Bug #86891 for Type-Tiny: Reduce boilerplate for inline

Fri Jul 12 09:29:19 2013 TIMB [...] cpan.org - Ticket created

Subject:

Reduce boilerplate for inline_as

Instead of: inline_as { my ($constraint, $varname) = @_; my $perlcode = $constraint->parent->inline_check($varname) . "&& ($varname >= -32768 && $varname <= 32767)"; return $perlcode; }, I'd really like to avoid all that boilerplate and be able to write something like: inline_as { return (undef, "$_ >= -32768 && $_ <= 32767"); }, This build on three things: 1: allow returning a list of strings that will be merged via: join "&&", map { "($_)" } to form the inline condition 2: allow an initial undef to mean "whatever $constraint->parent->inline_check($varname)" returns 3: call the inline_as code with $_ set to $varname

Fri Jul 12 17:52:55 2013 perl [...] toby.ink - Correspondence added

On 2013-07-12T14:29:19+01:00, TIMB wrote: Show quoted text

> 3: call the inline_as code with $_ set to $varname

This bit should already work. Overall, I'd like to keep inline_as much as it is, because that's how inline_as works in Moose and Mouse. However, inline_as is just a shortcut for setting the "inlined" property of the declared type constraint; and there's nothing to rule out having more than one shortcut function.

Fri Jul 12 17:52:55 2013 The RT System itself - Status changed from 'new' to 'open'

Fri Jul 12 17:52:55 2013 perl [...] toby.ink - Taken

Fri Jul 12 18:18:37 2013 perl [...] toby.ink - Correspondence added

That said, your suggestions could probably be implemented within the existing functions without breaking things.

Sat Jul 13 09:12:04 2013 perl [...] toby.ink - Correspondence added

Implemented this experimentally in 0.015_04. I will close this bug when a new stable release is available on CPAN.

Sat Jul 13 09:12:04 2013 perl [...] toby.ink - Status changed from 'open' to 'patched'

Tue Jul 16 11:36:26 2013 perl [...] toby.ink - Correspondence added

resolved in 0.016

Tue Jul 16 11:36:26 2013 perl [...] toby.ink - Status changed from 'patched' to 'resolved'

Thu Jul 18 10:41:53 2013 TIMB [...] cpan.org - Correspondence added

I'm puzzled by the "no guarantees that it will necessarily work with Moose/Mouse/Moo" in the updated docs. If undef is the same as $constraint->parent->inline_check($varname) shown in the example above, then why not flag that with the same qualifier in the docs? Or perhaps I'm misunderstanding what the qualifier relates to. Also no need for the outer parens in "($_ % 2 == 0)" as they're added by the code. Then the 10-line EvenInt example can be echoed by this equivalent 4-line version where the inline_as clause closely mirrors the where clause: declare EvenInt, as Int, where { $_ % 2 == 0 }, inline_as { (undef, "$_ % 2 == 0") }; What do you need to remove the "experimental" flag from this feature?

Thu Jul 18 12:09:39 2013 perl [...] toby.ink - Correspondence added

On 2013-07-18T15:41:53+01:00, TIMB wrote: Show quoted text

> I'm puzzled by the "no guarantees that it will necessarily work with > Moose/Mouse/Moo" in the updated docs.

If Moose/Mouse/Moo call $type->get_message($value), then it will work; no problem. I'm just worried that there might be some place in the Moose internals where it does something like: my $coderef = $type->inlined; my $message = $coderef->($value); And thus call the coderef directly, in scalar context, bypassing Type::Tiny::get_message, which is responsible for rewriting undef and joining the list with "&&". Grepping through the Moose source suggests not, but it's possible I've missed something; also there are myriad MooseX extensions that might. Show quoted text

> What do you need to remove the "experimental" flag from this feature?

Test cases. I'm unlikely to completely get rid of the feature, but if it turns out to be subtly broken because of above, it may need some implementation changes.

Thu Jul 18 12:19:42 2013 TIMB [...] cpan.org - Correspondence added

Show quoted text

> I'm just worried that there might be some place in the Moose internals > where it does something like: > > my $coderef = $type->inlined; > my $message = $coderef->($value); > > And thus call the coderef directly, in scalar context, bypassing > Type::Tiny::get_message, which is responsible for rewriting undef and > joining the list with "&&".

Could the rewriting undef and joining the list with "&&" be done earlier? Perhaps in the inline_as handler? Show quoted text

> > What do you need to remove the "experimental" flag from this feature?

> > Test cases.

Understood.

Fri Jul 19 07:39:03 2013 perl [...] toby.ink - Correspondence added

Show quoted text

> Could the rewriting undef and joining the list with "&&" be done > earlier? Perhaps in the inline_as handler?

There are basically three techniques: 1. Have get_message do the rewriting. This is what I've done for now as it seemed the least invasive technique. 2. Right now, inline_as basically just returns the exact coderef it's given as an argument. Instead it could return a wrapper around that coderef that does the rewriting. 3. Could rename the current $type->inlined method to $type->_real_inlined everywhere, and then create a new $type->inlined that returns a wrapper around the coderef returned by $type->_real_inlined. #2 means you only get the "return (undef, ...)" sugar if you use Type::Utils to declare your type constraints; you don't get it if you call Type::Tiny->new(...) directly. Also, #2 adds an extra sub call to each ->get_message, slowing it down a teeny bit. (Though ->get_message will not usually be in a "hot" code path in usual use cases.) #3 adds the extra sub call, but only for people who are calling ->inlined; people using ->get_message. So #3 is probably what I'll got for if #1 proves problematic. In fact, maybe even if it does not.

Bug #86891 for Type-Tiny: Reduce boilerplate for inline_as