Add a Podfile option to reuse an external source specification

[igor-makarov]

If a dependency was already defined using an external source, reuse that definition:

    # define an external source
    target 'TargetA' do
      pod 'JSONKit', :podspec => 'https://example.com/JSONKit.podspec'
    end
    # reuse it
    target 'TargetB' do
      pod 'JSONKit', :use_previous_external_source => true
    end

[segiddins]

Why does this need to be an option? It seems like we can default to this . behavior?

[dnkoutso]

Agreed. further more this can also be a method inside the Podfile to simplify this?

[igor-makarov]

@segiddins - Are we certain that this won't break any existing flow?
@dnkoutso - Can you specify what kind of interface you're imagining?

[dnkoutso]

def jsonkit_pod
  pod 'JSONKit', :podspec => 'https://example.com/JSONKit.podspec'
end

And then use that method across targets?

[igor-makarov]

That's what I'm using right now, but I thought a first-party way would be nicer for everyone.

I do think that @segiddins is right - it can be inferred and not need a separate option. What do you think?

[dnkoutso]

Inferring sounds good.

[igor-makarov]

Tried to hack it today, this seems non trivial.

I'll try to find time for it on the weekend.

[igor-makarov]

Done, figured it out!

Some of the tests were failing because iterating over dependencies mutated the target definitions.
I've changed some of the getters on TargetDefinition to not add a nil value to the hash in the getter. This didn't break anything AFAIK but please review it thoroughly.

[segiddins]

how will this linear scan perform on very large podfiled?

[igor-makarov]

There's a bit of a Shlemiel effect here, but ultimately it's all done in-memory and my profiling for 70 total pods shows an increase from 3% samples to 7%.

What do you think? Should I optimize now or can this be done later? I mean, I've made several optimizations recently that outweigh this by a margin 😂

[segiddins]

Probably fine for now, but we might want to eventually put it in a hash for constant-time lookup

[segiddins]

why these changes?

[igor-makarov]

I've changed several getters to avoid mutating the hash because iterating over deps mutated them.

The code assumed, previously, (and implicitly) that deps only get evaluated after the Podfile is completely scanned.

However, going over previously added deps before the entire Podfile is scanned mutated things and now it doesn't.

[segiddins]

👍. Is there a test that will fail if we accidentally go back to the old behavior later?

[dnkoutso]

want to continue this @igor-makarov ?

[igor-makarov]

Hadn't had the time to think about this yet.. Sorry 😭