op.c
|
|||
|
@@ -9243,6 +9243,8 @@ Perl_newMYSUB(pTHX_ I32 floor, OP *o, OP *proto, OP *attrs, OP *block)
|
|||
|
|
|||
|
PERL_ARGS_ASSERT_NEWMYSUB;
|
|||
|
|
|||
|
+ PL_hints |= HINT_BLOCK_SCOPE;
|
|||
|
+
|
|||
|
/* Find the pad slot for storing the new sub.
|
|||
|
We cannot use PL_comppad, as it is the pad owned by the new sub. We
|
|||
|
need to look in CvOUTSIDE and find the pad belonging to the enclos-
|
|||
|
diff --git a/t/op/lexsub.t b/t/op/lexsub.t
|
|||
|
index 3fa17acdda..f085cd97e8 100644
|
|||
|
--- a/t/op/lexsub.t
|
|||
|
+++ b/t/op/lexsub.t
|
|||
|
@@ -7,7 +7,7 @@ BEGIN {
|
|||
|
*bar::is = *is;
|
|||
|
*bar::like = *like;
|
|||
|
}
|
|||
|
-plan 149;
|
|||
|
+plan 150;
|
|||
|
|
|||
|
# -------------------- our -------------------- #
|
|||
|
|
|||
|
@@ -957,3 +957,6 @@ like runperl(
|
|||
|
{
|
|||
|
my sub h; sub{my $x; sub{h}}
|
|||
|
}
|
|||
|
+
|
|||
|
+is join("-", qw(aa bb), do { my sub lleexx; 123 }, qw(cc dd)),
|
|||
|
+ "aa-bb-123-cc-dd", 'do { my sub...} in a list [perl #132442]';
|
|||
|
--
|
|||
|
2.13.6
|
|||
|
|