Lazy load of locale data

[sergle]

Hi
In my code I have found that after removing unused locales it frees ~10Mb of RSS memory of process.
Here I'm trying to load locale data on demand - similar to DateTime::TimeZone
Commits includes full list of generated pm-per-locale files.

[autarch]

I wonder if there's a way to do save memory without having to move everything out into separate files. For example, rather than putting the locale data in __DATA__ it could go in the file some other way and we could manually read the data using seek as needed.

[autarch]

Hmm, I wrote the previous comment and then I remembered e1a4063

[sergle]

Maybe just store data into separate file, not in DATA. But the one problem I see - how to find it's location from code. Any suggestions?

[autarch]

I was thinking of that too. It could even be one file per locale. Mostly what I'd like to avoid is shipping hundreds of namespaces, since each namespace you create also uses memory. I think File::ShareDir would be the right tool for this job.

[sergle]

@autarch could you help me with Dist::Zilla and File::ShareDir? How to use local directory during tests? How to install this directory?
I have never used Dist::Zilla before.

[autarch]

I think you just need to add [ShareDir] to the dist.ini - see https://metacpan.org/pod/Dist::Zilla::Plugin::ShareDir

[sergle]

Sorry for the mess with commits. Should I create a new PR from the fresh fork?

[autarch]

You can just rebase the branch.

[autarch]

I ended doing my own version of this in #17. Thanks for suggesting this!