Double call with --doc

[finanalyst]

The Problem

Developing a new module to interface with the --doc option of perl6 CLI calls render twice.

This is an issue because processing pod to html can be quite time-consuming. Since render is called twice, the processing is done twice.

Expected Behavior

Given a simple module Pod::To::HTML2 containing

unit class Pod::To::HTML2;

method render( $pod-tree ) is export {
    say "At $?LINE tree is ", $pod-tree.perl;
}

and a test file simple.p6 containing

say "Hello World"
=begin pod
   this is some text
=end pod

I would expect the pod-block structure to be output once.

Actual Behavior

$ perl6 -Ip6-pod-render/lib --doc=HTML2 simple.p6
At 4 tree is $[Pod::Block::Named.new(name => "pod", config => {}, contents => [Pod::Block::Code.new(allowed => [], config => {}, contents => ["this is some text"])])]
At 4 tree is $[Pod::Block::Named.new(name => "pod", config => {}, contents => [Pod::Block::Code.new(allowed => [], config => {}, contents => ["this is some text"])])]
True

Steps to Reproduce

Nothing more to add

Environment

• Operating system: uname -a
  Linux merlin 4.18.0-13-generic Io additions #14~18.04.1-Ubuntu SMP Thu Dec 6 14:09:52 UTC 2018 x86_64 x86_64 x86_64 GNU/Linux

• Compiler version (perl6 -v): perl6 -v
  This is Rakudo Star version 2018.10 built on MoarVM version 2018.10
  implementing Perl 6.c.

[Altai-man]

The likely reason is that https://github.com/rakudo/rakudo/blob/master/src/Perl6/Actions.nqp#L1416 <- does not check if the phaser is already installed.
It is an always matching rule in https://github.com/rakudo/rakudo/blob/master/src/Perl6/Grammar.nqp#L833 and I am not sure how much is it matched, but given that the render is called twice...

Any ideas on how to neatly add a check to the method?

[Altai-man]

AST for perl6 --target=ast --doc=Text a.pm6 having the same two calls in a single place makes the phaser hypothesis more reliable:

            - QAST::Block(:cuid(3))  :outer<?> :in_stmt_mod<?> :code_object<?>
              - QAST::Stmts  
              - QAST::Op(if) 
                - QAST::Op(callmethod render)  
                  - QAST::Op(call &INDIRECT_NAME_LOOKUP) 
                    - QAST::Op(callmethod new) 
                      - QAST::WVal(PseudoStash) 
                    - QAST::Want 
                      - QAST::WVal(Str) 
                      - Ss
                      - QAST::SVal(Pod::To::Text) 
                  - QAST::Var(lexical $=pod)  
                - QAST::Op(call &say)  
                  - QAST::Op(callmethod render)  
                    - QAST::Op(call &INDIRECT_NAME_LOOKUP) 
                      - QAST::Op(callmethod new) 
                        - QAST::WVal(PseudoStash) 
                      - QAST::Want 
                        - QAST::WVal(Str) 
                        - Ss
                        - QAST::SVal(Pod::To::Text) 
                    - QAST::Var(lexical $=pod)  
              - QAST::Op(call &exit)  
            - QAST::Op(callmethod BIND-KEY) 
              - QAST::Var(lexical !INIT_VALUES) 
              - QAST::SVal(3) 
              - QAST::Op(call) 
                - QAST::WVal(Block) 
            - QAST::Var(lexical $=pod :decl(static)) 

[tbrowder]

Does that have anything to do with the optimizer? I think the source code gets a second parse pass and I've just noticed surprisingly some debug code in src/Perl6/Grammar.nqp outputs twice during a run.

UPDATE The ast stage is before optimization so we should look at the optimize stage.

I've also tried commenting out the always-matched line for my project and didn't see any difference. I know a little more now than when Richard first posted this, and the issue may have an intersection with my problems. (But looking at QAST causes my eyes to glaze over. Ah well, in with a penny, in with a pound.)