Test oneliner without double escaping dollar sign

[atoomic]

Fix #355

Double escaping '$' for Makefile usage is fine, but
we should not use such a syntax while testing and trying
to run the oneliner output.

This change is 'unescaping' the '$$' to '$' so we can perform
some extra checks while using oneliner.

Otherwise as shown in this example a simple 'my $foo' test will fail.

[Leont]

LGTM.

Technically it's wrong if the path to perl contains two dollar signs, moving it one line up would fix that unlikely edge-case.

[atoomic]

good point... was not considering it being an option, but you are right and I moved it above via db19381

[mohawk2]

Generally I favour this change.

Unfortunately, the CI fail reveals a real problem: dmake has to double up its braces. If you want to unquote them, you'll need to reverse that too.

I feel like try_oneliner is trying to be too clever by trying to de-quote the make quoting, and it's getting it wrong. Surely the right thing to do is create small makefiles that have these recipes, and just use make to run them?

[Leont]

I feel like try_oneliner is trying to be too clever by trying to de-quote the make quoting, and it's getting it wrong. Surely the right thing to do is create small makefiles that have these recipes, and just use make to run them?

And apparently VMS also does its own unique thing. Writing out a Makefile seems like the way to go.

[atoomic]

nice catch and this comes from the additional test using a hash I've added :-)
let me see if unescaping those braces fixes it

syntax error at -e line 1, near "{{"
Execution of -e aborted due to compilation errors.

#   Failed test '%h and $h'
#   at t/oneliner.t line 42.
#          got: ''
#     expected: '2'
# oneliner:
# "C:\strawberry\perl\bin\perl.exe"  -e "my %h = (1, 2); print $h{{1}}" --
# Looks like you failed 1 test of 17.
[21:25:20] t/oneliner.t .............. 
Dubious, test returned 1 (wstat 256, 0x100)
Failed 1/17 subtests 

[atoomic]

going to work on the Makefile solution

[Leont]

going to work on the Makefile solution

On VMS, the Makefile needs to be called Descrip.MMS AFAIK. Possibly other operating systems have similar quirks.

[atoomic]

I see `makefile_name` helper from t/lib/MakeMaker/Test/Utils.pm, also going to use the `run` helper let's wait for CI return

[craigberry]

On Jul 22, 2020, at 11:13 AM, Nicolas R. ***@***.***> wrote: I see `makefile_name` from t/lib/MakeMaker/Test/Utils.pm

Right, use that one in the tests and in MakeMaker itself use $self->{MAKEFILE}.

…

On Wed, Jul 22, 2020 at 10:11 AM Leon Timmermans ***@***.***> wrote: > going to work on the Makefile solution > > On VMS, the Makefile needs to be called Descrip.MMS AFAIK. Possibly other > operating systems have similar quirks.

________________________________________ Craig A. Berry "... getting out of a sonnet is much more difficult than getting in." Brad Leithauser

[Leont]

This will not be correct on all platforms

[atoomic]

Indeed you are right. This is kind a of a chicken egg issue: producing Makefile to test code which generate Makefile..
What would be your preference?

• do we have some helpers we could use for this?
• skipping some distro for this test?
• going back to the unescape solution?
• other?

thanks for your valuable input

[mohawk2]

Skipping and unescaping are not viable approaches. Probably the way to go here is to be making a sandbox dir, then minimal Makefile.PL. Look in the XS tests for functions to make this easy.