From: Father Chrysostomos <sprout@cpan.org>
Date: Fri, 6 Feb 2015 14:00:09 +0000 (+0100)
Subject: Compatibility fix for perl 5.21.6
X-Git-Tag: rt100202^0
X-Git-Url: http://git.vpit.fr/?a=commitdiff_plain;h=881d1aecb5e9948bbd7a384deffbab560eab4ef4;hp=881d1aecb5e9948bbd7a384deffbab560eab4ef4;p=perl%2Fmodules%2FLexical-Types.git

Compatibility fix for perl 5.21.6

Given

    my Int ($x, ($y, $z), $t) = ("x", "y");

In 5.20.1 it parses as:

6     <;> nextstate(main 2 -e:1) v:{ ->7
e     <2> aassign[t5] vKS ->f
-        <1> ex-list lKP ->a
7           <0> pushmark s ->8
8           <$> const(PV "x") s ->9
9           <$> const(PV "y") s ->a
-        <1> ex-list lKPRM* ->e
a           <0> padrange[$x:2,3] lRM/LVINTRO,1 ->b
-           <0> padsv[$x:2,3] lRM*/LVINTRO ->b
c           <@> list lKPRM* ->d
b              <0> padrange[$y:2,3; $z:2,3] lRM/LVINTRO,2 ->c
-              <0> padsv[$y:2,3] lRM*/LVINTRO ->-
-              <0> padsv[$z:2,3] lRM*/LVINTRO ->c
d           <0> padsv[$t:2,3] lRM*/LVINTRO ->e

In 5.21.6-to-be:

6     <;> nextstate(main 3 -e:1) v:{ ->7
b     <2> aassign[t5] vKS ->c
-        <1> ex-list lKP ->a
7           <0> pushmark s ->8
8           <$> const[PV "x"] s ->9
9           <$> const[PV "y"] s ->a
-        <1> ex-list lKPRM* ->b
a           <0> padrange[$x:3,4; $y:3,4; $z:3,4; $t:3,4] lRM/LVINTRO,4 ->b
-           <0> padsv[$x:3,4] lRM*/LVINTRO ->-
-           <1> ex-list lKPRM* ->-
-              <0> ex-pushmark sRM*/LVINTRO ->-
-              <0> padsv[$y:3,4] lRM*/LVINTRO ->-
-              <0> padsv[$z:3,4] lRM*/LVINTRO ->-
-           <0> padsv[$t:3,4] lRM*/LVINTRO ->-

The inner list is optimised away, but still visible in the op tree.

Also, I noticed that this comment in Types.xs is wrong:

   /* In a padrange sequence, either all lexicals are typed, or none are.
    * Thus we can stop at the first padsv op. However, note that these
    * lexicals can need to call different methods in different packages. */

$ perl5.20.1 -MO=Concise -e '(my main $a, my $b, my main $c) = ()'
6  <@> leave[1 ref] vKP/REFC ->(end)
1     <0> enter ->2
2     <;> nextstate(main 1 -e:1) v:{ ->3
5     <2> aassign[t4] vKS ->6
-        <1> ex-list lK ->4
3           <0> pushmark s ->4
-           <0> stub lP ->-
-        <1> ex-list lKPRM* ->5
4           <0> padrange[$a:1,2; $b:1,2; $c:1,2] lRM/LVINTRO,3 ->5
-           <0> padsv[$a:1,2] lRM*/LVINTRO ->-
-           <0> padsv[$b:1,2] lRM*/LVINTRO ->-
-           <0> padsv[$c:1,2] lRM*/LVINTRO ->-
-e syntax OK

I don’t know whether that has any bearing on the correctness of the code,
but I suspect there is a bug.

The attached patch gets all the tests passing, but I don’t know whether I
caught everything. Note that there are cases where the sibling of the
padrange is actually a null op :

$ ./perl -ILib  -MO=Concise -e '@a[my $x, my $y, my $z]'
7  <@> leave[1 ref] vKP/REFC ->(end)
1     <0> enter ->2
2     <;> nextstate(main 1 -e:1) v:{ ->3
6     <@> aslice vK ->7
3        <0> padrange[$x:1,2; $y:1,2; $z:1,2] l/LVINTRO,3 ->4
-        <1> ex-list lK ->4
-           <0> ex-pushmark s ->-
-           <0> padsv[$x:1,2] lM/LVINTRO ->-
-           <0> padsv[$y:1,2] lM/LVINTRO ->-
-           <0> padsv[$z:1,2] lM/LVINTRO ->-
5        <1> rv2av[t2] sKR/1 ->6
4           <#> gv[*a] s ->5
-e syntax OK

This fixes RT #100202.
---